Leaf Disease Detection of Multiple Plants using Deep Learning

CHAPTER-1

PREAMBLE

1.1 INTRODUCTION

In the realm of agriculture and plant sciences, a groundbreaking innovation

was on the horizon. A team of brilliant researchers and technologists had come
together to develop a revolutionary system: Leaf Disease Detection of Multiple Plants
using Deep Learning. This cutting-edge technology aimed to transform the way plant
diseases were identified and treated. With a focus on sustainability and crop health,
the team worked limitless to train a powerful deep learning algorithm to analyze and
interpret the subtlest changes in plant leaves. This innovative solution promised to
detect diseases with broad aspect of plants, from fruit-bearing trees to delicate flowers
and staple crops. The implications were profound. Farmers and gardeners worldwide
would soon have access to a reliable and efficient tool that could identify diseases
early, preventing the spread of infection and potential crop devastation. This
development marked a significant step forward in safeguarding global food security
and preserving the beauty of nature's bountiful gardens.

CONVULUTIONAL NEURAL NETWORK

Deep learning is a subfield of AI and ML that makes use of ANNs to learn

new material. During training, deep learning methods extract their characteristics and
split them up for categorization. Deep learning has several uses, such as in computer
vision, image categorization, restoration, voice recognition, video analysis, etc.

Nominal procedure in CNN makes detection and classification easy. Its multi-
layered design makes it effective at analyzing visual data and identifying key
elements for further analysis. Figure 1 depicts the four layers that make up a CNN.
Given as:

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 1
 Leaf Disease Detection of Multiple Plants using Deep Learning

Convolutional Layer
For learning biases and weights, convolutional layers save output of kernels
by previous layer. Optimization function's goal is to produce kernels which faithfully
reflect the data. This layer performs a series of calculations to get feature map from
the input picture.In Fig. 2, we see convolution layer in action, taking in 5x5 picture
and outputting 3x3 filter after scaling it down. Illustration also illustrates filter's
movement, which begins in top left corner of input image and progresses clockwise.
After multiplying every step's values by filter's, final values are sum of all steps'
values. Supplied picture is used to generate new, smaller matrix..

Pooling Layer
This layer decreases the neuron size in down sampling layer, which helps to
prevent overfitting. Fig. 3 depicts a sample of pooling process. This layer controls
overfitting & shortens training time while decreasing number of parameters used
in model [20]. If a model performs perfectly using training data but only 50% of time
using test data, it is considered overfit. For reducingquantity of dimensions into
feature maps, we used ReLU using max pooling. [21].

Activation Layer
Uses ReLU activation layer in every convolution layer. To avoid over fitting,
it even utilises dropout layers.

Fully Connected Layer

Input for classifier is based on results of this layer's examination of class

probabilities. In this layer, we apply popular Softmax classifier to problem of
sugarcane disease detection & classification.

India mostly depends upon agriculture. The options available to farmers for
choosing the most productive fruit and vegetable crops are many. Aim of this study is
to construct a state-of-the-art computational system for disease identification utilizing
photographs of different leaf spots. Using a mobile digital camera and image growing
for processing, we get images of the leafy sports field, which we subsequently utilize
for training and testing classification algorithms. The system's underlying technology

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 2
 Leaf Disease Detection of Multiple Plants using Deep Learning

is a hybrid of image processing and high-end computer methods. Farming lays the
foundation for all countries. The goal of agricultural research is to improve yield and
quality while decreasing costs and increasing revenue. Plant diseases pose a threat to
the agricultural product's quality. Pathogens including fungus, bacteria, and viruses
are responsible for these illnesses. Hence, it’ssignificant to identify the plant disease
and categorize it early on. Experts' continual surveillance of farmers might be too
costly and time-consuming to be practical. Numerous methods, using image
processing and other automated categorization technologies, have been suggested to
either eradicate or significantly mitigate such concerns, success depending on the
specific use cases.

The following are some examples of when image analysis might be useful:

 Identifying plant, stem, and fruit pathogens.

 The goal is to quantify disease-affected territory.
 To locate limits of the impacted region.
 To identify afflicted region's color.
 To ascertain leaf dimensions and form.
 To appropriately determine what Object is. Etc.

