Chapter -1

INTRODUCTION
1.1 INTRODUCTION

Agriculture is the backbone of the Indian economy. Since the beginning of

agriculture, great progress has been made to increase productivity and assist farmers
in dealing with problems such as plant disease. Earlier time leaf diseases is a massive
obstacle for farmers. The current method for plant disease diagnosis is simply human
eye monitoring, and ends in inaccurate detection and identification. This impact of
several plant diseases has been increasing in recent years. These symptoms have no
visible symptoms that are difficult to identify. Plant diseases include viruses, bacteria,
fungi, and some others. Therefore, the image processing and machine learning are
used in neural networks, fuzzy classifiers, and others. Symptoms are often diagnosed
individually, although human identification takes much longer time than automatic
identification. Automatic sickness identification is made much easier by the use of
image processing and machine learning.

Image processing is essential for reducing human work and achieving the most
effective outcomes. Convolutional neural networks (CNNs) having made significant
improvements in the identification of plant diseases in recent years. Rather than
determining the characteristics individually, CNNs allow end-to-end workflows to
continuously obtain more powerful features, and controlling the various efficiency of
plant leaf identification. The YOLOv5 algorithm has been developed to solve
problems and prevent the crop infections in large-scale production depending on each
of these classifications.

The YOLOv5 method was established during machine learning techniques, then
reduces the environmental impacts caused by pesticides consumption. Previously,
YOLOv3 and YOLOv4 are implemented, but they only obtained 75percent of total
production performance and accuracy, now YOLOv5 outperforms others. Within 50
to 74 seconds, the system reaches 99.3percent accuracy. YOLOv5 has become a CNN
it recognizes objects with a high accuracy in real time. The technique is processing
the entire image with a single neural network, then classify into parts to determine the

1
connection areas and correlations for every component. This method is applied to
learn algorithms for deep learning.

1.2 OVERVIEW

To identify pests, artificial intelligence and image recognition technologies are

integrated with sensor devices and the Internet of Things. Based on the intelligent pest
identification and environmental IoT data, real-time agricultural weather and pest
identification systems on mobile applications are evaluated. It is applied to smart
agriculture to integrate established performance of AIoT technologies with deep
learning. The crops are autolysis using image processing algorithms. It contains the
ability to detect several items. Crop pests grow in extremely dry and hot conditions.
The YOLOv5 model was used for deep learning and also the detection technique
alerts farmers to the presence of different pests before they multiply in large numbers.
It increases total agricultural economic value by providing proper pest management
strategies that reduce crop losses and costs.

1.3 EXISTING SYSTEM

1. To keep an eye on the plants and maintain the soil ideal moisture content.
2. It is applicable in the ONE-STAGE OBJECT DETECTOR METHOD.
3. The YOLOV5 ALGORITHM had been utilized to identify a single object.

1.4 REAL-TIME LEAF DISEASE DETECTION IN THE PLANT

Yolov5 has the potential to aid in the detection of leaf diseases by teaching the
model to recognize and categorize various types of unhealthy leaves in images. A
comprehensive dataset of annotated images can be used to train the model to correctly
identify different forms of leaf diseases, including their intensity levels. This
knowledge can then be used to optimize crop management procedures, reduce losses
in yield, and increase efficiency in the agricultural industry. The accuracy and speed
of Yolov5 render it a valuable tool in the field of leaf disease detection.

2
1.4.1 System Design

Fig.1.1 Block Diagram of Automatic Spraying System for Leaf Illness

At the start of the initial process, users can capture an image of an infected
plant leaf and store everything to your Raspberry Pi. Fig 1.1 explains that the
raspberry pi can transfer the information into image processing, the image processing
is needing a large amount of storage space and processing time throughout pre -
processing stage. The information contained in each pixel of the colour picture. Then
need to choose RGB picture convert to HSV picture scanning is capable of doing the
human eye receives and recognizes colour by detecting rays from the surroundings.
Colour is identified by wavelength healthy tissues in the eye, there are three types of
cells, one for red, one for green, and one for blue. Then HSV is the cylindrical
coordinate representation of RGB direct correlations.

HSV is also referred as HSB at times (hue saturation brightness). Now it

proceeds for image boundary fixing and k-mean algorithm is grouping method that
are separates an image into k numbers of groups or various types depending on input
data based on k centroid. The algorithm is simply straightforward. A collection of n
data points is used to determine initialised nodes. Each node can be selected to the
centre pixel close to it. After that, the cluster centres are changed to become the
average of every nodes is supplied to that cluster. Then proceed green pixel masking
is applied to indicate a specific type of colour range in a picture. The masking of

3
green colour is important for the identification of leaf disease. Various colours are
identified by masking the green colour that indicates which elements of the leaves
area remain unhealthy. After that change to Leaves Counting of Unhealthy Input
images.

