Available at www.sciencedirect.

com

INFORMATION PROCESSING IN AGRICULTURE 7 (2020) 183– 203

journal homepage: www.elsevier.com/locate/inpa

A role of computer vision in fruits and vegetables

among various horticulture products of agriculture
fields: A survey

Mukesh Kumar Tripathi *, Dr. Dhananjay D. Maktedar

Department of CSE, Guru Nanak Dev Engineering College, Bidar, Karnataka, India

A R T I C L E I N F O A B S T R A C T

Article history: Computer vision is a consistent and advanced technique for image processing, with the
Received 2 October 2018 propitious outcome, and enormous potential. A computer vision has been strongly adopted
Received in revised form in the heterogeneous domain including agriculture. During the study of existing research
8 June 2019 on the role of computer vision in fruits and vegetables among various horticulture products
Accepted 24 July 2019 of agriculture fields it is noticed that, the existing survey paper has not focused properly on
Available online 26 July 2019 mathematical framework, feature descriptor, defect detection on multiple datasets of fruits
and vegetables elaborately. This has motivated us to undertake an extensive survey. In this
Keywords: paper, we examine the paper broadly related to fruits and vegetables among various horti-
Computer vision culture products of agriculture fields, specific model, data pre-processing, data analysis
Agriculture method and overall value of performance accuracy by using a particular performance met-
Survey ric. Moreover, we study the different type of disease present in various fruit and vegetable.
Pre-processing We have also focused on the comparison of different machine learning approach with
Descriptor respect to different performance metrics on the same dataset. Thus, we have found that
Data-analysis among all existing machine learning techniques SVM give better classification accuracy.
Defect detection A generalized framework to grade the quality and defect detection of multiple fruits and
Monitoring vegetables is also proposed in this survey. This paper covers the survey of ninety-eight
papers closely related to computer vision in the agricultural field. By the survey, we have
found that computer vision plays an important role and has a large potential to address
the challenges related to the agricultural fields.
Ó 2019 China Agricultural University. Production and hosting by Elsevier B.V. on behalf of
KeAi. This is an open access article under the CC BY-NC-ND license (http://creativecommons.
org/licenses/by-nc-nd/4.0/).

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
2. Methodology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
3. Computer vision application in agriculture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188

* Corresponding author.
E-mail addresses: [email protected] (M.K. Tripathi), [email protected] (D.D. Maktedar).
Peer review under responsibility of China Agricultural University.
https://doi.org/10.1016/j.inpa.2019.07.003
2214-3173 Ó 2019 China Agricultural University. Production and hosting by Elsevier B.V. on behalf of KeAi.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
184 Information Processing in Agriculture 7 (2 0 2 0) 1 8 3–20 3

3.1. Recognition and classification systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188

3.2. Defect detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
3.3. Data preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
4. Feature extraction based on the various available descriptor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
4.1. Where and how to compute the descriptor? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
4.2. How to compare the descriptor? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5.1. Performance metric symbols used for classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5.2. Overall performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5.3. Performance comparison with another approach. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
5.4. Advantage and disadvantage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
6. Computer vision system to evaluate the quality level of fruits and vegetable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7. Future scope. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
8. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
Declaration of Competing Interest. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
Appendix A. List of abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
Appendix B. Publicly-available data-sets related to agriculture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200

1. Introduction The farmer in a rural area is illiterate, resulting unawareness

with the availability of advanced technology in the agricul-
India is a developing country. It is the sixth largest economy tural field, this lead one of the major challenge in adopting
in the world [1]. US holds the first rank in a world with the advanced technology.
19.39 trillion of GDP followed by China with 12.24 trillion of The development of agriculture is essential and should be
GDP, followed by Germany, Japan, UK and with 2.6 trillion of propositional to the population to fulfil the demand. Also,
GDP, India stood the sixth rank across the world. Agriculture India is one of a major country that exports many agriculture
is the backbone of Indian economy. In India, more than 70% products so it is important that the quality of agricultural
of the human resource directly or indirectly employed in agri- commodities must be sustained until it reaches to the end
culture [2]. In 1991, 64% of employment has been managed by user. The government of India has launched many fruitful
agriculture. After 1991, the Indian market had open the door and beneficiary schemes to enhance the economic condition
for industrialization, privatization etc. And adopt many mod- of farmers, but due to unawareness, only a few are able to
els of development, such as PPP, even still agriculture produc- take advantage of such scheme and able to employ this
ing 43% employment of the total population of the county [3] scheme for smart farming.
and also agriculture has a remarkable impact with an average The farmer has to face some other major challenges due to
18% towards Indian GDP [4]. the poor economic condition. One challenge is that most of
Due to the versatility of environment, India has the ability the farmer need to take help from the middleman to sell
to cultivate a large variety of horticulture product of fruits the product. Another problem is that major of warehouses
and vegetables. In 2017–18, the production reaches up to in India is situated at the district level. To reach to warehouse
305.4 million tones which makes 33% of the agriculture pro- the farmer needs to hire some vehicle by giving some money
duction [5]. Fruits and vegetables account to 90% share of the that is the extra financial burden on the farmer. Even though
total horticulture product. The other categories include flower, after reaching to the center, they are not getting some trained
aromatic, plantation crop and spices etc. In the last decade, the resource person to help this farmer. One other major chal-
production of fruits and vegetables increased by 35% [4]. In this, lenge is that the farmer is not able to adopt him in perceptive
State Uttar Pradesh ranks 1st in the production of vegetables of advanced technology. In order to address these challenges,
with 26.4 million tones, followed by state West Bengal with we require advanced technology in the agriculture field. The
25.5 million tones, which is 30% of the production of vegetables techniques based on computer vision to monitor, measure,
compared to other states of India. In the fruit production, state analysis continuously various aspects and phenomena in
Andhra Pradesh produces 120.98 lakh tones followed by state agriculture product [8].
Maharashtra 103.78 lakh tones which are together 24% of fruit In the view of the above challenges, many researchers
production compared to rest the state of India. have done the work. In that few of them are closely related
The population of India is 1,355,388,325 which is equiva- to computer vision domain are described in this survey paper.
lent to 17.744% of the total population of the world. Sub- The author [9] has done the detail review of quality grading of
sequentially, it is increasing day by day. China holds 1st rank fruits and vegetables applying computer vision techniques.
with 18.54%, followed by India. According to the report, only The principal focal point of a research paper is on feature
32.2% of the population of India are from an urban area, rest extraction based on Color, Texture, shape, and furthermore
remaining 67.8% belongs to the rural area [6]. The total liter- extracted feature is used by the classifier. The one major lim-
acy rate in India is 72% and the rate of literacy rate of male itation of this survey is, they have only considered the single
and female is 80.6% and 62.8% respectively [7]. It shows that image of fruits and vegetables for grading, sorting and recog-
the literacy rate of female is very less compared to the male. nition and classification of disease.
 Information Processing in Agriculture 7 ( 2 0 2 0 ) 1 8 3 –2 0 3 185

The author [10] presents a paper based on computer vision of classifying fruits and vegetables by color, texture, shape,
to grade the delicious apple. In this, an apple data set is divided weight, but other parameter such as good appearance and
into two categories such as health and defect. To evaluate the free from defect disease. The identification and classification
quality of apple different classifiers such as SVM, KKN. 92.5% of various available disease in citrus are still challenging task
and 82.2% accuracy rate has been achieved with SVM classifier during post harvesting.
for healthy and defect categories. The drawback of the present In the post-harvesting there are many factors that affect the
paper is that they have considered only one direction of apple quality and safety of the fruits and vegetables. The author [22]
images. A review for fruit grading system based on color, tex- discuss the processing step for the production of fruits in post-
ture, and size is also presented by the author [11]. harvest. They have additionally given the spotlight on a differ-
Quality evaluation of tomato based on a computer is also ent attribute that affect the quality of fruits and vegetables,
presented by the author [12]. They have described the color how to maintain quality of fruits, Chemical-Based Microbial
statistical feature, color texture feature. In the experiments, Control for Vegetables, role of physical treatment and impor-
they have gotten 100% and 96.47% accuracy rate for tance of computer vision for value-added monitoring product
defective/non-defective and ripe/unripe tomato image. A quality and microbial contamination during the handling
framework of orange sorting using pattern recognition is also along with meeting the traceability requirements. One another
presented by the author [13]. The one disadvantage present in author [23] present the detail review of postharvest quality
a paper, it could not able to identify the presence of disease in evaluation system based on computer vision and composition
orange. Another survey has been presented by the author [14]. of organically and conventionally produced fruits. The key
They all have highlighted the challenges and advantage of Postharvest quality parameters including physicochemical
collaboration between human and robots in agriculture fields. properties, postharvest storage performance, microbiological,
One other survey based on machine vision technology to sensory and nutritional quality. They have also discussed sev-
grade peels pistachios presented by the author [15]. In the eral important problems in organic fruits to maintain the qual-
experiments, they have utilized SVM as a classifier with an ity. One of the disadvantages of the review that it neglects to
accuracy rate of 94.33%. address significant of application of computer vision for stor-
Serval type of available disease of citrus is described by the age quality of fruits and vegetables. One another author [24]
author [16]. They have done a comparison of all steps discuss about the factor that impact the quality of fruits and
involved in a paper such as pre-processing, segmentation, vegetables. Appearance, Textural, Flavor, Nutritional, safety
feature extraction, classification is done. With the survey, is some factor that affects the quality during post-harvesting.
they have concluded that K-means is a more suitable method Recently, the author [25] also present the assessment of
for defect segmentation and texture feature is also a most mango quality, different technology during post-harvesting,
useful feature that is utilized by Support Vector Machine non-destructive assessment criteria during mango supply
and Neural –Network to detect the available disease in citrus chain. Heat treatment, cold storage management, controlled
leaves. A role and importance of statistical machine learning atmosphere storage, 1-methylyclopropen (1-MCP), ethylene is
method are discussed by the author [17]. They have suggested some assessment technique to quality grading during post-
many effective machine learning algorithms that can be uti- harvesting. With this we found that there are need to adopt
lized in agriculture fields. One limitation of the paper is that the application of computer vision to implement low cost
it covers only Machine Learning Techniques. It could not able structures for storage of fruits and vegetables. The author [26]
to describe supervised and unsupervised techniques such as has design and analysis the apple sorting machine the compo-
ANN and DL. A review of grading and sorting of fruits and veg- nent of the machine such as a slider, brush, roller, sliders. The
etables is presented by the author [18]. They have also high- available feature is classified by C4.5 classifiers. The author [27]
lighted the various steps such as pre-processing, analysis the big data tool and method in agriculture fields. For
segmentation, feature extraction, and classification involved the survey, we have chosen only fruits and vegetables from a
in grading and sorting. A grading system for strawberry is also variety of horticulture product. Comparisons of this survey
discussed by the author [19]. They have utilized the K-means and existing survey are presented in Table 1.
method for classification of a strawberry image based on We have analyzed the differing aspect and the role of com-
based on color, texture, shape feature. puter vision for fruits and vegetables. From the existing
Grading, sorting and disease recognition is a key factor to method, the most common one is satellite-based, Thermal
evaluate the quality of fruits and vegetables during posthar- and near-infrared cameras and mobile and sensor device
vest. Many research’s has used application of the computer etc. [28]. Some other techniques to extract the feature of an
vision to quality level evaluation of fruits and vegetable dur- image based on color, texture, and shape are the sum and dif-
ing postharvest. The author [20] present the role of Post- ferent histogram, GCH, CCH, CCV, CLBP, ZM, structure ele-
Harvest Management of Fruits and Vegetables Storage They ment [29,30]. At the last, the most popular techniques used
have additionally talked about significant of the key parame- to analyze image include Machine Learning which consists
ter for grading such as Packaging, Storage process and storage of various methods such as DL, K-Means, SVM, MSVM, ANN,
system. The author [21] present a review on the role and PNN, KNN, LP, WBF and RA [31,32]. Besides the all available
application of computer vision for quality grading of citrus techniques, Deep Learning techniques have recently gotten
fruits during post harvesting. They clarify the various avail- more attention [33].
able techniques to acquire the images and their use for the The above challenges present in agriculture motivated us
non-destructive inspection of internal and external features to do the survey on in fruits and vegetables among various
of these fruits. The role of machine vision is more effective horticulture products of agricultural fields. Based on the
186 Information Processing in Agriculture 7 (2 0 2 0) 1 8 3–20 3

Ref.