Controlling an epidemic of disease is no easy feat. Leaves and stems

are common sites where plant diseases manifest. The intricacy of visual patterns
prevents accurate measurement of illnesses, pests, and features detected visually.
Therefore, there exists a growing need for better and more nuanced recognition of
picture patterns. Leaf spot diseases come in many forms.

 Bacterial
 Fungal
 Viral

IRJMT, 1(5), 2019, 8-16, Sathya.V et al. need professional supervision at all
times, which may be too costly and time-consuming. Numerous methods, using image
processing and other automated categorization technologies, have been suggested to
either eradicate or significantly reduce suchconcerns, output is as per specific use

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 3
 Leaf Disease Detection of Multiple Plants using Deep Learning

cases.

The following are some examples of when image analysis might be useful:
 Identifying plant, stem, and fruit pathogens.
 The goal is to quantify disease-affected territory.
 To locate limits of impacted region.
 To identify afflicted region's color.
 To ascertain leaf dimensions and form.
 To appropriately determine what Object is. Etc.

Controlling an epidemic of disease is no easy feat. Leaves and stems are the
most common sites where plant diseases manifest. The intricacy of visual patterns
prevents accurate measurement of illnesses, pests, and features detected visually.
Therefore, growing need for better and more nuanced recognition of picture patterns.
Leaf spot diseases come in many forms.

 Bacterial
 Fungal
 Viral

Different Diseases, Figure 1 Fungi, bacteria, and viruses are the usual suspects
in the causation of leaf diseases. The reproductive structures of fungi are crucial to
their morphological identification. The life cycles of bacteria are lesser complicated
those of fungus. Bacteria, with rare exceptions, by splitting into two identical copies,
a process termed binary fission. Protein and genetic material with no linked protein
make up viruses, which are incredibly small particles. It’s uncommon for biological
scientists to produce hundreds of photos during a single experiment. Images may be
necessary for investigations including the classification of lesions, the scoring of
quantitative features, the determination of insect-eaten area, etc. Nearly every
responsibilities is handled either manually or with various pieces of software. It takes
a long time to process, and various people will interpret the results differently. For
doing high throughput studies, plant biologists need powerful computational tools that
can automatically extract and evaluate relevant data. In this case, image processing is

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 4
 Leaf Disease Detection of Multiple Plants using Deep Learning

crucial. This work includes image processtechniquesto identify leaf diseases. The
problem of fruit infections has arisen due todamage they may do to the quality and
yield of agricultural goods. Recognition The computer vision algorithms used by
healthy and diseased apple processing considered like color, shape, and texture.
Currently, specialists' naked-eye inspections are the gold standard for spotting
abnormal and diseased apples. However, technique is costly, time-consuming, and
requires a lot of specialized knowledge, so it's not always feasible. The early
identification and treatment of apple illnesses may be probableby using machine
learning based on detection and recognition of apple and its disorders. A risky factor
that reduces production and lowers quality of underdeveloped goods is fruit
infections. Several symptoms and behaviors associated with apple illnesses are
consistent and easily recognizable.

Hence, it’sessential to have a system providingintellectual assistance for the

Prevention process based on human expertise. The roundness, shape, and color values
represented by the extracted features aid in identifying apples and their structure. To
classify patterns in a sequential fashion, a pattern recognition system uses a number of
characteristics, such as entropy, color, shape, and border size. Other disciplines,
including as instruction, food packaging image retrieval, and plant science research,
could be advantageouswith this technique for image process tool for object
categorization and detection of normal and diseased fruit issues.

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 5
 Leaf Disease Detection of Multiple Plants using Deep Learning

1.2 PROBLEM STATEMENT

The problem of plant diseases has arisen because of damage they may do to
agricultural output, both in terms of quality & quantity. Because of potential benefits
of automatically inspecting indications of problems as soon as they occur on plant
leaves, automated identification of plant illnesses is crucial subject of study. The
suggested system is an automated software method for diagnosing plant leaf diseases.
The four primary points of this plan are as follows.The process begins with creating
the color transformation framework for input RGB image, then moving on to masking
while removing green pixels utilizing threshold value, categorization,
computing texture data analysis for user's segment, and finally running classifier
on extracted features. Disease diagnosis and treatment options are both considered for
classifier.

1.3 OBJECTIVE OF STUDY

 The goal of research is to develop a system for precise detection of agricultural

diseases & pests.