Fig.1.2 Flowchart of Steps to Detect the Disease on the Plant Leaf

The concept behind all this technique is move to horizontal inside the leaf
picture and identify the initial and final green pixel, then measure the unhealthy pixels
over this green pixel. If the number of unhealthy pixels exceeds 30% of the total
number of healthy pixels, the leaf has the disease is concluded with the help of the
flowchart mentioned in above Fig 1.2

4
1.5 AN IOT BASED SMART SOLUTION FOR LEAF DISEASE DETECTION

The internet of things is a system consisting of actuators, sensors, or both that

connect to the network either directly or indirectly. Internet of Things (IoT)
developments can be used in smart agriculture to increase agricultural productivity.
Agriculture, the pillar of the Indian economy, contributes to the overall economic
growth of a country. But, according to the usage of ancient agricultural equipment and
overall productivity is very poor in comparing to providing high quality service, and
people living in rural areas are already relocating to urban areas for much more
profitable industries, making it difficult to enable to concentrate on agricultural.
Farming innovation is nothing new, but the Internet of Things (IoT) is ready to take
smart farming to the next level. The internet of things is a system that connects
actuators, sensors, or both provides connectivity to the internet directly or indirectly.
It has several characteristics such as plant leaf disease, a server-based remote
monitoring system, the humidity and temperature sensing, moisture in the soil sensor
and so on. Instead of performing a manual check, it utilizes smart sensors to measure
moisture, temperature, and humidity.

Various sensors are placed at several agricultural fields to control each of the
sensors is a single controller known as the Raspberry PI has already been
implemented (RPI). Leaf disease can be identified using a camera interface to
communicate with an RPI. Current agricultural condition, such as leaf disease as well
as other environmental factors affecting crop such as humidity, temperature, and
moisture are communicated to farmers through Wireless internet Servers by RPI. The
IOT techniques in order to systems today the usage of agriculture technology.

1.5.1 WORKING METHODOLOGY

In any agricultural, various sensors such as soil moisture sensors, temperature

or humidity sensors, and cameras for detecting diseases on leaves have been
implemented. The Data is collected by sensors and transmitted towards the Raspberry
PI through wireless or wired connections. Server-side information is verified and
compared to desirable datasets including temperature, humidity and soil moisture. If
there is a difference in a specified threshold value, the signal is received to the farmer
using mobile or internet. The output of sensors is displayed on the webpage, allowing

5
him to obtain specific information of the crop and farming atmosphere from any
location. Crop diseases are detected using image processing. The camera is placed
towards the crop in order to capture a leaf. The collected image is transferred to the
cloud, and image-processing techniques are used to identify leaf diseases. The user
online store and smartphone application are communicating the condition of a leaves
to the farmers.

1.5.2 Hardware Components

Various hardware components can be used for leaf illness detection using the
YOLOv5 algorithm. These components include cameras, computers or embedded
systems like NodeMCU ESP8266 in Fig 1.4 or NVIDIA Jetson Nano, GPUs, and
microcontrollers such as Arduino.

Cameras are necessary for capturing high-quality images of the leaves, which
are then processed by the YOLOv5 algorithm. Computers or embedded systems can
run the algorithm efficiently to detect and classify leaf diseases in real-time. GPUs
can be used to accelerate the processing of large amounts of data, reducing the time
taken for training and inference of the YOLOv5 algorithm.

Fig.1.3 Architecture of Leaf Disease Detection System

In Fig 1.3 the Architecture of Leaf Disease Detection System is explained

clearly. Microcontrollers like Arduino can be used to control the hardware
components of an automatic spraying system based on the output of the YOLOv5
algorithm. For example, the microcontroller can trigger the spraying mechanism when

6
the YOLOv5 algorithm detects a leaf disease, allowing for the automatic spraying of
affected plants.

Fig.1.4 NodeMCU ESP8266

1.5.3 Software Used

1. Python: This programming language is a widely used tool for machine

learning and computer vision applications. Python's versatility allows for the
implementation of different machine learning algorithms and libraries,
including YOLOv5.
2. TensorFlow: This is an open-source software library created by Google and
is commonly utilized for building and training machine learning models,
including object detection models such as YOLOv5.
3. OpenCV: This open-source computer vision library is a powerful tool for
image processing and object detection tasks. It provides several functions
for image manipulation, such as filtering and thresholding, which can be
utilized to pre-process the images before being fed into the YOLOv5
algorithm.

7
1.5.4 Background Related Work

The research of leaf disease and categorization based on field crop images has
progressed greatly as the amount of publicly available public datasets has increased
and image processing and object recognition technologies have expanded. The first
stage of the detection procedure is disease identification that depends mainly on
feature information produced by the model to identify disease spots and determine the
level of infection. For example, a fully connected CNN network for weed detection,
could discover and separate weeds in complex settings. It used a dataset of 18,222
pictures to create a convolutional neural network model to diagnose maize leaf
infection with 96.7% accuracy.