[17]

[16]
[14]
[10]
[15]
[12]
[13]
[26]
[18]
[19]
[11]
[9]
Start
Performance
Evaluation
Metric
U
Collect data

U
✕

✕
✕
✕
✕
✕

✕
✕
✕
✕
✕
Similarity
Measure

Choose Feature Descriptor

U
✕

✕
✕
✕
✕
✕
✕
✕
✕
✕
✕
✕
Choose Model
Classification

Training and Classification

U
U

U
U
U
U
U
U
U

U
U
U
✕
detection

Training by machine
Defect

learning
U
U

U
U

U
✕

✕
Size

Fig. 1 – Flowchart of the proposed methodology.

U

U
✕

✕
✕
✕
✕
✕

✕
Shape

discussion above, this review paper will try to give the com-
U

U
U

U
U
U
✕

✕
✕
✕
✕
✕

parison of the present survey with existing survey. This paper

Feature Extraction

covers the numerous application of computer vision in agri-

Texture

culture fields. This survey has conjointly potential to compare

the various method of different phases of computer vision
U

U
U

U
U
U

U
U
U
U
✕

such as Pre-processing, Segmentation, Feature extraction,

Color

Defect detection, Classification, Similarity measure, Perfor-

U

U
U

U
U
U

U
U
U
U
✕

mance evaluation metric. One limitation of the present sur-

vey that it could not able to cover the require time to
Segmentation

recognition and classification of fruits and vegetables. It

would be a great challenge and scope for other research’s to
do the deep literature survey and implementation for less
recognition time.
U
U

U
U
U
U
U
U
U
U
U
U
U

The all procedure is represented in the form design cycle

Processing

refer Fig. 1. The rest paper is organized as follows: Section 2

describes the methodology of the survey followed by Sec-
Table 1 – Comparisons of this survey and existing survey.

Pre-

tion 3, which give broadly cover all aspects of the application

U
U

U
U
U
U
U
U
U
U
U
U
U

of computer vision in fruits and vegetables of agricultural

Application of statistical machine learning in

Detection and Classification of citrus disease

fields. This session covers the various subtopic from 3.1 to

Sorting and Grading of Fruits and Vegetables

3.3. In Section 3.1 Recognition systems are described, fol-

Human-robot interaction in Agriculture

lowed by Defect detection, Data preprocessing in 3.2 and 3.3

Orange Sorting by pattern recognition

respectively. Section 4 explains distinguish available feature

Strawberry quality grading system
Grading of golden delicious Apple

extraction based on the various available descriptors followed

Grading of Fruits and Vegetables

Evaluation of peeled pistachios

by Performance Metric symbols used for classification, Over-

Quality evolution of Tomato

all Performance, Performance comparison with another

approach, advantage, and disadvantage in respective Sections
Fruits Grading System

5.1–5.4 presented. Section 6 present the deep review of assess-

Apple sorting system

ment quality of fruits and vegetables using computer vision.

In Section 7, describe the future scope in details and the last
conclusion is presented in Section 8. Also, a list of abbrevia-
This Survey
Application

agriculture

tions is represented by Appendix A.

2. Methodology
2019
2019

2018
2018
2018
2017
2017
2016
2016
2016
2012
2010
2011
Year

All the reference paper is collected and deeply analysis in

two-step:
Table 2 – Containing the answers to all the questions listed in the methodology section.
Phase Techniques This paper Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref.
[50] [66] [30] [56] [57] [60] [58] [64] [69] [40] [38] [90] [37] [45] [36]

Year 2019 2018 2018 2017 2017 2017 2016 2015 2015 2015 2014 2013 2012 2010 2010 1992
Segmentation Thresholding U U U
K-Means U U U U U U U U U
Spatial Weighted Method U U
Median Filter U U
Watershed Method U U
Grab Cut U U
Otsu’s Thresholding U U

Information Processing in Agriculture

Feature GCH U U U U U U
extraction CCH U
CCV U U U U U U U
CDH U U
LBP U U U
CLBP U U U
SSLBP U U
LTP U
SEH U U
HIST U U
UNSER U U U
ISHAD U U
GLCM U U U U
ZM U U
BIC U U U U U U
WDH U U
SIFT U U U
Statistical Feature U U U U

7 ( 2 0 2 0 ) 1 8 3 –2 0 3
Classification KNN U U U U
SVM U U U U U U U U
Non-Linear SVM U U
MSVM U U U U
ANN U U U
PNN U U
GA U
Similarity Euclidean distance U U
measure L1 Distance U U
Cosine distance U U
Canberra distance U U
The Di distance U U
Chi-square Distance U U
Performance Precision U U
evaluation Recall U U
metric Mean U U
Variance U U
Equal Error rate U U
Classification Accuracy U U U U U U U U U U U U U U U

187
188 Information Processing in Agriculture 7 (2 0 2 0) 1 8 3–20 3

1: Collection of all paper by using keyword based on a distance, Di distance, Chi-square Distance is some few meth-
search of journals, conference, a new paper by a different ods to measure the similarity. Finally, Precision, Recall, Mean,
site such as Google Scholar, ACM, Science direct, Research Variance, Equal Error rate, Classification Accuracy Metric are
gate, IEEE explore, Google etc. used to evaluate the performance of the system. As we know
that computer vision covers numerous method in all phase.
We have used the following keyword to search the paper is But in this survey, author could not able to consider other
represented by syntax such as: techniques due to limitation of the application. So, this would
be great opportunity for research’s to explore other tech-
[‘‘Recognition”] AND [‘‘Classification”] AND [Fruits or niques in agriculture fields.
Vegetables or Agriculture] = Results.
3. Computer vision application in agriculture
With this syntax, we have gathered numerous papers
through given link. We have downloaded three hundred fifty In this part, we have described a computer vision related
papers by using all keyword. Out of all research papers, we research in fruits and vegetables, indicating various findings
have access to two hundred thirty papers. In that one hun- for recognition and classification system, defect detection,
dred thirty-three paper has been discarded and 97 papers data preprocessing.
are cited which are relevant to our domain and finally, all
selected papers have been deeply analysis as per requirement 3.1. Recognition and classification systems
and suitability of the domain.
In the section, we have focused on a brief review of the work
2: In this step, all the selected paper has been analyzed, done by many researchers in the domain of the recognition
based on all research questions. Some important question and classification of fruits and vegetables. Many researchers
is listed below: are working on recognition of image, but initially, the author
[34,35] proposed the effective image recognition approach

What is the major problem for recognition of the image of based on color, texture, classification. The author [34] com-
agriculture fields? pute the to process the CCV, the technique finds the associ-

Which techniques have been used to detect the fruits and ated parts of the image and characterize the pixels inside a
vegetable disease? given shading container either coherent or incoherent. Subse-

What is the suitable approach for classification of disease quent to grouping the CCV pixels, CCV obtains two color his-
of fruits and vegetables? tograms: one for coherent pixels and another for incoherent

Why the preprocessing method is required? pixels. The two histograms are put away as a solitary his-

How to extract the feature of images using color, shape,
togram. Image pixel is classified by border/interior classifica-
texture?
tion [35], considering two color histogram, one as border pixel

What are the available datasets in agriculture field?
and second one as only interior pixels.

Which type of the metric has been adapted to measure the
Veggie Vision is the first automatic produce ID system
performance of the system?

Is an analysis of their performance on the different introduce by [36]. They have used a Histogram, Color and tex-
datasets? ture method to evaluate the accuracy of the recognition sys-

Is any author comparing their own method accuracy with tem. For the experimental purpose, they have considered
another? apple and orange as a case study and results are evaluated

Is the author applying similarity measure method, evalua- by the histogram method in the form of Hue, Saturation,
tion criteria to match the performance criteria? and Intensity. Color features having accuracy 72% and 90%

What is the major disease in a different category of fruits correct choice in top four selections. They also used the two
and vegetables? texture method named as Texture Measure A and Texture
Measure B to measure the convolves and center surrounding
All the answer to the question asked in the methodology respectively. The combine Texture produces accuracy with
section is presented in Table 2. In that, we have shown the dif- 33% and 63% correct choice is in the top four selections and
ferent method used by researchers in all phases. All the a combination of color, texture feature having 84% and 96%
answer to the question asked in the methodology section is top four choices.
presented in Table 2. In that, we have shown the different The author [37] uses many descriptors for feature extrac-
method used by researchers in all phases. In the segmenta- tion. They have also fused some descriptor and implemented
tion section, various method has been analysis for prepos- on supermarket dataset which consists of 15 types of fruits
sessing. This method is Thresholding based, K-Means, and vegetables with a different pose, illumination, and vari-
Spatial Weighted Method, Median Filter, Watershed Method, ability. In the experiments, BIC, CCV and UNSER feature are
Otsu’s Thresholding. GCH, CCH, CCV, CDH, LBP, CLBP, SSLBP, fused, for a SVM classifier, more than 95% accuracy rate has
LTP, SEH, HIST, UNSER, ISHAD, GLCM, ZM, BIC, WDH, SIFT, been achieved in the proposed system. The author [38] pro-
Statistical Feature method is used to analysis in feature posed a framework for recognition and classification of fruits
extraction phase. For the classification KNN, SVM, Non- and vegetables. The framework consists of three steps such as
Linear SVM, MSVM, ANN, PNN, GA classifier has been utilized. Image Segmentation, Feature Extraction, and final classifica-
Euclidean distance, L1 Distance, Cosine distance, Canberra tion of fruits and vegetables using MSVM techniques. The dis-
 Information Processing in Agriculture 7 ( 2 0 2 0 ) 1 8 3 –2 0 3 189