 Apply CNN algorithm to data set and generate model for prediction.

 Detection of plant sickness from given input and image and display disease.

 Classifying images of leaves into diseased categories as per pattern of the defect.

1.4 SCOPE OF PROJECT

Disease detection often involves a visual check by farmers or professionals.

This approach, however, is cumbersome and often wrong. This study may provide a
method for employing a deep CNN to identify plant illnesses. With little training,
CNN can quickly and correctly identify leaf diseases in plants. This method
guarantees more precise results.

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 6
 Leaf Disease Detection of Multiple Plants using Deep Learning

1.5 METHODOLOGY USED

CNN Model Steps:

 Conv2D:Its Convolution Layer in which a specified number of kernel filters convolve

the input picture while simultaneously shifting kernel by a stride of size = 2 or 3.

 Max pooling: Its pooling procedure that selects top feature from the feature map area

that is hidden from view by the filter. Therefore, a feature map consisting of the most

salient characteristics from the preceding feature map would be the result of using the

max-pooling level.

 Flattening: A 'Flatten' layer sits within convolutional layer & fully linked layer. A

two-dimensional feature matrix is "flattened" into an array and sent with fully linked

layer.

 Epochs: In machine learning, an epoch is a unit of measurement for quantity of times

an algorithm has iterated through the whole training dataset. While handling huge of

data, it is common practice to process the data in batches.

 Training and Testing Model—Data is processed, including image scaling, cropping,

and array conversion. The test picture further undergoes processing inalike vein. To

evaluate the efficacy of the program, it is common practice to randomly choose a

picture from collecting over 32,000 photographs of plant leaves.In this setup, Train&

Validate sections each get 30% of the budget.

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 7
 Leaf Disease Detection of Multiple Plants using Deep Learning

CHAPTER-2

LITERATURESURVEY

1) T.Vijaykanth Reddy, et.al. “DLDPF with Transfer Automatic Leaf Disease

Detection”, 2021 ICCMC 2021, The agricultural sector in India is crucial to the

country's economy and overall development. Precision Agriculture (PA) is a

relatively new technological advancement in agriculture. However, there are

major enhancements as a result of technological developments. The advent of

deep learning as a component of AI has paved the way for the incorporation of

technology into farming for the first time. Leaf disease detecting is one task that

may be mechanized. This field of study has gained a lot of attention due to

scalability of cloud computing resources. Current CNN-based methods have flaws

in how they handle adaptability and reusing previously acquired knowledge. In

this research, we bridge that gap by using CNN with deep models previously pre-

trained by transfer learning. Combining CNN technology with AlexNet and

GoogLeNet cascade inception has resulted in the proposed DLDPF. The

fundamental method is a Cascade Inception-based CIDCNN-TL that incorporates

Transfer Learning. For its implementation, the suggested framework makes use of

the Keras and TensorFlow deep learning libraries, as well as the Python data

science platform. AlexNet, GoogLeNet, VGGNet-16, and ResNet-20 are only few

of the deep learning module compared to the proposed framework DLDPF. The

empirical research of apple leaves makes use of a large dataset. Output of

experimentation showed that the DLDPF is more accurate than the state-of-the-art

deep learning module in the automated prediction of lead illnesses.

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 8
 Leaf Disease Detection of Multiple Plants using Deep Learning

2) Roy, Arunabha M.,et.al. "A Deep Learning Enabled Multi Disease Detection

Moduleas per Computer Vision" 413-428. As per cutting-edge computer vision

method, this research proposes a deep learning-enabled object identification

model for identifying many classes of plant diseases. While most current models

only allow for large-scale illness diagnosis, current module portrays for reliable

fine-grained multi-scale early disease detection. The proposed model has

undergone fine-tuning to enhance its performance in detecting apple plant diseases

in a multi-class setting, particularly in uncontrolled environments. The

optimization process has focused on achieving a balance between speed and

accuracy. Detection model demonstrated mAP of 91.2% & F1-score of 95.9% at a

detection rate of 56.9 FPS. In general, the findings indicate that the current

methodology surpasses the existing detection model significantly, exhibiting a

notable increase of 9.05% in accuracy & 7.6% enhancement in F1-score. The

suggested model provides a practical and economical means of diagnosing a

variety of illnesses affecting apple plants in realistic orchard settings.