The disease identification step is the second stage of object detection. The
disease classification phase is the second step of object detection, and it consists of
matching the supplied input feature vector with one of all the classes learned during
training to identify so that class the item belongs to provide a data-balanced, quicker
R-CNN-based identification strategy based on several shooting angles for winter
jujubes of different maturity in natural environments. By randomly adding the
background database to the leaf pictures, the PANet-50 residual network for the
automated diagnosis of wheat illnesses was used to increase the variety of the
datasets. To achieve rapid identification of tea buds, Agriculture 2022 combined the
rapid detection capabilities of YOLOv3 with a detection and classification approach
based on a two-level fusion network with a changeable universe, and also the
recognition accuracy achieved 95.71%.

1.6 REAL-TIME PEST DETECTION USING YOLOV5

Given the expanding population and nutritional demands, we should develop new
agricultural production methods and systems that include environmental
considerations as well as maintaining efficiency and sustainability. Inadequate pest
management measures might result in a 70% loss of production. The invasive pest
that can reduce production by consuming plant and tree leaves, shoots, fruit, and floral
components. Pesticide spraying is the most widely used pest management method due
to its efficiency of action and scalability. Similarly, the public knowledge of
environmental and health issues increases, less pesticide use will be required.

8
Spraying pesticides directly if they are needed is a critical method for reducing
pesticide usage. To execute spot spraying, the position of the pest needs to be
determined. Pests can be identified using computer vision algorithms.

Using the YOLOv5 object detection technology, they constructed an object

detection system that uses video to identify pests that are observed and may cause
damage to sunflower production. it used 2416 photos taken under various lighting and
backdrop conditions. The object recognition system was trained in two way there
are deep learning and learning from scratch, and the outcomes were compared. The
results indicate that the system is functioning and capable of technical and
behavioural.

1.6.1 YOLOV5 using Image-Processing

YOLOv5 is a highly effective tool for detecting leaf illnesses. This advanced
algorithm for object detection has been utilized for numerous purposes, including
identifying plant diseases present in leaves. YOLOv5 represents an updated version of
the YOLO algorithm that boasts exceptional enhancements in both accuracy and
speed. This algorithm can be trained to identify and categorize diverse leaf diseases,
such as bacterial spot, powdery mildew, and rust. In order to use YOLOv5 for leaf
illness detection, it is necessary to train the model on a collection of labeled images
that feature both healthy and diseased leaves. The labeled images are critical for
instructing the model to distinguish between various types of leaf diseases. Once the
model has been trained, it can be applied to new images of leaves to detect and
classify diseases. Ultimately, YOLOv5 is a potent algorithm with the potential to
revolutionize leaf illness detection, and it is a valuable area of research to explore.

1.7 OBJECTIVE

To create a system capable of identifying crop disease and pests that must be
supplied. Create a pesticide database for each pest and disease, in addition to provide
a proper treatment for the identified diseases.

1.7.1 Motive of The Work

9
 1. Farmer economic growth is totally dependent on the plant’s growth and
yield. Plants are particularly susceptible to disease, and leads to losses.
2. To prevent plant disease and avoid huge losses for farmers, it is necessary to
detect the disease at the very early stage to avoid additional damage of the leaf.
3. Because human detection is a painful process, you need an automated and
computerized version to make it simpler and much less time-consuming.

1.8 PROBLEM STATEMENT

Automatic detection of plant diseases is an important to study the problem

because it could be useful for monitoring large fields of crops and moreover
automatically detecting disease symptoms are quickly they appear on plant leaves. It
is also valuable information in the field of machine learning. It is a technique that uses
machine learning technology to collect images and evaluate how many of the images
contain diseases or pests. Visible plant disease are measure by human rates,
microscopic inspection, and many other. But without a human mechanism, how can
you detect the disease.

1.9 ORGANIZATION OF THE PROJECT

Chapter - I includes a presentation of the pest detected as well as a few information

in to current system. It also includes problem statement, motivation, and the task
function.

Chapter - II provides clear and concise information about the writing review.

Chapter - III configurations using the current model, that provides basic overview on
the operation and functioning of the Image processing based on YOLOV5.

Chapter - IV the outcomes of the work are described.

Chapter - V describes additional work for improving the proposed system and
analyses completed work.

10
 Chapter 2

LITERATURE SURVEY
2.1 OVERVIEW

The identification of leaf diseases is essential for reducing productivity and

quantity losses in agricultural products. Leaf research and treatment involves the
study of respect to various patterns on the leaves. Monitoring and detecting diseases
on leaves is necessary for long-term sustainable agriculture. It is quite tough and
Continuously monitor the leaf diseases. It takes a huge amount of effort, as well as
knowledge of leaf diseases require a lengthy processing time. As a result, image
processing is used in the identification of leaf diseases Image capture, image pre-
processing, image segmentation, and feature extraction are all phases in detection as
well as classification This study described the methods used to diagnose leaf diseases
using images of the leaves.