tinctive feature extraction method such as GCV, CCV, BIC, the feature, which represent an image as an order less distri-
UNSER, ISADH is implemented and evaluated in HSV color bution of features. Two different descriptors Scale Invariant
space. 93.62%, 95.27%, 95.82%, 96.96%, 98.90% accuracy rate Feature Transform and SPIN is employed and evaluated in dif-
is achieved for respective descriptor. The learned classifier ferent categories of the dataset. Non-linear SVM is utilized for
is applied to detect the image used to compare the accuracy classification. The highest accuracy is achieved by (HS + LS) -
rate of images [39]. To assess the outcomes different execu- SIFT descriptor with a value of 96%. The SPIN technique is
tion metric, for example, Recall, Precision, F-measure, False- additionally evaluated by an author, however excluded in a
Positive has been used for the experiment. They utilize two paper since it fails to produce a quality outcome. The parallel
different set of image, one is for single scale case with con- and distributed techniques also play important role in agri-
taining 170 images of car, second dataset case with multi- culture field by [46]. In their work, they have done a compar-
scale case with 108 images of car with different size and rota- ison between parallel and serial execution. The results show
tion. With the test, they express that recall- precision curves that parallel execution is quite useful for image recognition
are more appropriate than ROC curves for measuring the per- compares to serial execution and also They have also pro-
formance of object detection approaches. posed image processing on a massively parallel computer
The author [40] has used 15 different categories of fruits using single-instruction, multiple-data (SIMD) computer and
and vegetables for a case study. This dataset has also different multiple-instruction multiple-data (MIMD) parallel processing
pose, variability, cropping, and partial occlusion effect. Differ- computers. The result indicates MIMD implementation is at
ent descriptor based on color, texture, shave has been used to least four times faster than the SIMD implementation.
extract the feature of the image. This extracted feature is The author [47] present progressive randomization
used for classification. The MSVM is used for training and descriptor for image categorization. One most important
classification. Finally, they have shown 93.84 accuracy rates. characteristic, its low dimensionality and its unified approach
The author [41] also used various feature extraction tech- for different applications the advantage of progressive ran-
niques such as ACC, BIC, CCV, GCV, LAS, QCCH, EOAC based domization descriptor is its low dimensionality and its unified
on color, texture, and shape on the 15 different categories of approach for different application. One major drawback is
fruits and vegetable dataset. They have combined two low- that it uses LSB Value of the image which does not contain
cot classifier such as support vector machine (SVM) and all information under MSB. The learning algorithms for the
learned classifier such as Naive Bayes, Decision Tree, Simple constellation model to recognize the image is proposed by
Logistic, Naive Bayes Tree, k-Nearest Neighbors (KNN). The the author [48]. The accuracy rate is 87%, 90% in the case of
proposed framework to combine two classifiers is totally face and car image separately. They have mentioned key fac-
depending on term diversity measures to find detect which tor is how to derive optimal detector, this detector important
base learning classifiers are suitable to be consolidated. because of two reasons, first, it requires to inspect a possible
A mathematical framework has been introduced for the joint extensive number of conceivable joint part positions,
detection of the image based on the probability method by second it is not invariant as for interpretation, pivot, or scale.
the author. The probability method is useful for decision mak- The main advantage of the proposed method is that it can
ing. They have also used other parameter such as shape and detect the image with illumination, occultation and also
appearance to detect the images. For the experimental pur- invariance effect. The cost of the experiment is very high, is
pose 101 different categories of object and each category con- one of the disadvantages of the proposed system.
tains between 45 and 400 images. The proposed system is able TFRS fruits recognition system as proposed by the author
to produce a performance of around 70% to 95% for different [49]. They have tested 123 images of thirty-nine species of
category of a dataset. Another author [43] describes unsuper- Thai fruit. For feature extraction they utilize three various
vised learning techniques for image categories. In this features based on color, size, edge shape. RGB histogram
approach, image categorization is based on the clustering and Sobel edge detection based on color and edge shape fea-
method. Based on this approach unrelated or inhomogeneous ture is used to extract the feature of images. With the exper-
image are categorized. The one limitation of this proposed iment they are able to achieve 93.5%, 86.67% accuracy rate for
system is that it totally depends on the shape and color of trained and untrained dataset respectively. Also, they calcu-
the local feature and also how to weight of different type of late the average access time with a value is 54.68 s per image.
feature in the case of local or global. Also, the author [44] used The author [50] presents a different form of KNN classifier
effectively code-blocks for visual recognition and also used such as Fine-KNN, Medium-KNN, Coarse -KNN, Cosine- KNN,
SVM and Naive Bayes two linear classifier for categorization Weighted-KNN. In the experiment, the feature of fruits is
of the image. In this, they have used three different classes extracted by first and second order statistical method. The
of the dataset. First contain side views of cars. Precision and accuracy rate of the proposed classifier is 96.3%, 93.8%, 25%,
recall metric is used to evaluate the performance on the same 83.8%, 90% and 95% respectively. The results clearly show that
dataset. The second data set is in seven different categories Fine-KNN has quality output, whereas coarse- KNN has the
by name Xerox 7 contain 1776 images. The third dataset lowest accuracy rate. The footboard machine is proposed by
ETH80 is in four categories such as cars, horses, dogs and the author [51]. They have used SVM for classification. To
cows. The best performance has been achieved on the Xerox evaluate the performance of a proposed machine, Precision
7 dataset by the linear SVM classifier. With 4.8% error rate and and recall metric is used. The performance of the machine
a standard deviation is 1.2%. is 82.76% and 78.77% respectively.
The author [45] also presents a method for an extension of Deep Learning base architecture is also an effective
image categorization and also classification based on a bag of method to recognize the fruits and vegetables [52,53]. Neural
190 Information Processing in Agriculture 7 (2 0 2 0) 1 8 3–20 3

Network based fruits recognition system is also presented by fruits and vegetables and other limitation is utilization of
the author [54]. They have utilized Deep Learning architecture new trend classifier such as ML, ANN, Fuzzy logic, Artificial
and achieved 85.11% accuracy rate. Also a novel approach to intelligence etc. In future, all researchers should utilize vari-
recognition of images based on DL techniques has been also ous and current feature extraction and classification method
proposed by the author. The author also [56] describe the for recognition and classification.
color and texture descriptor for recognition of fruits. Color
Features (Mean, Standard Deviation, Skewness, and Kurtosis) 3.2. Defect detection
and GLCM feature extraction method with SVM classifier is
evaluated to test the accuracy of the system. The author In agriculture, the disease is one major factor for loss of qual-
has achieved quality, accuracy rate with a value of 91.67% ity of fruits and vegetables. The economic condition of the
accuracy for apple dataset, with 75% strawberry dataset able farmer is directly proportional to the quality of fruits and veg-
to produce a lower accuracy rate. etables. For example, soybean rust is a fungal disease in soy-
The author [57] has also used texture and shape feature for bean has significant economic losses to the farmer. If we are
green apple recognition. Some parameter such as contrast, able to identify the disease, the report says that the farmer
entropy, means used to describe the characteristics of the tex- may get benefit approx. 11 million-dollar profit [63].
ture of image 90% accuracy rate has been achieved by using a In a recent year, a lot of work has been a car out to detect
combination of color and texture descriptor for recognition of and classify the disease. The disease can appear in both pos-
green apple. A framework for plant species detection pro- sibilities such as pre-harvesting and post-harvesting. With
posed by author [58]. They have utilized the color, texture, the help of pesticide and other chemical product, we can con-
shape-based descriptor to extraction the feature of images. trol the disease. In the aspect of economic various factors is
Five different categories of a dataset are evaluated by SVM responsible for the loss, such as storing and sorting manage-
classifier. With fruits dataset, the system shows poor accu- ment system. As early as possible if we could detect and clas-
racy rate with an estimation of 67.3% and highest exactness sify the disease, we can prevent the loss of fruits and
rate 98% is accomplished in leaf subcategory datasets. Also vegetables. That can lead to a strong economic background
Hybrid RGBD feature is proposed by [59] to detect the image so their life will be with health and prosperity. One major
of the fruit. Additionally, they have presented the comparison problem is that disease may be spread over the part of the
table to recognize the accuracy based on color, fused color image of fruits and vegetables within short spam. Due to this,
with shape descriptor. The accuracy of color and fused color, the application of computer vision to the detection and clas-
shape with a value of 97% and 99% respectively. sification of the different type of disease is crucial in agricul-
The author [60] present the modified fruits–fly optimiza- ture fields. Many researchers are trying to design and develop
tion techniques to recognize the fruits. Classification Accu- effective and computer vision model to detect the disease and
racy, True Negative, False Positive are utilized as a to overcome those challenges.
performance evaluator metric. The highest accuracy rate with The traditional approach to detect the disease is totally
a value 98.95% is achieved on a WBC dataset with Classifica- based on an expert opinion by using eye, smell, and observa-
tion Accuracy metric by GAFOVFS techniques. tion. One major drawback of this it is time-consuming, the
The Watershed Image method for background subtraction cost will be more and the last most time we need to rely on
is described by [61]. They have done the comparisons between expert opinion. One limitation of this method is that results
neural network, Naive Bayes and decision tree algorithms. cannot be authenticated to overcome this limitation many
The decision tree has the highest accuracy rate with a value researchers have widely used advanced techniques to detect
of 93.13% using CA as the metric. The Naı̈ve Bayes and a neu- the disease in agriculture [30,56,64,65].
ral network have produced 91.94%, 92.84% accuracy rate indi- A survey has done on the application of image processing
vidually to identify orange image condition such as ripe, of fruits disease detection and classification by the author
unripe, and scaled. Precision and Sensitivity are also used [29]. In that, they have proposed a framework to detect the
as a performance metric to evaluate the system for all three disease based on color, texture and shape descriptor. Addi-
classifiers. Decision tree classifier with Precision and Sensitiv- tionally, they have used the Multiple Support Vector Machine
ity metric has the highest accuracy rate 93.45% and 93.24% classifier to achieve an average accuracy more than 99%. Aver-
compared to the neural network, Naive Bayes classifier. age accuracy more than 99%. Author [56] proposed an auto-
The author [62] used a CNN to classify the fruits and veg- matic system to recognize various categories of fruits based
etables. The outcomes demonstrate a 95.6% accuracy rate on color and texture feature. All images are classified by the
has been accomplished by VGG model. The Common Support Vector Machine classifier and produces high accuracy
Publicly-available datasets related to agriculture are shown rate with a value of 83.33%. Also, the author [64] have pro-
in Appendix B. The outline of this section is presented in posed fused descriptor based on color, Texture, shape to clas-
Table 3. This table highlight description of the problem, Pre- sify the disease of apple. The fusion of CCV + CLBP descriptor
processing techniques, feature extraction method, various has accomplished the most noteworthy accuracy with an
classifiers, metric and value of accuracy of the system. In this estimation of 95.93%.
section, various feature extraction method is used to extract Some of the authors [30,65] has described the basic param-
the feature and also discuss about different classifier to clas- eter of disease detection such as Ratio of Infected Area, Lesion
sify the image. In this, author has also described the existing Color Index, Damage Severity Index. In the experiments, they
problem in details. One major disadvantage of recognition have calculated the total infected region and the total area of
and classification system is the unavailability of a dataset of images. They have also given the details of soybean disease
Table 3 – Application of computer vision in fruits and vegetable and used techniques.
Year Application Problem description Data pre-processing dataset used Descriptor details Data analysis Performance Value of metric Ref.
method metric used used in %.

2018 Fruits and vegetable Classification based on CNN Learning 26 different set of CNN CNN CA 95.60 [61]
classification model. Techniques fruits and Veggies.