3) Sardoğan, Melike, et.al. “Plant Leaf Disease Detection and Classification Based

on CNN with LVQ Algorithm.” UBMK (2018), pp. 382-385. Artificial systems

have made it simpler to identify diseases in modern times. It seems cumbersome

to keep vigil over plants around the clock, go to the agricultural office, or consult

an expert for spotting signs of illness. An easy method to tell whether a plant is

sick or healthy is necessary, thus scientists have set out to create such a system.

Diseased plants will also be reported so that a treatment strategy may be

implemented. Researchers in this study gathered data on vegetables from a field,

used many different Neural Network Algorithms, and developed a novel

ModCNN that achieved 97.69% accuracy. The authors have also organized signs

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 9
 Leaf Disease Detection of Multiple Plants using Deep Learning

of the bean leaf illnesses into four groups to aid in precise diagnosis.

4) N. Goel, et. al. "Predicting Module for Automatic Leaf Disease Analysis," 2018

IEEE IACC, 2018, pp. 360. When weather patterns shift, it may have a

devastating effect on crops, reducing agricultural output. As the situation

deteriorates, crops may become more susceptible to illnesses from pathogens.

Identifying plant diseases in real time is one strategy for minimizing crop failure.

Our proposed algorithm uses a trained dataset of pomegranate leaf pictures to

automatically identify plant illness. Using the data from the test set, we can

determine whether a user has uploaded a healthy picture or one that includes

illness. If not, the leaf is regarded healthy, and automated recommendations for

preventing plant disease are made. picture analysis done on the picture validated

by biologists and scientists identifies the disease-causing rodent. This model

delivers experimentally tuned accuracy for picture segmentation utilizing

numerous cluster sizes. We may use our model to estimate and forecast the spread

of illness while taking preventative measures.

5) Sandeep Kumar, et al. “Leaf Disease Detection and Using Machine Learning for

Classification: Proceedings of the International Conference, CSTCEE 2020. Plant

disease detection is a crucial activity in the agricultural industry. The economy

relies heavily on this factor. Because of prevalence of plant illnesses, finding new

plant infections is an important task in the agricultural sector. Constantly

observing your plants is essential for spotting leaf diseases. Keeping a close eye

on plants around the clock requires a lot of time and energy from humans. A

scheduled method of plant monitoring is necessary for its simplification. The

diagnosis of plant diseases using computer programs facilitates the detection of

damaged leaves with less time and effort expended by humans. The suggested

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 10
 Leaf Disease Detection of Multiple Plants using Deep Learning

algorithm can more precisely recognize plant diseases and categorize them than

current methods.

6) P. Jiang, et.al, "Infectious Apple Leaf Diseases Identified utilising an Enhanced

Deep Learning Approach," in 59069-59080, 2019. There are 5kinds of apple leaf

infections that have a significant negative impact on apple yield. To ensure the

continued success of the apple business, however, researchers need a quick and

reliable method of detecting illness in apples. For identifying illnesses in apple

leaves in real time, the authors of this research suggest a deep learning technique

based on enhanced convolutional neural networks. Into this study, we first use

data augmentation and image annotation tools to build ALDD, which includes

both controlled laboratory photos and complicated images captured in real-world

field settings. In light of this, we present a novel deep CNN-based model for

detecting apple leaf infections. Data demonstrates that the INAR-SSD model

attains 78.80 percent detection rate.

7) K. P. Panigrahi ,e al. “Maize Leaf Disease Detection and Classification Using

Machine Learning Algorithm”, 2020, Springer, pp.659 The main reason why

agricultural output is poor is due to plant diseases. The most common problems

that farmers face are those of disease detection and management. Therefore, it

would be helpful for farmers to discover these illnesses early so that they can

minimise losses. We test and contrast the aforementioned classification methods

to find the most effective one for predicting plant diseases. When compared to

other classification methods, the RF algorithm achieves the best accuracy

(79.23%). As a preventative strategy, farmers will employ all of the

aforementioned trained models for the early identification & classifying of new

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 11
 Leaf Disease Detection of Multiple Plants using Deep Learning

image illnesses.