The use of soil texture data to identify plant leaf disease is already defined.
First, RGB is converted into HSV form using a colour transformation structure
because HSV is an excellent colour identifier. Green pixels are masked and recovered
using a per-calculated threshold level. The training procedure was carried out on each
YOLO model, taking use of the presented configuration. During training, the models
results on the test dataset are automatically calculated. Deep learning in agriculture
has various applications, including plant diagnostic techniques, disease identification,
and production calculation.

Traditional yield assessment in wine production often is relies on the hand

harvesting of growth of selected, that is a time-consuming procedure. The use of
automated bunch counting could reduce the time required to estimate production. In
the current work, four variants of You Only Look Once (YOLO) object identification
models have been trained and evaluated for real-time cluster detection and counting

11
Compared to certain other YOLO models such as YOLOv4 and YOLOX, the
YOLOv5 model is used to develop applications for mobile devices, because of its
small size and model parameters, computation speed and high accuracy.

2.2 LITERATURE REVIEW

Y. Liu, et al. [1] The agricultural production is greatly affected by various plant
diseases. Classifying the severity of crop diseases is the requirement for formulating
disease prevention and control strategies. However, the differences between different
severity of the same crop disease are very tiny. It increases the difficulty of correct
crop disease recognition. For example, at the early stage of the disease, the lesions on
the leaves are not obvious. The performance of the proposed model was evaluated
using the AI challenger 2018 plant disease recognition dataset, and the recognition
accuracy can reach 91.94%, which is 3.02% point higher than the original Squeeze
Next model. In order to verify the effectiveness of the proposed model, comparative
experiments were carried out using ReseNet50, Xception and mobilenetv2. The
experimental results showed that the accuracy of the proposed method was slightly
better than Xception, while the model size is only 2.83 MB, which is only 3.45% of
Xception. The proposed method balances the performance and efficiency very well.
Thus, it is suitable for deployment on mobile terminals and other embedded resource-
constrained devices, which help to promote the popularization of smart agriculture
application.

Yoshua Diller et al. [2] Fruit flies (Diptera: Tephritidae) are invasive and prevent
the spread species that impact global fruit and vegetable supply. The current work
covers a freshly designed McPhail-type electronic trap (e-trap) and its field
performance in surveying three important invasive FF species (Ceratitis capitata,
Bactrocera dorsalis and B. zonata) The FF classification code was initially built using
a machine-learning algorithm and FF photos collected from two of the species'
experimental colonies (C. capitata and B. zonata). Field experiments were
subsequently carried out to assess the e-electronic, trap's communication, and
appealing capabilities, as well as the model's accuracy in classifying FFs.The FF
family (Diptera: Tephritidae), which encompasses about 4000 species worldwide,

12
includes a group of economically important insect pests that infest fruits and plants.
The e-trap is engaged twice a day at this point, sending two photographs every day.
Image analysis algorithms desire to fit for this purpose process uploaded photos. The
outcomes are controlled and incorporated into risk models, as well as alerting
stakeholders.
Robert Brodschneider et al. [3] Beekeepers have several choices for
controlling the parasite mite Varroa destructor in honey bee colonies, but there is little
empirical evidence on the strategies they use in reality. The most common operations
were Varroa infestation monitoring, drone brood removal, different oxalic acid
treatments, and formic acid applications. The mite predominantly eats fat body tissue,
which results in lower weight, reserve protein levels, and adult lifespan in afflicted
larvae, among other things. Furthermore, the mite transmits honey bee viruses. The
study was carried out using the programme R version 4.0.4. (R Core Team 2021).
Effective control is achieved by the selection of appropriate procedures in relation to
colony and environmental circumstances, as well as the combination of approaches.
Oxalic acid crystals are produced.
Livia Zapponi et al [4] Behavioural manipulation is a pest control technique
that's also commonly used across the world to protect many crops from various insect
species. It focuses on the use of organic components (i.e., semiochemicals) that aim to
interrupt, block, or enhance behavioural patterns, therefore reducing or eliminating
crop losses. Brown marmorated stink bug (BMSB) is an invasive species that has
become a main agricultural problem inside the regions that has expanded. The
experiments investigated even if the rotational signal might increase the number of
BMSB entering the trap under controlled conditions (greenhouse), measuring males,
females, and larvae separately. Next, to evaluate the performance on the field of the
two - layer trap. Four different kinds of traps were tested: 1) VON trap enticed with
Trécé Inc. Dual Lure; 2) VOFF trap (transducer switched off) lured with Trécé Inc.

W. Chen et al. [5] an AI-based pest detection system that uses images to
identify levels of pests. Deep-learning-based object identification algorithms, such as
quick and efficient region-based convolution networks, single-shot multi box
detectors, and YOLOv4, are used to recognize and identify specific levels of pests in
images. YOLOv4 exhibited the greatest classification accuracy among the algorithms,
with 100% in insect vectors, 89% in Coccidae, and 97% in Lives either directly or

13
indirectly. A smartphone application based on the learned level of insect detection
model has been developed to help farmers in identifying pests and applying suitable
pesticides to decrease crop losses.