Information Processing in Agriculture

2017 Automatic fruit Recognition of fruits based Grab cut 8 different set of GLCM SVM CA 91.67 [56]
recognition system. on color and texture. segmentation fruits.
2017 Recognition of apple Recognition of apple based Thresholding Apple Data-set. Contrast, Entropy, SVM CA 88.25 [57]
on fused texture and shape. Mean
2017 Fruit recognition Fruits recognition Using Discrete Cosine 32 different set of Hybrid RGBD Hierarchical multi CA 97 [59]
Multi-class classifier. Transform fruits data-set. feature classification
2016 Image Recognition Image recognition and Learning Handwritten data- Bilinear-Deep model SVM Mean 87.25 [55]
classification. Techniques set.
2016 Recognition of Ripe, Recognition of Ripe, Unripe Watershed method Orange data-set. Border Interior Naı̈ve Bayes, ANN, CA 91.94,92.84,93.13 [60]
Unripe and Scaled and Scaled Condition based Decision Tree
Condition on ML techniques.
2015 Plant species recognition plant species recognition K-means Fruits Data-set. SIFT SVM CA 67.33 [58]
based on Texture feature
discriptor.
2013 Detection of fruits and Detection of fruits and K-means Supermarket GCH,CCV, ISHAD, SVM CA 96 [38]
vegetables vegetables using ISADH Produce data-set. BIC
feature algorithm.
2010 Classification of fruits Fruits and vegetables K-means Supermarket GCH,UNSER, CCV, MSVM CA 95 [37]

7 ( 2 0 2 0 ) 1 8 3 –2 0 3
and veggies classification based on Produce data-set. BIC
multi-class classification.
2010 Recognition of image Recognition of image using Spatial weighting Sets of car, bicycles, SIFT Nonlinear SVM Equal error rate 96 [45]
Spatial Weighting for Bag-of- and motorbikes.
Features.
2007 Recognition of fruits and Recognition of fruits and LBB Based method Arts Scenario. PR NBDC_BLDA CA 88.30 [47]
image image using PR Statistical
descriptor
2000 Identification of fruits Identification and Median Filter Fruits Data-set. Statistical Feature Fine-KNN, Coarse- Precision & 96.3, 25 [48]
classification of fruits KNN. KNN Recall
1992 Recognition system Recognition of image based Thresholding Fruits image data- RGB Color KNN CA 84 [36]
on color and Texture. set

191
 192
Table 4 – Method to detect & classify the disease.
Year Application Problem Description Data Pre-processing Data-set used Descriptor Data analysis Disease Type Performance Value of metric Ref.
Details method metric used used in %.

2018 Disease Classification of K-means Data-sets of papaya GLCM SVM Black Spot, Brown CA 95.2 [66]
Classification. disease in papaya. Spot, Powerdary
mildew
2018 Recognition and Recognition and Weighted Fruits data set GRB,GLCM SVM Anthracnose, Black CA 96.9 [72]

Information Processing in Agriculture

classification classification to of segmentation spot, Canker, Scab,
citrus diseases. Melanose.
2017 Disease detection. Detect the disease Otsu’s Thresholding Soybean leaf BIC,CCV, CDH, SVM Soybean Rust, CA 95 [30]
based on image method diseased image LBP, SSLBP, LAP Bacterial Blight,
processing Sudden Death,
techniques. Downy Mildew,
Brown Spot,
Frog Eye
2016 Disease detection & Classify disease K-means A diseased apple GCH, CCV, RF Classifier Blotch, Rot, Scab CA 76.25 [70]
classification. based on the various LBP.CLBP
descriptors.
2016 Classification of Compute the Clustering-based data set of diseased RGB Feature SVM Black Pod Rot, CA 85 [71]
defect infected level of Segmentation cacao pods Vascular Streak
cocoa farms Dieback, Cacao Pod
Borer,
2015 Classification of Classification of K-means Data-set of diseased GCH, CCV, LBP, MSVM ROT, SCAB, BLOTCH CA 95.94 [64]
disease. disease in apple apple CLBP,
based on color, ZM

7 (2 0 2 0) 1 8 3–20 3
texture
2015 Fruits Classification. Classification of K-means Data-sets of GLCM NEURAL Bacterial light, Rot, CA 90 [69]
pomegranate fruits diseased NETWORK Spot
and leaf. pomegranate
2014 Detection of disease. Detection of disease K-means Apple Data-set. GCH, CCV, SVM ROT, SCAB, BLOTCH CA 92.98 in RGB, [40]
based on color, UNSER, 97 in HSV
texture. BIC
2012 Apple Disease Apple Disease K-means Apple data set GCH, CCV, LBP, MSVM Blotch, Rot, Scab CA 95 [90]
classification classification based CLBP
on color, Texture,
Shape.
2006 Detection of the Classify the healthy NA Arable diseased crop NA NEURAL Yellow rust on NA 99 [68]
defect. and disease with NETWORK wheat
crop.
 Information Processing in Agriculture 7 ( 2 0 2 0 ) 1 8 3 –2 0 3 193

categorize in six major forms such as Soybean Rust, Bacterial 3.3. Data preprocessing
Blight, Sudden Death, Downy Mildew, Brown Spot, Frog Eye. In
the experiments, they classified the soybean disease with dif- Preprocessing is an essential step in computer vision. In our
ferent classifiers such as SVM, KNN, and PNN Classifier. With finding, the large amount of related work 80% involved some
the experiment, they found that CCV and BIC descriptor deli- basic preprocessing step. Further, process image is applied in
ver 90%, 95% performance accuracy. With, KNN classifier and next step such as extraction of the feature based on color, tex-
90%, 92% accuracy rate have been accomplished. ture, shape. The most common preprocessing techniques
Common diseases in papaya fruit are a black spot, pow- were image resizing such as 256  256, 128  128, 96  96
dery mildew, brown spot, phytotron blight, and anthracnose and 60  60 pixel. K-mean is very effective and popular image
are described by the author [66]. A framework based on a segmentation techniques related to agriculture
machine based on an agro medical expert system is pre- [40,44,55,66,70,75,76]. In our survey, we have observed that
sented. Additionally, for the experiments, evaluation, they 70% the author has used K-mean technique. Some author
have implemented the Co-occurrence matrix feature and has also used thresholding techniques [46,77]. Some other
statistic feature. SVM, Decision tree and Naive Bayes classifier techniques are also categories such as region based, edge-
have been utilized for classification of disease. The average based, attribute-based, model-based and graph-based
accuracy of all the classifiers 97.03%, 88.93%, and 80.89%. method [25].
The author [67] has done the detail review the losses in Background subtraction [65,74] and foreground pixel
fruits and vegetables in both pre-harvest and post-harvest extraction [78] is also used to process the image. Watershed
due to fungal infection. They have stated that nears about segmentation [79,80] edge-based segmentation techniques
25% loss of total production. An automatic identification of [81,82] were used to detect the boundary of the image. In
plant stresses and diseases in arable crops using proximal histogram-based segmentation, the histogram is calculated
optical sensing and self-organizing maps is proposed by from the frequency of occurrence of the various color levels
[68]. They have utilized the neural network architecture to in the image [83,84,85]. Other than that, some author has used
train the feature. Additionally, Artificial Neural Networks this process images into feature extraction [65] such as Gray
has utilized by the author [69] to detect and classify the dis- level co-occurrence method, Local linear model, linear dis-
ease in pomegranate fruit. They also describe the type of dis- criminate analysis. Some author has also used the processed
ease present into pomegranate fruit such as bacterial blight, image to extract the feature by combing color, texture, shape
fruits spot, fruit rot, and leaf rot, they have achieved a good [64].
accuracy rate of more than 90%. In this section, various strategy is involved for background
A hybrid approach to detect the disease in apple fruit is subtraction of image. Most of author has used segmented
proposed by [70]. With the experiment, CLBP and LTP with techniques in one color space. To overcome of this limitation,
Gabor classifier give better result compare with another it is advisable to all researcher to perform background sub-
method. Gabor classifier with CLBP descriptor produces the traction of image in number of available color space for same
highest accuracy at 70%, 80%, 80% and 100% for a respective data-set. One major disadvantage of pre-processing that it is
disease present in apple such as apple rot, apple scab, apple time consuming and further phase of recognition is totally
blotch, normal apple. Due to disease 40% loss has been depended on prepossessing. In future, all researcher should
recorded due to the only presence of disease in cacao farms apply effective and efficient solution for background subtrac-
[71]. tion of image.
The author proposed a framework for recognition and
classification of citrus disease based on color, texture, geo- 4. Feature extraction based on the various
metric feature. RGB, GLCM and area, solidity, orientation fea- available descriptor
ture is used to extract the feature. They have also discussed
the disease present in citrus such as Anthracnose, Black spot, In the present survey, we have given focus to extract the
Canker, Scab [72]. Also, some framework to detect the disease images feature are mostly based on color, texture, shape
is introduced by the author [73,74]. The Table 4 provides the descriptor. Color is most basic feature extraction descriptor
different disease present in fruits and vegetables. The differ- contain a different variety of methods such as RGB histogram,
ent classifier method and evaluated metric with accuracy rate Opponent histogram, Hue histogram, RG histogram, Trans-
is also presented in Table 4. form Color Distribution, Color Moment and Color Moment
The present work in this section deal with description of Invariants and others [86].
various fruits and vegetables. Different segmentation Texture descriptor is also having a various category of
method, feature extraction method and data analysis to methods such as mean, variance, energy, correlation, entropy,
obtain the accuracy of the system. Author has also listed var- contrast, cluster shade. Fourier descriptor, Space descriptor,
ious available data-set of fruits and vegetables with various angular transform, image moment etc. are categorized under
disease. One limitation of present survey, it could able to the shape descriptor [87].
detect and classify the defect on only train data-set. It is Many descriptors are available and review based on a large
advisable to all the researchers to consider this limitation number of the application using color, texture, and shape.
and try to come with new experiential solution to recognize The descriptor is very helpful in answering some question
and classify the defect in real-time. listed such as.
194 Information Processing in Agriculture 7 (2 0 2 0) 1 8 3–20 3

How to recognize the different variety of fruits and

Table 5 – Similarity measure.
vegetables?

How to detect the disease available in fruit image? Euclidean distance
P 2 1=2

How will it differentiate the face of human from the Euclident ðDa ; Db Þ ¼ ð dim b a
k¼1 D ðkÞ  D ðkÞ Þ
image?
L1 distance P  

How will it do the pattern recognition?  b 
L1 ðDa ; Db Þ ¼ dim a
k¼1 D ðkÞ  D ðkÞ

Cosine distance Pdim b 1=2

The accuracy of training and classification depends upon a ð D ðkÞDa ðkÞ2 Þ
Cosine ðD ; Db Þ ¼ Pdim k¼1 1=2 Pdim 1=2
the goodness of the descriptor. That’s why we need various ð b 2
D ðkÞ Þ a
ð D ðkÞ2 Þ
k¼1 k¼1
descriptors which can able to extract the feature of images EMD distance P  
 b a 
efficiently. Thus the good quality of descriptor plays a major Emd ðDa ; Db Þ ¼ dim
k¼1 cdf ðD ðkÞÞ  cdf ðD ðkÞÞ

role in the most recognition and classification problem. Canberra distance

P jDb ðkÞDa ðkÞj
These are the following three questions to answer to Canberra ðDa ; Db Þ ¼ dimk¼1 jDb ðkÞþDa ðkÞj
develop an effective and powerful descriptor. The question The Di distance  b
P  D ðkÞDa ðkÞ 
is listed below. di ðDa ; Db Þ ¼ dim
k¼1 1þDb ðkÞþDa ðkÞ

Chi-square distance  

Where to compute the descriptor? P ðDb ðkÞDa ðkÞÞ2 

How to compute the descriptor? chisq ðDa ; Db Þ ¼ 12 dim
k¼1  Db ðkÞþDa ðkÞ 

How to compare the descriptor?