8) S. Gayathri, et al. “Image Analysing& Detecting of Tea Leaf Disease using deep

Learning, 2020,” (ICESC 2020), pp.398 Research into image recognising

technology, that allows the area of automated picture classification and discovery

of plant diseases, has increased in response toprevalent development of deep

learning. India is both the greatest user and producer of tea in the world. Damage

to plant's health from diseases might stunt its development. Several diseases

impact tea leaves and slow crop development. The goal of our research is to create

CNN named LeNet, and use it to identify illnesses in tea plants based on images

of their leaves. This makes LeNet the ideal CNN model for improving diagnostic

assessment of tea leaves and other plant leaves in future applications.

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 12
 Leaf Disease Detection of Multiple Plants using Deep Learning

2.1 EXISTING & PROPOSED SYSTEM

2.1.1 Existing system & Disadvantages

Current system will list pesticide suppliers geographically, while it

additionally includes details such as name of illness affecting afflicted plant's
leaves, exact time and date of disease's appearance, & prediction for the region.
Apple, potatoes, maize, tomatoes, grapes, and soya beans are only ten of crops
included in collection.

 Disadvantages

 Only detects disease

2.1.2 Proposed system & Advantages

Image processing is at the heart of the system we've presented.

Furthermore, system lists plant's name, soil kind it prefers, location where it
thrives, ideal temperature & precipitation levels, & reasons & recommended
treatments for any diseases it may have.

Advantages

 Save energy
 Save time
 More accuracy
 Detects phases of disease

2.2 TOOLS & TECHNOLOGIES UTILISED

Deep Learning for Leaf Disease Detecting in Multiple Plants research made
use of several cutting-edge resources and methods. Among them are:

 In order to teach the system to identify patterns and traits that are diagnostic of
certain plant diseases, we used a set of sophisticated algorithms called deep
learning.
 In order to accurately diagnose plant diseases, researchers employed CNNs to

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 13
 Leaf Disease Detection of Multiple Plants using Deep Learning

extract relevant information from photos of plant leaves.

 To ensure that the deep learning models had the best possible input, we used
sophisticated image processing methods to improve quality and clarity of leaf
pictures.
 To create deep learning models and build the whole system, Python was the major
programming language.
 TensorFlow and PyTorch, two of the most well-known deep learning frameworks,
were used to facilitate the creation and rollout of the DL models.
 To efficiently train deep learning models, researchers used a wide variety of visual
datasets depicting both healthy and damaged plant leaves.
 Using high-performance GPUs sped up training and inference phases, cutting
down on computational time.
 To meet the computational needs of training massively parallel deep learning
models and to make them more amenable to scaling, we turned to cloud
computing platforms.
 Easily uploading leaf photos and receiving instantaneous disease diagnosis
findings is now possible for farmers and gardeners thanks to a user-friendly online
or smartphone interface.

2.3 HARDWARE & SOFTWARE NECESSITIES

HARDWARE PREREQUISITE

 Processor : Pentium core I5 and Higher

 RAM : 4GB or more
 Hard Disk : 500 GB or more
 Monitor : 15 inch Color monitor
 Keyboard : 102/104 Keys
 Mouse : Optical mouse

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 14
 Leaf Disease Detection of Multiple Plants using Deep Learning

SOFTWARE PREREQUISITE

 Operating System : Windows 10

 Front End : Phyton

 Back End : SQLite3

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 15
 Leaf Disease Detection of Multiple Plants using Deep Learning

CHAPTER-3

SYSTEM REQUIREMENT SPECIFICATION

3.1 USERS

The photos in the Kaggle Plant Village dataset represent a wide variety of
plant diseases. There are around 32,000 photos in the collection, depicting healthy and
sick plant leaves. Sugarcane, cotton, potato, & guava are only few commercial/cash
crops, & fruit plants considered in this method. The suggested model was used to
build and construct an automated system used to identify plant illnesses, which is then
used to determine whether or not the plant is afflicted. These are the procedures that
make up the method: 1. Collecting a plant picture dataset encompassing 38 classes of
plant illnesses shown in 14 distinct types of plant photos. The second step is a
convolutional-layer-based preprocessing of the picture. Third, a system for
categorizing plant illnesses based on whether or not a picture of a plant's leaf seems to
be sick

3.2 FUNCTIONAL NECESSITIES

1. The use of CNN algorithm for crop disease detection.
2. There is a well-trained & tested dataset

3.3 NON-FUNCTIONAL REQUIREMENTS

1. Performance Prerequisite
 Function & module efficiency has to be satisfactory.
 Software's high performance will allow its users to get more done in
less time.
 Reaction times must be very short.
 Provided virtual environment must have excellent efficiency.