Lena Durocher et al.[6] Identify the leaf damage assessment levels for
autumn armyworm are most regularly used for which purposes, that may also give the
greatest opportunities for statistical analysis, but that it would be an appropriate
balance of information and workload. Common leaf damage gauges include the
nominal "yes-no damage scale," which just analyses damage incidence, and difficult-
to-understand ordinal scales that incorporate incidence and severity information, such
as the "Simple 1 to 5 entire plant damage scale." During detection approaches, many
software languages such as Python and C++ are employed, as well as servers and
image processing. The research has been completed, and a website has been created to
display farmers the present state of their crops. A certain source states that the
approach may enhance agricultural output and farmers' total revenue.
Channamallikarjuna Mattihalli et al. [7] In this method, images of leaves
are captured and compared to images of healthy leaves in a database that is per-stored
in the device. If the plants are determined to be infected after image analysis, this
device automatically opens the valves, allowing or blocking medicines supply to the
plant area using a sprinkler or drip irrigation method. In addition, soil moisture and
temperature sensors are used to prevent disease transmission caused by changes in
climatic conditions. If the moisture/temperature measured values above the specified
range, the Beagle Bone Black enables the plants to self-medicate. Farmers are notified
of plant and valve operations through GSM.Deep learning methods are increasingly
has been used in agricultural extension. even if they can automatically learn the deep
convolution information of images and also have faster frequency and accuracy levels
than standard algorithms. Deep learning has also been used to detect plant diseases in
visible light images.
G. Jocher et al. [8] YOLOv5 is a single-stage object detection system. Object
detection is considered a recurrent problem in one-stage object identification systems.
In a single step on the input image, it calculates the detection model and the
dimensions of the structuring element that will include the item. The three major
components are the backbone, neck, and head. The head layer is also known as
YOLO. In YOLOv5, the model backbone is a CSPNet structure. The CSPNet