4.1. Where and how to compute the descriptor?

Depends upon the type of the application, there is a various Table 6 – Evaluation criteria.
way to compute the descriptor. In our paper, we have limited
Average retrieval precision (ARP):
our scope based on color, texture, and shape parameter to P
ARP ¼ N1 Ni¼1 PðXi Þ
compute the descriptor. Effective descriptor should have fol-
PðXi Þ ¼ IInt  100
lowing characteristics [88,89].
Average retrieval rate (ARR):
P

Locality ARR ¼ N1 Ni¼1 RðXi Þ

Pose invariance RðXi Þ ¼ IIdbr  100

Distinctiveness

Repeatability Average retrieval accuracy (ARA):
P
ARA ¼ N1 Ni¼1 AðXi Þ,

4.2. How to compare the descriptor? where AðXi ) is the accuracy for query image Xi and define as:

100; where Ir is in majority among It
AðXi Þ ¼
0; else
Similarity measure and evaluation criteria are two major cat-
egories to compare the performance of the descriptor. Simi- mon classifiers used by many researchers [30,37,40]. Another
larity measures are calculated by finding the distance classifier such as a PNN, GA, Decision Tree, DL, ANN, etc. are
measure between the descriptor. In our finding, we have iden- widely used a classifier to measure the accuracy of the pro-
tified some important and different distance [30] are listed in posed system and also we have found that none of the
Table 5. researchers have been given the definition and symbol of per-
We also have searched some method for evaluation crite- formance, due to unavailability we have to describe the per-
ria. This method is helpful to identify the top match. Average formance metric is shown in Table 7.
retrieval precision (ARP), Average retrieval rate (ARR), Average
retrieval accuracy (ARA) method category under evaluation 5.2. Overall performance
criteria is defined following expression [30] such as (see
Table 6). In this section, we will present how the authors show the final
performance value of the evaluated algorithms. The overall
5. Discussion performance that is considered as a keyword of this survey
utilizes distinguish performance metric to measure the accu-
In this part, we will present a computer vision related racy of the proposed system. Selection of performance metric
research in fruits and vegetables, indicating various findings totally depends upon the type of application and dataset used
for Performance Metric symbols used for classification, Over- by the author.
all performance, Performance comparison with another The author [90] used a variance metric (r) to get the recog-
approach, Advantage and disadvantage. nition accuracy of the dataset with consist 15 different sets of
fruits and vegetables. The overall accuracy is more than 90%,
5.1. Performance metric symbols used for classification which indicate good performance. Also, they used the CA
metric to recognize the disease of apple. In this, they have
In our finding, we have to survey many papers to obtain the implemented many feature extraction methods in both RGB
accuracy of the proposed system. SVM and KNN are two com- and HSV color space.
 Information Processing in Agriculture 7 ( 2 0 2 0 ) 1 8 3 –2 0 3 195

Table 7 – Description of various metric symbols used to calculate the accuracy of the system.
Sr. No. Performance metric Symbol used Definitions

1 Precision P It is Called positive value and defines by P = TP/ (TP + FP)

where TP denote correct Prediction and FP denotes False
Positive
2 Recall R It is calculated by R = TP/(TP + FN) Where TP denotes correct
prediction and FN denotes False Negative.
3 Quality Measure QM It is calculated by QM = TP2/(TP + FP) (TP + FN) where FN is
false negatives, FP is False Positive and TP denotes correct
prediction
4 Mean m It is calculated by m = 1/n R(Xi), Where R(Xi) is a total number
of the sum of all other number and n is a number of counts
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
5 Variance r  
It is calculated by r  1=nðX  XÞ,X Where is a mean value, X
is sample value 2
6 Standard Deviation X2 It is calculated by X2 = ðn1ÞS
r2
. where S is sample size, n is a
count of sample and r is variance
7 Mean Relative Error MRE The mean error between estimated and predicted values, in
percentage
8 Mean square error MSE It is an error between the estimated and predicted values
9 Root Mean Square Error RMSE Standard Deviation of difference between estimated and
predicted values
10 Intersection over union IOU IOU is calculated by the union of the overlapping box,
intersection box. Where is the intersection of the box is
divided by the union of the area box and it is defining by
IOU = TP/(FP + TP + FN), where FP, TP, FN are True positive,
False Positive, False Negative respectively
11 Classification accuracy CA The percentage of the total number of the image correctly
classified divided by the number of the image for Testing

ISADH feature extraction produces a higher accuracy rate metric among all available metric. We likewise have found
(>93%) in both RGB and HSV color space and UNSER has the that among all existing machine learning techniques SVM
lowest accuracy rate among all feature extraction. Mean and produce higher accuracy. In future, researcher should have
standard variance performance metric are also used by the fused the various data analysis techniques to improve the
author [91]. performance of the recognition system.
Mean and standard variance are applying to get a perfor-
mance by the author [92] They have also compared the per- 5.3. Performance comparison with another approach.
formance of the proposed system with another metric such
as sensitivity, specification, accuracy, mean and last Square One objective of this paper is to examine how computer
error. The CA metric is used by the author [66] that show vision using various techniques performs in comparison to
good accuracy rate with 96.70%. Various performance accu- other existing methods. In this, we focus on comparative
racy metric such as Precision, FPR, FNR to recognize the dis- analysis between techniques used for the same dataset as
ease in papaya applying various classifiers such as SVM, well as including various performances metric in the same
Decision Tree, Naive Bayes classifier. The results show that paper. Many authors have used machine learning techniques
System produces better accuracy compare to the other to obtain the accuracy of the proposed system. MSVM show
classifier. 93.84% where 86.04% accuracy rate compared to minimum
ARP and ARR performance metric used by the author [65] distance on the same dataset with the same performance
to detect the disease of Soybean. The author [93] use classifi- metric [29], which show 7–8% higher accuracy rate.
cation accuracy metric to identify the fruits images using Evaluation of performance based on different metric
ANN_ABC and ANN_HS techniques and achieved 96.70%, applying three classifier SVM, Decision Tree, Naı̈ve Bayes
94.28% accuracy. The author [74] use precision and recall met- has been used. All this classifier with different metric is
ric to measure the accuracy of the proposed system. He has tested on Papaya dataset. An SVM classifier with the Specifi-
considered 3811 amount of litchi as a dataset. One drawback cation metric has produced highest accuracy and metric such
is that when a number of images are less for training and test- as sensitivity, precision using Naı̈ve Bayes to show worst per-
ing is less than its precision and recall proportionally will be formance by the author [66].
less. The same performance metric has been used by other The author [93] has used KNN, ANN-ABC, ANN-HS classi-
researcher’s [39,76,94,95,96]. fier with CA metric to evaluate the performance of the sys-
Various performance metric has been used to evaluate the tem. ANN-ABC has achieved the highest accuracy rate with
overall accuracy of the system. With the review, it has a value of 96.70% and ANN-HS, KNN produced 94.28%,
concluded that classification accuracy (CA) is most popular 70.88% respectively. The most effective method is deep learn-
196 Information Processing in Agriculture 7 (2 0 2 0) 1 8 3–20 3

ing which is part of machine learning produced 15% higher difference histogram [40]. In the only case, deep learning
accuracy rate compared to another method by the author has shown worst performance with 20% accuracy compared
[42]. Color based method gives better results computer to tex- to other Classifier such as KNN [98]. Table 8 show difference
ture [56]. in performance using the same performance metric in agri-
Precision metric with elimination algorithms is superior culture. Different performance metric is used for recognition
compared than Recall and F-Measure [39]. The accuracy of and classification of fruit and vegetables. Different combina-
K-means classifier is 2–20% higher compared to other KNN, tion of classifier is also additionally implemented on the same
LDA, and SVM. Also, neural network has shown better results dataset in this section. There exist numerous tools and plat-
in many experiments [67]. RIA, RSI based techniques [65], GA form permitting researchers to experiment with computer
and random decision forest-based techniques [95,67], LDA vision for fruits and vegetables. Deep leaning is more popular
Classifier [74], ANN- based algorithms [62] are effective tech- for data analysis now days. In future work, researchers must
niques for identification of the image. Border and interior utilize the current available techniques to achieve good per-
pixel classification [35] also be superior compared to the formance accuracy.
unsupervised learning method [43].In the survey, it has found
that fused descriptor is superior compared to a single with 5.4. Advantage and disadvantage
SVM classifier [19]. Also, Deep learning is getting more atten-
tion from all inter-discipline and intra-discipline domain. The In this Survey, our principle concentration is the accuracy of
author [97] use multimodel deep learning for the detection of recognition and classification of fruits and vegetable and also
the image. detection of disease in fruits and vegetable among the horti-
They have utilized multimodel deep learning on public culture product under agriculture field (see Sections 3.1 and
avail dataset such STISEN, GAIT, Sleep-Stage, Indoor- 3.2). The advantage of computer vision is that provides rapid-
Outdoor. In some case, it is also observed that MSVM showed ness, persistence, and non-destructiveness. Detection of
poor results with GCH descriptor and improved sum and fruits and vegetables [38] the method is robust under a

Table 8 – Performance comparison with another approach.

Year Application in agriculture Performance metric Difference in performance (%) Ref.

2018 Recognition of Papaya disease CA SVM:95.2, Decision tree:88.6 [66]

Naı̈ve Bayes:77.78
Sensitivity SVM:85.6, Decision tree:66.0 [66]
Naı̈ve Bayes:33.33
Specification SVM:97.12, Decision tree:92.00 [66]
Naı̈ve Bayes:86.67
Precision SVM:85.6, Decision tree:60.0 [66]
Naı̈ve Bayes:33.0
False Positive rate SVM:2.88, Decision tree:8.0 [66]
Naı̈ve Bayes:13.3
False negative rate SVM:14.4, Decision tree:40.0 [66]
Naı̈ve Bayes:66.67

2018 Identification of orange varieties CA KNN:70.88, ANN-ABC:96.70 [93]

ANN-HS:94.28

2017 Apple Recognition CA GA:92.30, SVM:87.18 [95]

KNN:80.34, RF:83.76

2017 Branch Recognition CA GA:88.03, SVM:84.61 [95]

KNN:90.59, RF:85.47

2017 Identification of leaf CA GA:80.03, SVM:75.21 [95]

KNN:74.35, RF:77.78

2014 Recognition of leaf image CA KNN:64.4, LDA:67.4, SVM:82.4, [96]

K-Mean:84.6, MSVM:84.6

2011 Fruits recognition and classification CA SVM: GCH + LBP:93.84 [11]

KNN: GCH + CLBP:92.98

2004 Recognition of image Recall Elimination:80.58 (Recall) [39]

Precision Elimination:100 (Precision) [39]
F-measure Elimination:44.4 (Measure) [39]
 Information Processing in Agriculture 7 ( 2 0 2 0 ) 1 8 3 –2 0 3 197