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 16
 Leaf Disease Detection of Multiple Plants using Deep Learning

CHAPTER-4

SYSTEM DESIGN

4.1 SYSTEM PRESPECTIVE

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 17
 Leaf Disease Detection of Multiple Plants using Deep Learning

MODULEDESCRIPTION

1. DATA COLLECTION: The photos in the Kaggle Plant Village dataset

represent a wide variety of plant diseases. There are around 32,000 photos in
the collection, depicting healthy and sick plant leaves. Sugarcane, cotton,
potato, & guava are only some of the commercial/cash crops, & fruit plants
taken considered for method. The suggested model was used to build and
construct an automated system used to identify plant illnesses, which is then
used to determine whether or not the plant is afflicted. These are the
procedures that make up the method: 1. Collecting a plant picture dataset
encompassing 38 classes of plant illnesses shown in 14 distinct types of plant
photos. The second step is a convolutional-layer-based preprocessing of the
picture. Third, a system for categorizing plant illnesses based on whether or
not a picture of a plant's leaf seems to be sick

2. IMAGE PRE-PROCESSING:Data enrichment & annotating are utilisedfor

improving the pictures given into data.

3. IMAGE ANALYSIS: The primary objective of our technology is to

recognize the kind of sickness present in a picture. We need to not only detect
the item, but also determine what category it falls into. To better identify
illnesses inside a picture, we extend the idea of an object identification
framework by adding other feature extractors.

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 18
 Leaf Disease Detection of Multiple Plants using Deep Learning

4.2 CONTEXT DIAGRAM

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 19
 Leaf Disease Detection of Multiple Plants using Deep Learning

CHAPTER-5

DETAILED DESIGN

5.1 USE CASE DAIGRAMS

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 20
 Leaf Disease Detection of Multiple Plants using Deep Learning

CHAPTER-6

SYSTEM IMPLEMENTATION

Key phases in the system development of project LDDMP utilizing Deep

Learning are as follows:

Images of healthy and damaged plant leaves have been gathered into variety of
sources and compiled into many datasets for use in training deep learning models.

To improve data quality and training sample variety, we preprocessed the

obtained leaf pictures by methods including shrinking, normalizing, and augmenting.

• Model Development: Common frameworks like TensorFlow and PyTorch were

used in the creation of deep learning models, most notably CNNs. To achieve their
aim of disease classification accuracy, these models were trained on preprocessed leaf
image datasets.

Using the prepared datasets, the resulting models were trained on high-
powered GPUs for testing and validation. Training required feeding the models leaf
photos and repeatedly tuning the model's parameters to reduce prediction errors. To
gauge the efficacy of the model and guarantee its applicability, validation sets is
utilised for test.

• Model Evaluation and Fine-Tuning: We used independent test datasets to assess

the trained models' accuracy, precision, recall, indicators. The models were fine-
tuned so that they could identify diseases more accurately.

To facilitate the uploading of photographs of plant leaves, a user-friendly

online or mobile interface was designed and deployed. Disease detection in submitted
photos was accomplished by running them through trained algorithms. The
technology returned reliable information quickly, which aided in making important
choices about illness management in real time.

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 21
 Leaf Disease Detection of Multiple Plants using Deep Learning

The system was constantly checked for errors and updated with fresh data to
make it more precise and secure. The algorithms' illness identification skills were
improved over time thanks to user input and real-world data.

The goal of project Leaf Disease Detection of Multiple Plants using Deep
Learning was to develop and deploy system which mightproficientlyclassify& control
diseases across numerous plant species having high degree of reliability.

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 22
 Leaf Disease Detection of Multiple Plants using Deep Learning

6.1 SCREEN SHOTS

Fig 6.1 Read Image

Through this we are inputting the leaf image

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 23
 Leaf Disease Detection of Multiple Plants using Deep Learning

Fig 6.2: Preprocessing

Converts color image to grayscale image

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 24
 Leaf Disease Detection of Multiple Plants using Deep Learning

Figure 6. 3: Converts to different channel

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 25
 Leaf Disease Detection of Multiple Plants using Deep Learning

Figure 6. 4 : Displays the leaf disease % and leaf is diseased or not

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 26
 Leaf Disease Detection of Multiple Plants using Deep Learning

Figure 6. 5: Accuracy Graph

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 27
 Leaf Disease Detection of Multiple Plants using Deep Learning

CHAPTER-7

SOFTWARE TESTING

7.1 TESTING STRATEGIES

1. Dataset Testing: Evaluate the quality and diversity of data utilised to train &

test deep learning model.