14
technique separates the feature map in the base layer into two sections they are half
reach the transition layer through the convolution layer, and the other half is directly
integrated. This not only decreases model size but also allows faster prediction.
Y.Wang et al.[9] Compared to certain other YOLO models such as YOLOv4
and YOLOX, the YOLOv5 model is used to develop applications for mobile devices
because of its small size and model parameters, computation speed and high accuracy,
and lack of development of high hardware configuration. The YOLOv5 model is
quite small in size, such as a network parameter of 7.3M and a storage capacity of
14.2 MB, thereby making it suitable for mobile devices with limited hardware
configuration. The YOLOv5 calculation speed is sufficient. Especially compared to
previous YOLO models, the indications indicate the YOLOv5 model speed. To
develop the agricultural pests dataset for the machine learning models, approximately
3.5 pictures of 10 different pest types are collected from mobile data sources. After
that, the dataset was divided into the following proportions are 70% of the samples
used for training, 20% for model assessment, and the remaining samples have been
used for testing.
J. Redmon et al. [10] A YOLO (You Only Look Once) detection model was
described as a one-stage detection model. Apart from Faster R-CNN that requires
repeated training of the RPN (Region Proposal Network), YOLO only requires one
training course. YOLO is a technique used to predict, as compared to R-CNN, it
divides detection results into two steps are object type and object location. Because
the output layer is a fully linked layer, the YOLO training model can only identify
images with the same input quality as the training image. Additionally, even if every
detected cell contains several objects, only one of them may be detected at a time. The
combined training strategy is used in YOLOv2, and more items are detected faster
and more accurately.
Sanjay B et al. [11] It describes how to identify leaf diseases in plants by
using a characteristic parameter. To begin, convert the captured RGB picture to HSV
image and then use masking to mask the image green pixels. After masking,
segmentation is applied, and these segments are used for edge detection. Finally, the
surface parameter is compared to the optimal surface parameter of the leaf. on image
processing techniques used on a variety of crops including fruit, vegetable, industrial
and rice production. The proposed approach is for identifying fungal disease
symptoms on various types of crops. The database is kept up to date in order to store
15
fungal disease patterns on crops. The technology is used to remotely monitor the crop
and provide notifications to the owner in order to prevent additional loss.
Apeksha Thorat et al. [12] a method for using image processing to identify
diseases in plant leaves. The plant is first captured as an RGB picture. Color values
are transformed to the domain after an image is converted and a pattern has been
created during pre-processing. To mask green pixels, K-means is applied. Useful
image elements are produced through image classification. For edge, texture, colour,
and feature, the colour texture features technique is applied. Neural networks are used
to recognize and classify diseases. By converting RGB images to HSV images, HSV
videos are produced. Cross matrices are used to conduct color features object
identification. Skin image evaluation can be done using statistical or structured
methods.
Lena Durocher-Granger et al. [13] It concentrates mostly on FAW (Fall
Armyworm), a pest disease, and the pesticides used to control FAW in Zambia. Small
- scale farmers suffered major losses in maize as a result of FAW. Increased pesticide
use raises worries about health and environmental risks. Evaluation of local natural
predators attacking FAW based on location, data collection, quantity of plants, and
amount of damage produced. They have used a variety of methods, including field
location, egg masses, larva collecting, and parasitism identification. Statistical
analysis was performed on parasitism level, maize development phase, and FAW
density. There are eleven parasitic organisms that attack the eggs and larvae to
prevent the spreading of FAW through preservation and biological activities.
C.Zhu et al. [14] a method of image processing and a smartphone application
for identifying and detecting insects A sliding cluster segmentation method is used to
minimize the non - homogeneous illumination of pest images captured using mobile
phones, then connected to domain-based statistical characteristics are used to identify
and number the pests in agricultural crops. Finally, testing using a Smartphone
application indicates that the methodology performs better than the previous method
in collecting sample pest images through mobile phones with 95% accuracy.
Vijai Singh et al. [15] The method of identifying a disease involves multiple
processes, but the four most important parts are as followed by, a colour
transformation structure is applied to the input RGB picture. Then next, the green
pixels are masked and quickly removed to use a certain threshold value, then followed
by a segmentation process and finally, surface parameters are calculated to provide
16
informative segments. Finally, a classifier is used to identify the disease using the
parameters that have been collected. With a 94% accuracy rate in the successful
detection and identification of the tested diseases, the methodology proves its
efficiency. The stability of the specific flow is identified using experimental data
collected from a database that contains outputs from around 400 plant leaves.
Amreen Abbas et al. [16] On the Plant leaf dataset, the PANet model has an
accuracy of 94.3%, resulting in an accuracy of 95.65%. They identified three disease
are early huge drain, late plague, and leaf fungus using two deep learning processes
on a dataset of plant leaves. because the first model uses feed-forward CNN in ways
similar to max - pooling layers, even though the second model uses CNN in parallel to
long-short-term memory and observed simply learning. The resultant CNN
architecture, that was attention-based, used to have the greatest accuracy of 98%. It
defines the mobile application method used by farmers to receive agricultural real
time updates. The model makes use of sensors such as soil moisture, temperature and
humidity.
Arti N. Rathod et al. [17] In neural network it difficult to understand
structure of algorithm and to determine optimal parameters when training data is not
linearly separable explains many data classification techniques for plant leaf disease
identification. To providing learning area, the K-nearest neighbour technique appears
to be the most suitable and easiest way of the algorithms for class prediction. If the
training data is not present in the data, determining optimal parameters in SVM
becomes difficult, that is one of its disadvantages.
Smita Naikwadi et al. [18] Better result of detection can be obtained with the
large database and advance feature of colour extraction. To identify plant diseases that
appear on the leaf, image matching is performed using an edge detection technique
and a colour feature. The layers separation technique is used in the training process,
that involves training these samples to separate the layers of an RGB picture into red,
green, and blue layers, as well as the edge detection technique, and detects the edges
of the layered images. The colour founder image enhancement method is developed
by using spatial gray-level dependence matrices.
S. Arivazhagan et al. [19] The training samples can be increased and shape
feature and color feature along with the optimal features can be given as input
condition of disease identification uses an artificial neural network (ANN) and several
image processing techniques to identify plant diseases quickly and accurately.
17
Because the best accuracy is based on an ANN classifier used for classification and a
Gabor filter for extracting features, it produces better results, with up to 91%
recognition rate. An ANN-based classifier identifies several plant diseases by
combining textures, colour, and information to identify certain diseases.
Abah Bashir et al. [20] Bayes classifier, K-means clustering and principal
component classifier can be used to classify various plant diseases. To provide disease
detection in Malus domestic a using an efficient method such as K-mean clustering,
texture and colour analysis. It uses texture and colour traits that are common in both
normal and damaged areas to identify and differentiate distinct crops. K-means
clustering, Bayes classifier, and principle component classifier all could be used for
classification in the future.
2.3 SUMMARY
The training procedure was carried out on each YOLO model, taking use of the
presented configuration. During training, the models' results on the test dataset are
automatically calculated. Deep learning in agriculture has various applications,
including plant diagnostic techniques, disease identification, and production
calculation. Traditional yield assessment in wine production often is relies on the
hand harvesting of growth of selected, that is a time-consuming procedure. The use of
automated bunch counting could reduce the time required to estimate production. In
the current work, four variants of You Only Look Once (YOLO) object identification
models have been trained and evaluated for real-time cluster detection and counting.
A heterogeneous dataset was used to train YOLO models. YOLO models were trained
using a diverse dataset consists of images collected from open databases and obtained.
Based on the testing results, YOLOv5 and YOLOv4 models outperformed YOLOv3
models in terms of F-1 score and mAP50. The validation based on the actual number
of clusters indicated that YOLO models could identify (True Positive) between 58%
and 83% of the actual number of clusters of plants. A False Positive-False Negative
adjustment, that reduced the error for YOLOv3 models, improved the final estimation,
that also included a correction for blocked clusters and an F-1 score. Using these
factors, YOLOv5x performed the best in rates of successfully identified clusters, and
YOLOv4 performed well but at a slower detection speed Including an average
inaccuracy of 13.3%, YOLOv5x was able to predict the number of clusters per plant.
Although YOLOv5x displayed real-time detection, a faster detection speed may be
necessary for on-field use. In this testing, YOLOv4-tiny obtained the best
18
combination of accuracy and speed. The leaf obstruction and False Positive-False
Negative correction should be analysed further to improve the object identification
algorithm's performance in agriculture.