challenging condition such as illumination, different poses, manual selection. One limitation of the system is how to fix
variability, cropping, and partial osculation. Also in the detec- the ideal temperature for storage of grapes.
tion of disease in soybean [30] robust to the image effect with From the literature, it has found that poor administration
blurring, scaling, illumination. of distribution center is one of the key factors for losses. To
The one other advantage of using computer vision in fruits solve this problem, the author [105] has come with a solution
and vegetables in the term of reducing effort/labor. The con- with the Universal Turing Machine. This technique is used to
ventional procedure required adequate time, automatic pro- calculate the effect of cultivar stacking heading, stacking
cessing takes place in computer vision with reducing the bearing, stacking speed, the anxiety and energy proportion.
time. More ever fruits and vegetables are tested shown in The one limitation of the present system that determines
Tables 3 and 4. The accuracy of all framework is generally the properties of only one class of fruits and vegetables. That
high. By the survey, it is seen that SVM classifier is mostly why it is advisable for all researchers to explore this new
used for testing compared to other classifiers such as KNN, trend topic in future and try to provide some solution to mea-
PNN, and ANN etc. And also KNN widely used for the segmen- sure the mechanical properties of various fruits and
tation of fruits and vegetables. The paper [8] indicates that DL vegetables.
has the potential to apply in a variety of agriculture product. Extended to storage issues of quality of fruits and vegeta-
The testing time is also less with DL compared to other meth- bles, the author [106] presents reviews of the application of
ods such as SVM, KNN, PNN etc. internal and external quality analysis of fruits. The various
One other advantage of computer visions is the availability mechanisms have been utilized such as NIRS, Optical tomog-
of a vast range of feature descriptor. The accurate feature of raphy, Multispectral and hyperspectral imaging. These all
fruits and vegetables is more important that is utilized by available techniques have the capability to measure the inter-
the classifier to obtain the accuracy. With the survey, we nal and external quality. In India, due to a large population,
can say that color and texture based descriptor is mostly used there is a huge demand for better quality, safe nutritious
to compare to other descriptors. The other advantage of com- and fresh product without defect. That is why in future there
puter vision in fruits and vegetables is that it detects the dis- is a considerably more requirement for non-damaging esti-
ease earlier. The author [64] classifies the disease available in mation and low-cost techniques for quality assessment.
apple data set, due to that farmer can save the cost of the Another author [107] also uses a non-destructive method
product. to assess the quality of fruits and vegetable. They have
One major disadvantage in the use of Computer Vision is explored the various methodologies to obtain the size of the
the preparation of dataset, that is Hercules task and it is too fruits. They have also explored the application of computer
much time-consuming. Another disadvantage is the high vision to determine the size of the image. A novel approach
computation time to process the image with an effect of is proposed by the author to estimate the feature of potato
osculation. image. In this, they have determined some feature of potato
In the computer vision, a number of tasks such as pre- image that is length and width, thickness, maximum square
processing, segmentation, feature extraction, training, and area, and volume. The experiment results show 93% showed
testing. Each individual step depends on the previous task. 93% accuracy grading in volume thickness modal and 73%
At any step, if we could not able to measurable action we could by area region thickness model.
not able to recognize and classify the fruits and vegetables in Author presents [108] a novel algorithm to estimate the
agriculture fields. One major disadvantage is that fruits and quality of citrus fruits. They have considered volume and
vegetables have more dependability on the environment, mass of the product are two key attributes to assess the qual-
due to that it may possible that same approach of computer ity. The conventional way to deal with to calculate the volume
vision produces different accuracy on the same dataset. and mass is time-consuming. To overcome this problem, an
image processing based technique is used. First, all fruits
6. Computer vision system to evaluate the image is in RGB get converted into HSI. The value of hue will
quality level of fruits and vegetable be probably constant. Water displacement method (WDM) is
taken to calculate the volume and R2 value has been calcu-
Computer vision has rich application and potential to evalu- lated that indicate the mass information of fruits and vegeta-
ate the quality level of fruits and vegetable. Many researchers bles in multi-product sizing product system. To satisfy the
[99–101] have proposed numerous techniques to grade the consumer need we all should adopt the new techniques for
agriculture product based on computer vision. An author quality assessment, sorting and proper management of the
[102] has proposed an automatic detection and classification warehouse. In this, the author [109] has given the solution
system to grade the quality of apple. SVM data analysis tech- to assess the quality based on computer vision. As we are
niques have been used to recognize apple fruits. One advan- all aware that colour is a key attribute to identify the quality.
tage of the proposed system that it consumes less Manual inspection of colour will be time-consuming and have
recognition time. An interesting work presented by author a high chance of error. To overcome this author has given the
[103] for quality inspection. The test has been done dynami- automated system to the recognition of colour.
cally. The experiment results show that 95% image is classi- The author [110] has done the classification of cashew pro-
fied. A quality infection of grape-based on computer vision duct to access the quality by using computer vision. First,
system is presented by author [104]. In the experiment auto- background subtraction and enhancement operation are
matically feature selection give better accuracy compare to done on capture images. They have utilized the morphology
198 Information Processing in Agriculture 7 (2 0 2 0) 1 8 3–20 3

operation to extract the background. In this, three different has been asked in methodology section with efficiency, accu-
class of cashew is present such as whole, split up and split racy, lost cost and user friendly.
down. After that various feature of the product is extracted.
These features are max, min, mean, standard derivation, area
under a curve and finally the length of the curve. Further, all 7. Future scope
the extracted feature used for training and classification. The
system produces a 100% accuracy rate in all three different Within the study, we have mostly covered the various existing
class of cashew. applications of computer vision in fruits and veggies among
Presence of the defect is responsible to decrease the qual- various horticulture products of agricultural fields.
ity. The author [111] presents an efficient and effective We have described in brief about how to extract the fea-
methodology to detect the defect in citrus. They have used ture using the various descriptors. We have additionally
two different classifiers such as neural network and CART examined the various machine learning techniques to find
for training and testing. Both classifier produces good the accuracy of the proposed system. With the existing study,
accuracy rate with a value 98.30% and 93.71% respectively. we have developed the framework for recognition and classi-
Computer vision framework speaks to an appropriate instru- fication of veggies and fruits for future work described in
ment for a quality assessment of fruits and vegetable in agri- Fig. 2.
culture fields. It covers to large domain of agriculture. The In the study, we have observed following future scope
computer vision able to give answer of all question which listed below.

Input in the
form of capture
image

Image of training Image of Testing

Background Subtraction Background Subtraction

Feature Extraction Feature Extraction

Color Texture Color Texture

Training by machine Classification by Machine

learning learning

Recognized the different

categories of fruits and veggies

Fig. 2 – Framework for recognition and classification of veggies and fruits.

 Information Processing in Agriculture 7 ( 2 0 2 0 ) 1 8 3 –2 0 3 199

Consider the different area of dataset under agriculture current technique on the same dataset with the same per-
field. formance metric.

Fused the color, texture, shape descriptor to extract the We have also given concern on the technical aspect of var-
feature. ious frameworks employed, data preprocessing, details of

Same performance metrics can be used in the different descriptors used to extract the feature of images and also
dataset to measure accuracy. detection of a defect of multiple fruits and vegetables using

Recently, deep learning getting more focus of many various techniques. More ever, we also discuss a different
researchers due to its potential to produce high and accu- type of disease present in various fruits and vegetables. Dif-
rate results. This may be one of the future works for recog- ferent performance metric symbols used to calculate the per-
nition and classification problems. formance of the proposed system has been discussed in the
survey. With this survey, SVM gives a better accuracy rate
and classification accuracy is a widely used performance met-
8. Conclusion ric outperforms other metrics.
In future work, we want to use a different descriptor based
In this paper, we have done a survey based on the role of on color, texture, shape, size and fused them to achieve more
computer vision in fruits and vegetables among various hor- accuracy, also apply the deep learning techniques concept to
ticulture products of agricultural fields. We have identified detection and classification. Our main aim of this survey is to
ninety-eight paper related to fruits and vegetables in the motivate the most researchers to gain knowledge of com-
agricultural domain, identify the particular area, the dataset puter vision. Researchers should use this information to
used and various challenges in the agriculture domain. In address the challenge in agriculture fields. The overall advan-
the previous work reported by researcher’s in literature, tage of computer vision is motivating for its future implemen-
quality grading and defect detection of fruit and vegetable tation of modern farming, more accurate and efficient sorting
are done on a single dataset of fruit and vegetables. In this, and monitoring systems.
a generalized framework is proposed to grade the quality
and defect detection of multiple fruits and vegetables. We
Declaration of Competing Interest
have discussed comparative analysis of the performance of
particular single feature like color, texture, shape with fused
The authors declare that they have no known competing
feature. Likewise, a comparative performance study is an
financial interests or personal relationships that could have
analysis of various machines learning method with another
appeared to influence the work reported in this paper.

Appendix A. List of abbreviations

Name Abbreviation Name Abbreviation

ANN Artificial Neural Networks ACC Color Autocorrelogram

ACM Association for Computing Machinery ARP Average Retrieval Precision
ARA Average Retrieval Accuracy ARR Average Retrieval Rate
BIC Border Interior Classification CCH Color Coherence Histogram
CDH Color Difference Vector CCV Color Coherence Vector
CLBP Complete Local Binary Pattern DL Deep Learning
DSI Damage Severity Index EOAC Edge Orientation Autocorrelogram
FNR False Negative Rate FPR False Positive Rate
GDP Gross Domestic Product GCH Global Color Histogram
GLCM Gray Level Co-occurrence Matrix GA Genetic Algorithms
ISHD Improved Sum and Difference Histogram KNN k-Nearest Neighbors
LAS Local Activity Spectrum LTP Local Ternary Patterns
LDA Linear Discriminant Analysis LBP Local binary pattern
LP Linear Polarization LDA Linear Discriminant Analysis
LSB Least Significant Byte LCI Lesion Color Index
MSVM Multiple Support Vector Machine MSB Most Significant Byte
PPP Public Private Partnership PR Progressive Randomization
PNN Probabilistic Neural Network QCCH Quantized Compound Change Histogram
RIA Ratio of Infected Area RNN Recurrent Neural Networks
RA Regression Analysis RSI Severity Index for Rust
SD Standard derivation SEH Structure Element Histogram
SVM Support Vector Machine SSLBP Scale selective Local binary pattern
SIFT Scale Invariant Feature Transform TFRS Thai Fruit Recognition System
WBF Wavelet Based Filtering WDH Wavelet Decomposed Color Histogram
200 Information Processing in Agriculture 7 (2 0 2 0) 1 8 3–20 3

Appendix B. Publicly-available data-sets related to agriculture

No. Organization/data-set Description of data-sets Source

1 Supermarket Data-set 15 different categories of fruits http://www.ic.unicamp.br/

and vegetables ~rocha/pub/downloads/tropical-
fruits-DB-1024x768.tar.gz.
2 Image-Net Data-set Image of various plants (trees, http://image-net.org/explore?
Vegetables, flowers) wnid=n07707451
3 Leaf snap Data-set Leaves of 185 tree species http://leafsnap.com/data-set/
4 Life CLEF Data-set Geographical and use of plants http://www.imageclef.org/2014/
lifeclef/plant
5 Image-Net Large Scale Visual The image that allows image http://imagenet.org/challenges/
Recognition challenge (ILSVRC) localization LSVRC/2017/#det
6 Crop/weed Field Image Data-set Crop/weed images https://github.com/cwfid/data-
set
7 Flavia leaf Data-set Leaf image of 32 plants https://flavia.sourceforge.net/
8 Syngenta crop Challenge 2017 Corn images https://ideaconnection.com//
syngenta-crop-Challenge/
challenge.php
9 Plant data-set Herbicide injury image data-set https://plants.uaex.edu/
herbicide
https://www.uaex.edu/yard-
garden/resource-libary/disease/
10 Malaya Kew Data-set. Image of leaves from 44 species https://web.fsktm.um.edu.my/
class ~cschan/downloads_
mkleaf_data-set.html