2. Model Performance Testing: Measure the accuracy and performance of deep

learning module in detecting leaf diseases using a separate validation or test

dataset.

3. Robustness Testing: Assess the model's capability for handling variations in

lighting, camera angles, and image resolutions to ensure it can generalize to

different conditions.

4. Real-World Testing: Deploy the system in a live environment and gather

feedback from users and domain experts to assess its usability and

effectiveness.

7.2 LEVEL OF TESTING

7.2.1 Unit Testing

1. Unit Testing: Test individual components and functions of the leaf disease

detection system to ensure they work correctly in isolation.

2. Model Input Testing: Verify the preprocessing module's ability to properly

resize, normalize, and enhance leaf images as inputs for deep learning.

3. Model Prediction Testing: Validate precision of disease predictions made by

deep learning module using a representative set of labeled leaf images.

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 28
 Leaf Disease Detection of Multiple Plants using Deep Learning

4. Error Handling Testing: Test the system's capability for handling exceptions or

errors during preprocessing, model training, or prediction phases and ensure

proper error handling and messaging.

7.2.2 Integration Testing

1. Integration Testing: Verify the interactions and compatibility between

different modules and components of the leaf illness detecting system.

2. Preprocessing and Model Integration Testing: Validate the integration of

the preprocessing module with deep learning module and ensure proper

data flow and compatibility between the two.

3. Data Exchange Testing: Verify the seamless exchange of preprocessed leaf

images and disease predictions between the preprocessing module, deep

learning model, and inference module.

4. End-to-End Testing: Perform comprehensive tests to ensure the entire

system functions correctly, from leaf image input to disease prediction

output, and validate the accuracy and performance of the integrated

components.

7.2.3 System Testing

1. System Testing: Evaluate the leaf disease detection system as a whole to

ensure it meets the specified requirements and functions correctly.

2. Input Validation Testing: Verify the system's ability to handle various

types of input, including different image formats, sizes, and qualities.

3. Disease Classification Testing: Assess the accuracy and reliability of the

system in correctly identifying and classifying leaf diseases across

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 29
 Leaf Disease Detection of Multiple Plants using Deep Learning

multiple plants.

4. Performance and Scalability Testing: Evaluate the system's performance

metrics, such as inference time and resource utilization, and test its

scalability to handle a large volume of leaf images efficiently.

7.2.4 Validation Testing

1. Validation Testing: Verify that the leaf disease detection system meets the

specified requirements and performs as intended.

2. Accuracy Validation: Evaluate the system's accuracy in correctly identifying and

classifying leaf diseases by comparing its predictions against manually verified

disease labels.

3. Robustness Validation: Assess the system's ability to handle variations in leaf

images, lighting conditions, and disease severity to ensure reliable and consistent

performance.

4. Generalization Validation: Test the system's ability to detect leaf diseases across

multiple plant species and geographical locations to ensure its effectiveness in

real-world scenarios.

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 30
 Leaf Disease Detection of Multiple Plants using Deep Learning

CHAPTER-8

CONCLUSION

In conclusion, Leaf Illness Detecting of Multiple Plants utilizing Deep

Learning study significantly advanced plant disease management. The study used

deep learning algorithms to provide a successful & cost-effective approach for

diagnosing plant illnesses. This cutting-edge method provides growers and gardeners

with a solid resource for early disease detection, which is crucial for

stopping development of blight & limiting subsequent crop losses. Project's intuitive

design & ongoing updates help ensure that farmers everywhere have access to safe,

nutritious food while also promoting environmentally responsible farming methods.