Chapter-3

DETECTION OF CROP INFECTION BASED ON

YOLOv5 ALGORITHM

3.1 OVERVIEW

Recent Century is full of technology, none field can’t be remained without use of
the technology, Agriculture is one of those fields. More than 42% of the total
population in the world has chosen agriculture as their primary occupation. The
farmers have to do many works such of them are also not good for their health. Some
problems about also their crops as some insects can damage their crops. To decease or
remove these damages different types of pesticides available. So, the project aims to
build a system that automatically irrigates only that part of the Farm which needs
water based on the acquired information from the Soil regarding the control. The
Farm is divided into microclimate regions equipped with smart motes with special
sensors and integrated wirelessly (Lora-IOT) into automated irrigation controllers
with wireless networking capability. Also, it can be Equipped with Camera for
Disease Detection. Sensor motes senses the local climatic conditions such as
temperature and soil moisture level for that part of the irrigation Farm.

GSM have to done work also an unequal surface, so it is hard to do work in

the Terrace fields due to signal Coverage. Sensing distance, these GSM are works in

19
particular distance sets by user. Plant Disease Detection Not Available in Existing
system. The irrigation controller node receives periodical updates from the sensor
nodes and based on the dryness of the region it decides when to open the valve of the
motor to irrigate the field associated with it can be controlled by Soil also. The
controller node has a user interface in the form of a navigation keys to set the mode of
operation and to enter the password and an LCD display to view the recorded sensor
data. It is optionally connected with an IOT module that enables the farmer to send
control commands to the irrigation controller through SMS. Monitoring the sensor
data is not feasible as it may have large volumes of data. The wireless sensor network
is based on IEEE 802.15.4 networking protocol that operates in 2.4 GHz ISM band.
Each node has a RF transceiver and a low power Raspberry Pi.

20
Fig. 3.1 Image Explaining the Process of Proposed System

21
 The proposed system uses the 3d segmentation scheme used to detect the Rice
plant disease in remote sensing images in Fig.3.1. Here we are applying the RCNN
classification for the disease level prediction. The images are obtained by the remote
sensing database. The results showing the relatively we are achieving the better
accuracy when compare to other systems.

3.2 PROCESS FOR DETECTING THE INFECTED CROP

Fig.3.2 Circuit Diagram of Proposed System

The circuitry in Fig 3.2 is required for leaf illness detection using the
NodeMCU board with YOLOv5 algorithm will depend on the specific components
used and the system's design. However, here are some general guidelines to follow:

1. Connect the NodeMCU board to a power source, such as a USB cable or

battery. Also, connect it to a computer or a WiFi network for programming and data
transmission.

2. Connect the camera module to the NodeMCU board using the appropriate
pins. Ensure to choose a compatible camera module that can capture high-quality
images.

3. Optional: Add a sensor module, such as a temperature or humidity sensor, to

monitor the plant's environment and detect any abnormalities.

22
 4. If including an automatic spraying system, connect the system to the
NodeMCU board using appropriate pins or a separate controller.

5. Wire all the components together using appropriate jumper wires or

breadboard, following the manufacturer's instructions for each component.

6. Program the NodeMCU board to capture images, pre-process them if needed,

and send them to a computer or cloud server for disease detection using the YOLOv5
algorithm. You can also program the board to trigger the automatic spraying system
when a leaf disease is detected.

3.3 COMPONENTS USED FOR DETECTING THE INFECTED CROP

1. Camera or mobile device: To capture images of the leaves for analysis.

2. YOLOv5 algorithm: A deep learning algorithm for object detection, which is

trained to detect leaf diseases and pests.

3. Computer or server: To run the YOLOv5 algorithm and analyse the images
captured by the camera or mobile device.

4. Automatic spraying system: To treat the crops after detecting the presence of
leaf diseases and pests. The spraying system can be equipped with sensors and
actuators to detect the presence of pests and apply the appropriate treatment.

5. Pre-processing software: To enhance the quality of the captured images by

removing noise and distortion.

6. Feature extraction software: To extract features from the pre-processed images

using algorithms such as colour-based segmentation, texture analysis, and edge
detection.