R E F E R E N C E S agriculture industry. In: Proc ICCCV ’16 Proceedings of the

2016 International Conference on Communication,
Computing and Virtualization, Mumbai, India; 2016. p. 426–
433.
[1] The economic times. GDP data; 2018. Link <https://
[13] Jhawar J. Orange sorting by applying pattern recognition on
economictimes.indiatimes.com/news/economy/indicators/
colour image. In: Proc. ICISP ’15 Proceedings of the 2015
view-being-the-worlds-6th-largest-economy-means-little-
International Conference on Information Security & Privacy.
for-indias-future/articleshow/64966415.cms>.
Nagpur, India; 2016. p. 691–697.
[2] India at a glance. FAO in India. Food and Agriculture
[14] Vasconez JP, Kantor GA, Auat FA. Human -robot interaction
Organization of the United Nations; 2018. Link: <http://
in agriculture: a survey and current challenges. Biosyst Eng
www.fao.org/india/fao-in-india/india-at-a-glance/en/>.
2018;179:35–48.
[3] Singh A, Ganapathysubramanian B, Singh AK, Sarkar S.
[15] Nouri-ahmadabadi H, Omid M, Mohtasebi SS. Design,
Machine learning for high-throughput stress phenotyping in
development and evaluation of an online grading system for
plants. Trends Plant Sci 2016;21(2):110–24.
peeled pistachios equipped with machine vision technology
[4] Mamta S, Hemanga B, Bhawna T, Sweta J, Moreshwar K,
and support vector machine. Inf Process Agric 2017;4
Ranbir S, Pankaj G. Horticultural statistics at a glance. Link:
(4):333–41.
<http://nhb.gov.in/statistics/Publication/Horticulture%20At
[16] Iqbal Z, Attique M, Sharif M, Hussain J, Habib M, Javed K. An
%20a%20Glance%202017%20for%20net%20uplod%20(2).
automated detection and classification of citrus plant
pdf>.
diseases using image processing techniques: a review.
[5] Press Information Bureau Government of India. ECONOMIC
Comput Electron Agric 2018;153:12–32.
SURVEY; 2018. Link: <http://www.pib.nic.in/indexd.aspx>.
[17] Rehman TU, Mahmud S, Chang YK, Jin J, Shin J. Current and
[6] India Population. Worldometers n.d data; 2018. Link: <http://
future applications of statistical machine learning
www.worldometers.info/world-population/india-
algorithms for agricultural machine vision systems. Comput
population/>.
Electron Agric 2019;156:585–605.
[7] Literates and Literacy Rates; 2018. Link: <http://www.nlm.
[18] Mahendran R, Gc J, Alagusundaram K. Application of
nic.in/literacy01_nlm.htm/>.
computer vision technique on sorting and grading of fruits
[8] Kamilaris A, Prenafeta-Boldú FX. Deep learning in
and vegetables. J Food Process Technol 2012;3(8):1–7.
agriculture: a survey. Comput Electron Agric 2018;147:70–90.
[19] Liming X, Yanchao Z. Automated strawberry grading system
[9] Bhargava A, Bansal A. Fruits and vegetables quality
based on image processing. Comput Electron Agric 2010;71
evaluation using computer vision : a review. J King Saud
(S1):S32–9.
Univ – Comput Inf Sci 2018;1–15.
[20] El-ramady HR, Domokos-szabolcsy E, Abdalla NA, Taha HS,
[10] Moallem P, Serajoddin A, Pourghassem H. Computer vision-
Fari M, editors. Sustainable agriculture reviews: postharvest
based apple grading for golden delicious apples based on
management of fruits and vegetables storage. Switzerland:
surface features. Inf Process Agric 2017;4(1):33–40.
Springer Cham; 2013. p. 65–152.
[11] Al Ohali Y. Computer vision based date fruit grading system:
[21] Cubero S, Lee WS, Aleixos N, Albert F, Blasco J. Automated
design and implementation. J King Saud Univ – Comput Inf
systems based on machine vision for inspecting citrus fruits
Sci 2011;23(1):29–36.
from the field to postharvest—a review. Food Bioprocess
[12] Arakeri MP, Lakshmana. Computer vision based fruit
Technol 2016;9(10):1623–39.
grading system for quality evaluation of tomato in
 Information Processing in Agriculture 7 ( 2 0 2 0 ) 1 8 3 –2 0 3 201

[22] Erkan M, Yildirim I. editors. Minimally processed [41] Faria FA, Dos Santos JA, Rocha A, Da S. Torres R. Automatic
refrigerated fruits and vegetables. Postharvest Quality and classifier fusion for produce recognition. In: Proc. SIBGRAPI
Safety of Fresh-Cut Vegetables, USA: Springer; 2002. p. 271– ’12 Proceedings of the 2012 IEEE Conference on Graphics,
326. Patterns and Images, Ouro Preto, Brazil; 2012. p. 252–9.
[23] Mditshwa A, Magwaza LS, Tesfay SZ, Mbili N. Postharvest [42] Fei-Fei L, Fergus R, Perona P. One-shot learning of object
quality and composition of organically and conventionally categories. IEEE Trans Pattern Anal Mach Intell 2006;28
produced fruits: a review. Sci Hortic (Amsterdam) (4):594–611.
2017;216:148–159. [43] Heidemann G. Unsupervised image categorization. Image
[24] Brasil IM, Siddiqui MW, editors. Preharvest modulation of Vis Comput 2005;23(10):861–76.
postharvest fruit and vegetable quality: quality of fruits and [44] Jurie F, Triggs B. Creating efficient codebook for visual
vegetables: an overview. UK: Academic Press Elsevier; 2017. recognition. In: Proc. ICCV ’05 proceedings of the 2005 IEEE
p. 1–40. international conference on computer vision, Beijing,
[25] Ntsoane ML, Zude-Sasse M, Mahajan P, Sivakumar D. China; 2005. p. 604–10.
Quality assessment and postharvest technology of mango: a [45] Marszaek M, Schmid C. Spatial weighting for bag-of-
review of its current status and future perspectives. Sci features. In: Proc. CVPR ’06 Proceedings of the 2006 IEEE
Hortic (Amsterdam) 2019;249:77–85. computer society conference on computer vision and
[26] Sofu MM, Er O, Kayacan MC, Cetis B. Design of an automatic pattern recognition, San Francisco, California; 2010, p. 2118–
apple sorting system using machine vision. Comput 2125.
Electron Agric 2016;127:395–405. [46] Fakhri A Nasir A, Nordin M Rahman A, Rasid Mamat A. A
[27] Kamilaris A, Kartakoullis A, Prenafeta-Boldu FX. A review on study of image processing in agriculture application under
the practice of big data analysis in agriculture. Comput high performance computing environment. Int J Comput Sci
Electron Agric 2017;143:23–37. Telecommun 2012;3(8):16–24.
[28] Ishimwe R, Abutaleb K, Ahmed F. Applications of thermal [47] Rocha A, Goldenstein S. PR: More than meets the eye. In:
imaging in agriculture—a review. Adv Remote Sens 2014;3 Proc.ICCV ’ 07 Proceedings of the 2007 IEEE international
(3):128–40. conference on computer vision. Rio de Janeiro, Brazil; 2007.
[29] Dubey SR, Jalal AS. Application of image processing in fruit p. 1–8.
and vegetable analysis: a review. J Intell Syst 2015;24 [48] Weber M. Unsupervised learning of models for recognition.
(4):405–24. Doctor thesis. California Institute of Technology; 2000. Link:
[30] Shrivastava S, Singh SK, Hooda DS. Soybean plant foliar <http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.
disease detection using image retrieval approaches. 336.5737&rep=rep1&type=pdf>.
Multimed Tools Appl 2017;76(24):26647–74. [49] Pornpanomchai, Chomtip, Khomkhwan Srikeaw, Voranun
[31] Saxena L, Armstrong L. A survey of image processing Harnprasert KP. Thai Fruit Recognition System (TFRS). In:
techniques for agriculture. In: Proc. AFITA ’14 Proceedings of Proc. ICIMCS ’09 proceedings of the 2009 proceedings of the
the 2014 Asian Federation for Information Technology in first international conference on internet multimedia
Agriculture, Perth, Australia; 2014.p. 401–13. computing and service. New York, USA, 2009. p. 108–112.
[32] Dubey SR, Jalal AS. Robust approach for fruit and vegetable [50] Nosseir A, Eldin S, Ahmed A. Automatic identification and
classification. In: Proc. ICMOC ’12 Proceedings of the 2012 classifications for fruits using k-NN. In: Proc ICSIE ’18
International Conference on Modelling, Optimization and Proceedings of the 2018 ACM international conference on
Computing, Tamil Nadu, India; 2012. p. 3449–53. software and information engineering Cairo, Egypt; 2018. p.
[33] Lecun Y, Bengio Y, Hinton G. Deep learning. Nature 62–67.
2015;521:436–44. [51] Pham C, Jackson D, Schoning J, Bartindale T, Plotz T, Olivier
[34] Pass G, Zabih R, Miller J. Comparing images using color P. Food board : surface contact imaging for food recognition.
coherence vectors. In: Proc ICOM ’96 Proceedings of the 1996 In: Proc. UbiComp ’13 proceedings of the 2013 ACM
ACM international conference on multimedia, New York, international joint conference on pervasive and ubiquitous
USA; 1996. p. 65–73. computing .Zurich, Switzerland; 2013. p. 749–752.
[35] Stehling RO, Nascimento MA, Falcao AX. A compact and [52] Fan S, Wang X. Vegetation recognition based on deep
efficient image retrieval approach based on border/interior learning with feature fusion. In: Proc. ICAIP ’17 proceedings
pixel classification. In: Proc. CIKM ’02 Proceedings of the of the 2017 ACM international conference on advances in
2002 ACM CIKM International Conference on Information image processing, Bangkok, Thailand; 2017. p. 19–23.
and Knowledge Management. McLean VA, USA; 2002. p. 102– [53] Patil O, Gaikwad PV. Classification of vegetables using tensor
109. flow. Int J Res Appl Sci Eng Technol 2018;6(4):2926–34.
[36] Bolle RM, Connell JH, Haas N, Mohan R, Taubin G. [54] Sundararajan K, Woodard DL. Deep learning for biometrics:
Veggie vision: a produce recognition system. In: Proc. a survey. ACM Comput Surv 2018;51(3):1–34.
WACV ’96 Proceedings of the 1966 IEEE Workshop on [55] Zhong S-H, Liu Y, Hua KA. Field effect deep networks for
Applications of Computer Vision, Sarasota, FL, USA; image recognition with incomplete data. ACM Trans
1996. p. 244–251. Multimed Comput Commun Appl 2016;12(4):1–22.
[37] Rocha A, Hauagge DC, Wainer J, Goldenstein S. Automatic [56] Jana S, Basak S, Parekh R. Automatic fruit recognition from
fruit and vegetable classification from images. Comput natural images using color and texture features. In: Proc.
Electron Agric 2010;70(1):96–104. devLC ’17 Proceedings of the 2017 IEEE conference on
[38] Dubey SR, Jalal AS. Species and variety detection of fruits Devices for Integrated Circuit, Kalyani, India; 2017. P. 620–4.
and vegetables from images. Int J Appl Pattern Recognit [57] Li D, Shen M, Li D, Yu X. Green apple recognition method
2013;1(1):108–26. based on the combination of texture and shape features. In:
[39] Agarwal S, Awan A, Roth D. Learning to detect objects in Proc. ICMA ’17 Proceedings of the 2017 IEEE International
images via a sparse, part-based representation. IEEE Trans Conference on Mechatronics and Automation. Takamatsu,
Pattern Anal Mach Intell 2004;26(11):1475–90. Japan; 2017. p. 264–9.
[40] Dubey SR, Jalal AS. Fruit disease recognition using improved [58] Purohit S, Viroja R, Gandhi S, Chaudhary N. Automatic plant
sum and difference histogram from images. Int J Appl species recognition technique using machine learning
Pattern Recognit 2014;1(2):199–220. approaches. In: Proc. CoCoNet ’15 proceedings of the 2015
202 Information Processing in Agriculture 7 (2 0 2 0) 1 8 3–20 3