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 31
 Leaf Disease Detection of Multiple Plants using Deep Learning

CHAPTER-9

FUTURE ENHANCEMENT

Multi-Plant Leaf Disease Detecting utilizing Deep Learning project has need

for improvement in following areas:

i) Improve the system's capacity to reliably identify a broader variety of plant

health concerns by updating and increasing the database of plant illnesses,

including uncommon and emerging diseases.

ii) By integrating sensors and IoT devices in the field, we can monitor plant

health data in real time, allowing for early disease identification and

proactive disease treatment.

iii) Incorporating multi-spectral imaging methods may offer more data on

biochemical & physiological changes in plants, leading to better disease

identification and more accurate diagnosis.

iv) Reduce need for larger training data and boost the system's performance

on new or lesser-known plant types by using transfer learning approaches,

in which pre-trained models are fine-tuned on particular plant species or

illnesses.

v) Adding a smartphone app that lets farmers and gardeners take pictures of

leaves and analyze them later without an internet connection would greatly

improve accessibility.

vi) Algorithms to track and forecast illness development over time might

improve disease management practices and make more efficient use of

available resources.

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 32
 Leaf Disease Detection of Multiple Plants using Deep Learning

vii) By providing a forum for researchers, farmers, and other specialists to pool

their expertise, we may better understand plant diseases as a group and

work to develop better methods of disease detection and control.

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 33
 Leaf Disease Detection of Multiple Plants using Deep Learning

CODING:

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 34
 Leaf Disease Detection of Multiple Plants using Deep Learning

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 35
 Leaf Disease Detection of Multiple Plants using Deep Learning

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 36
 Leaf Disease Detection of Multiple Plants using Deep Learning

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 37
 Leaf Disease Detection of Multiple Plants using Deep Learning

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 38
 Leaf Disease Detection of Multiple Plants using Deep Learning

REFERENCES

[1] T.Vijaykanth Reddy, et.al. “Deep Leaf Disease Prediction (DLDPF) with

Transfer Learning for Automatic Leaf Disease Detection”, 2021 Fifth

International Conference on Computing Methodologies and Communication

(ICCMC 2021),

[2] Roy, Arunabha M.,et.al. "A Deep Learning Enabled Multi Disease Detection

Model Based on Computer Vision" 413-428.

[3] Sardoğan, Melike, et.al. “Plant Leaf Disease Detection and Classification

Based on CNN with LVQ Algorithm.” International Conference on Computer

Science and Engineering (UBMK) (2018), pp. 382-385.

[4] N. Goel, et. al. "Prediction Model for Automated Leaf Disease Detection &

Analysis," 2018 IEEE 8th International Advance Computing Conference

(IACC), 2018, pp. 360 10.1109/IADCC.2018.8692116

[5] Sandeep Kumar, et al. “Leaf Disease Detection and Classification based on

Machine learning”, International Conference on Smart Te in Computing,

Electrical and Electronics (ICSTCEE 2020)

[6] P. Jiang, Y. Chen, B. Liu, D. He and C. Liang, "Real Apple Leaf Diseases

Using Deep Learning Approach Based on Improved Convolutional Neural

Networks," in 59069-59080, 2019.

[7] K. P. Panigrahi ,et al. “Maize Leaf Disease Detection and Classification Using

Machine Learning Algorithm”, 2020, Springer, pp.659

[8] S. Gayathri, et al. “Image Analysis and Detection of Tea Leaf Disease using

deep Learning, 2020,” International Conference on Electronics and

Sustainable Communication Systems (ICESC 2020), pp.398

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 39
 Leaf Disease Detection of Multiple Plants using Deep Learning

[9] Sammy V. Militante et al. “Plant Leaf Detection and Disease Recognition

using Deep Learning, 2019,” IEEE Eurasia Conference on IOT,

Communication and Engineering, 2019, pp.579

[10] Prajwala TM, el al. “Tomato Leaf Disease Detection using Convolutional

Neural Networks,” Eleventh International Conference on Contemporary

Computing (IC3), 2018.

[11] S. Pawar, P. Bagal, P. Shukla and A. Dawkh Cancer using Machine Learning

Classifier," Innovation in Technology (ASIANCON) 10.1109/ASIANCO

N51346.2021.9544767.

[12] Sk Mahmudul Hassan et al., “Identification of Plant Leaf Diseases Using CNN

and Transfer Learning Approach, 2021,” Electronics, 2021, pp.1-19

M.Tech. in Artificial Intelligence & Data Science, FOET(Co-Edu.), Sharnbasva University Kalburgi 40