IEEE international conference on computing and network [73] Padol PB, Yadav AA. SVM classifier based grape leaf disease
communications. Trivandrum, India; 2015. p. 710–9. detection. In: Proc. CASP ’16 proceedings of the 2016 IEEE
[59] Rachmawati E, Supriana I, Khodra ML. Toward a new conference on advances in signal processing, Pune, India;
approach in fruit recognition using hybrid rgbd features and 2016. p. 175–179.
fruit hierarchy property. In: Proc. EECSI ’17 Proceedings of [74] He ZL, Xiong JT, Lin R, Zou X, Tang LY, Yang ZG. A method of
the 2017 international conference on electrical engineering, green litchi recognition in natural environment based on
Computer Science and Informatics, Yogyakarta, Indonesia; improved LDA classifier. Comput Electron Agric
2017. p. 202–7. 2017;140:159–67.
[60] Ye FEI, Lou X, Han MIN. Evolving support vector machine [75] Faria FA, Dos Santos JA, Rocha A, Torres RDS. A framework
using modified fruit fly optimization algorithm and genetic for selection and fusion of pattern classifiers in multimedia
algorithm for binary classification problem. In: Proc. recognition. Pattern Recognit Lett 2014;39:52–64.
ICCWAMTIP ’16 Proceedings of the 2016 IEEE International [76] Jawale D, Deshmukh M. Real time automatic bruise
Computer Conference on Wavelet Active Media Technology detection in (Apple) fruits using thermal camera. In: Proc.
and Information Processing .Chengdu, China; 2016. p. 38–46. ISSCP ’17 proceedings of the 2017 IEEE international
[61] Wajid A, Singh NK, Junjun P, Mughal MA. Recognition of conference on communication and signal processing.
ripe, unripe and scaled condition of orange citrus based on Chennai, India; 2017. p. 1080–1085.
decision tree classification. In: Proc. iCoMET ’18 proceedings [77] Tripathi MK, dr. Maktedar D. A framework with OTSU’ S
of the 2018 IEEE international conference on computing, thresholding method for fruits and vegetables image
mathematics and engineering technologies, Sukkur, segmentation. Int J Comput Appl 2018;179(52):25–32.
Pakistan; 2018. p. 1–4. [78] Yuan Y, Liu Y, Dai G, Zhang J, Chen Z. Automatic foreground
[62] Zeng G.Fruit and Vegetables classification system using extraction based on difference of Gaussian. Sci World J.
image saliency and convolutional neural network. In: Proc. 2014;2014:1–9.
ITOEC ’17 proceedings of the 2017 IEEE information [79] Li J, Chen L, Huang W. Detection of early bruises on peaches
technology and mechatronics engineering conference, (Amygdalus persica L.) using hyperspectral imaging coupled
Chongqing, China; 2017. p. 613–7. with improved watershed segmentation algorithm.
[63] Roberts Michael J, Schimmelpfennig David E., Ashley Postharvest Biol Technol 2018;135:104–13.
Elizabeth, Livingston Michael J, Ash Mark S, Vasavada U. The [80] De Smet P. Optimized high speed pixel sorting and its
value of plant disease early-warning systems: a case study application in watershed based image segmentation.
of USDA’s soybean rust coordinated framework. Econ Res Pattern Recognit 2010;43(7):235923–66.
Rep 7208;2006:1–38. [81] Marmanis D, Schindler K, Wegner JD, Galliani S, Datcu M,
[64] Dubey SR, Jalal AS. Apple disease classification using color, Stilla U. Classification with an edge: improving semantic
texture and shape features from images. Signal Image Video image segmentation with boundary detection. ISPRS J
Process 2015;10(5):819–26. Photogramm Remote Sens 2018;135:158–72.
[65] Shrivastava S, Kumar S, Hooda DS. Color sensing and image [82] Wang X, Deng Y, Duan H. Edge-based target detection for
processing-based automatic soybean plant foliar disease unmanned aerial vehicles using competitive Bird Swarm
severity detection and estimation. Multimed Tools Appl Algorithm. Aerosp Sci Technol 2018;78:708–820.
2015;74(24):11467–84. [83] Caraiman S, Manta VI. Histogram-based segmentation of
[66] Habib MT, Majumder A, Jakaria AZM, Akter M, Uddin MS, quantum images. Theor Comput Sci 2014;529:46–60.
Ahmed F. Machine vision based papaya disease recognition. [84] Qin K, Xu K, Liu F, Li D. Image segmentation based on
J King Saud Univ – Comput Inf Sci 2018. histogram analysis utilizing the cloud model. Comput Math
[67] Liu J, Sui Y, Wisniewski M, Droby S, Liu Y. Review: Utilization with Appl 2011;62(7):2824–33.
of antagonistic yeasts to manage postharvest fungal [85] Siang Tan K, MatIsa NA. Color image segmentation using
diseases of fruit. Int J Food Microbiol 2013;167(2):153–60. histogram thresholding Fuzzy C-means hybrid approach.
https://doi.org/10.1016/j.ijfoodmicro.2013.09.004. Pattern Recognit 2011;44(1):1–15.
[68] Moshou D, Bravo C, Wahlen S, West J, McCartney A, De [86] Van De Sande K, Gevers T, Snoek C. Evaluating color
Baerdemaeker J. Simultaneous identification of plant descriptors for object and scene recognition. IEEE Trans
stresses and diseases in arable crops using proximal optical Pattern Anal Mach Intell 2010;32(9):1582–96.
sensing and self-organising maps. Precis Agric 2006;7 [87] Amanatiadis A, Kaburlasos VG, Gasteratos A, Papadakis SE.
(3):149–64. Evaluation of shape descriptors for shape-based image
[69] Dhakate M, Ingole AB. Diagnosis of pomegranate plant retrieval. IET Image Process 2011;5(5):493–9.
diseases using neural network. In: Proc. NCVPRIPG ’15 [88] Lowe DG. Distinctive image features from scale invariant
Proceedings of the 2015 National Conference on Computer keypoints. Int J Comput Vis 2004;60(2):91–110.
Vision, Pattern Recognition, Image Processing and Graphics, [89] Tuytelaars T, Mikolajczyk K. Local invariant feature
Patna, Bihar; 2015. p. 1–4. detectors: a Survey. Found TrendsÒ Comput Graph Vis
[70] Samajpati BJ, Degadwala SD. Hybrid approach for apple fruit 2007;3(3):177–280.
diseases detection and classification using random forest [90] Dubey SR, Jalal AS. Adapted approach for fruit disease
classifier. In: Proc ICCSP ’16 proceedings of the 2016 identification using images. Int J Comput Vis Image Process
international conference on communication and signal 2012;2(3):44–58. https://doi.org/10.4018/ijcvip.2012070104.
processing Melmaruvathur, India; 2016. p. 10015–1019. [91] Choi D, Lee WS, Ehsani R, Schueller J, Roka FM. Detection of
[71] Tan DS, Leong RN, Laguna AF, Ngo CA, Lao A, Amalin D. A dropped citrus fruit on the ground and evaluation of decay
framework for measuring infection level on cacao pods. In: stages in varying illumination conditions. Comput Electron
Proc. TENSYMP ’16 Proceedings of the 2016 Region 10 Agric 2016;127:109–19.
Symposium, Bali, Indonesia; 2016. p. 384–389. [92] Kumar RA, Rajpurohit VS, Nargund VB. A neural network
[72] Sharif M, Attique M, Iqbal Z, Faisal M, Ullah MI, Younus M. assisted machine vision system for sorting pomegranate
Detection and classification of citrus diseases in agriculture fruits. In: Proc. ICECCT ’17 Proceedings of the 2017 Second
based on optimized weighted segmentation and feature International Conference on Electrical, Computer and
selection. Comput Electron Agric 2018;150:220–34. Communication Technologies. Coimbatore, India; 2017. p. 1–9.
 Information Processing in Agriculture 7 ( 2 0 2 0 ) 1 8 3 –2 0 3 203

[93] Sabzi S, Abbaspour-Gilandeh Y, Garcı́a-Mateos G. A new [103] Benalia S, Cubero S, Prats-Montalbán JM, Bernardi B,
approach for visual identification of orange varieties using Zimbalatti G, Blasco J. Computer vision for automatic
neural networks and metaheuristic algorithms. Inf Process quality inspection of dried figs (Ficus carica L.) in real-time.
Agric 2018;5(1):162–72. Comput Electron Agric 2016;120:17–25.
[94] Singh N, Dubey SR, Dixit P, Gupta JP. Semantic image [104] Pietro Cavallo D, Cefola M, Pace B, Logrieco AF, Attolico G.
retrieval by combining color, texture and shape features. In: Non-destructive and contactless quality evaluation of table
Proc. ICCS ’12 proceedings of the 2012 IEEE international grapes by a computer vision system. Comput Electron Agric
conference on computing sciences. Phagwara, India; 2012. p. 2019;156:558–64.
116–120. [105] Jafari Malekabadi A, Khojastehpour M, Emadi B, Golzarian
[95] Tao Y, Zhou J. Automatic apple recognition based on the MR. Development of a machine vision system for
fusion of color and 3D feature for robotic fruit picking. determination of mechanical properties of onions. Comput
Comput Electron Agric 2017;142:388–96. Electron Agric 2017;141:131–9.
[96] Shao MW, Du JX, Wang J, Zhai CM. Recognition of leaf image [106] Magwaza LS, Opara UL, Nieuwoudt H, Cronje PJR, Saeys W,
set based on manifold-manifold distance. In: Proc. ICIC ’14 Nicolaı̈ B. NIR spectroscopy applications for internal and
Proceedings of the 2014 international conference on external quality analysis of citrus fruit – a review. Food
intelligent computing. Nanning, China; 2014. p. 332–7. Bioprocess Technol 2012;5:425–44.
[97] Radu V, Tong C, Bhattacharya S, Lane ND, Mascolo C, Marina [107] Moreda GP, Ortiz-Cañavate J, Garcı́a-Ramos FJ, Ruiz-Altisent
MK. Multimodal deep learning for activity and context M. Non-destructive technologies for fruit and
recognition. In: Proc. IMWUT ’18 proceedings of the 2018 vegetable size determination – a review. J Food Eng
ACM on interactive, mobile, wearable and ubiquitous 2009;92:119–36.
technologies. New York, USA; 2018. p. 1–27. [108] Omid M, Khojastehnazhand M, Tabatabaeefar A.
[98] De Los Reyes A, Augenstein TM, Wang M, Thomas SA, Estimating volume and mass of citrus fruits
Drabick DAG, Burgers DE. The validity of the multi- by image processing technique. J Food Eng 2010;100:
informant approach to assessing child and adolescent 315–21.
mental health. Psychol Bull. 2015;141(4):858–900. [109] Pace B, Pietro Cavallo D, Cefola M, Colella R, Attolico G.
[99] Ilic ZS, Fallik E. Light quality manipulation improves Adaptive self-configuring computer vision system for
vegetable quality at harvest and postharvest: a review. quality evaluation of fresh-cut radicchio. Innovative Food
Environ Exp Bot 2017;139:79–90. Sci Emerg Technol 2015;32:200–7.
[100] El Khaled D, Castellano NN, Gazquez JA, Garcı́a Salvador RM, [110] Sunoj S, Igathinathane C, Jenicka S. Cashews whole and
Manzano-Agugliaro F. Cleaner quality control system using splits classification using a novel machine vision
bioimpedance methods: a review for fruits and vegetables. J approach. Postharvest Biol Technol 2018;138:19–30. https://
Cleaner Prod 2017;140:1749–62. doi.org/10.1016/j.postharvbio.2017.12.006.
[101] Lin H, Ying Y. Theory and application of near infrared [111] Gomez-Sanchis J, Martı́n-Guerrero JD, Soria-Olivas E,
spectroscopy in assessment of fruit quality: a review. Sens Martı́nez-Sober M, Magdalena-Benedito R, Blasco J.
Instrum Food Qual Saf 2009;3:130–41. Detecting rottenness caused by Penicillium genus fungi in
[102] Ji W, Zhao D, Cheng F, Xu B, Zhang Y, Wang J. Automatic citrus fruits using machine learning techniques. Expert Syst
recognition vision system guided for apple harvesting robot. Appl 2012;39:780–5.
Comput Electr Eng 2012;38:1186–95.