7. Disease classification software: To classify the leaf into healthy or diseased

categories based on the extracted features. The software can use machine learning
algorithms such as convolutional neural networks (CNNs) or support vector
machines (SVMs) for classification.

8. Decision-making software: To make decisions based on the disease

classification and recommend treatment options.

23
 9. Reporting software: To generate reports that include information about the
disease classification and recommended treatment methods.

3.3.1 NodeMCU ESP8266

The NodeMCU showed in Fig 3.3 is an inexpensive and compact

microcontroller board that can connect to WiFi and be programmed with the Arduino
IDE. Although commonly used in IoT projects, it can also be utilized with the
YOLOv5 algorithm for leaf illness detection.

By incorporating a camera module with the NodeMCU, leaf images can be

captured and sent to a computer or cloud server for analysis via the YOLOv5
algorithm. The board's WIFI capability allows for remote crop monitoring and
convenient data transmission.

To use the NodeMCU with the YOLOv5 algorithm, it must be programmed to

capture and pre-process images if necessary, and transmit them for disease detection.
Upon detecting a leaf disease, the NodeMCU can also be used to activate an
automatic spraying system to treat the affected plants. Overall, the NodeMCU is a
versatile and cost-effective option for implementing a leaf illness detection and
automatic spraying system using the YOLOv5 algorithm.

Fig.3.3 NodeMCU ESP8266 Pinout

24
3.4 FLOWCHART FOR THE DETECTION OF INFECTED CROP

Fig.3.4 Flowchart for the Proposed System

Here is an alternative way to describe the flowchart in Fig 3.4 for a leaf
illness detection system:

1. Image acquisition: The system acquires an image of a leaf using a camera or

other imaging device.

2. Pre-processing: The acquired image is pre-processed to enhance its quality and

remove any unwanted noise or distortion.

25
 3. Feature extraction: Using algorithms such as colour-based segmentation,
texture analysis, and edge detection, the system extracts relevant features from the
pre-processed image.

4. Disease classification: The extracted features are used to classify the leaf as
healthy or diseased using machine learning algorithms such as CNNs or SVMs.

5. Decision-making: Based on the classification result, the system decides

whether the leaf needs treatment or not.

6. Treatment recommendation: If the leaf is identified as diseased, the system

recommends a suitable treatment method, such as spraying with a specific pesticide or
herbicide.

7. Reporting: The system generates a report that includes information about the
disease classification and the recommended treatment method.

3.5 SUMMARY

The proposed system can be integrated to check the quality of the soil and the
growth of crop in each soil. The sensors and microcontroller are successfully
interfaced and wireless communication is achieved between various nodes. All
observations and experimental tests prove that this project is a complete solution to
field activities and irrigation problems. Implementation of such a system in the field
can definitely help to improve the yield of the crops and overall production.

26
 Chapter-4

RESULTS AND DISCUSSION

4.1 EXPERIMENTAL RESULTS AND COMPARATIVE ANALYSIS

The YOLOv5 algorithm is capable of detecting leaf illness and automating

spraying systems. This is achieved through the use of a large dataset of labelled
images, which trains the algorithm to accurately identify various types of diseased
leaves and assess their severity. Once detection has taken place, an automatic spraying
system in Fig 4.1 is activated and targeted towards the affected areas with the correct
treatment. The automation process can significantly reduce crop losses and improve
crop management efficiency. YOLOv5's accuracy and real-time performance are
highly valuable features for the development of advanced agricultural technologies.

Fig.4.1 Demonstration of Proposed System

27
4.1.1 Experimental Dataset Building

To create an experimental dataset for leaf illness detection and automatic

spraying, you need to collect images of leaves that display different types of diseases
as well as healthy leaves. The images are then labelled with their corresponding
disease status using available tools to automate the process. After labelling, the
dataset is organized by creating folders for each category of leaf, such as healthy
leaves, leaves with fungus, leaves with pests, and so on. The labelled images are then
sorted into their respective folders. To improve the accuracy of the model, you can
augment the dataset by using image augmentation techniques to create additional data
from the existing images. Once the dataset is complete, it is split into training,
validation, and testing sets. The training set is used to train the model, while the
validation set is used to fine-tune the model's hyperparameters, and the testing set is
used to evaluate the model's performance. Once the model is trained, it can be used to
detect leaf diseases and automatically spray the affected leaves.

Table 4.1 Number of Samples of Each Disease Type

Serial Disease/Pest Sample size No.of. Percent of

Number labeled bounding box
samples samples

1 Early blight 1,209 12,187 8.30%

2 Leaf Curl 1,198 11,734 7.99%

Disease

3 Late blight 1,303 12,362 8.41%

4 Brown spot 1,348 11,726 7.98%

5 Leaf miner 1,228 11,580 7.88%

4.1.2 Comparative Analysis of Experiments

In general, the Yolo V5 algorithm is superior to the other three algorithms in

detection accuracy and speed, especially the detection time is the shortest. Therefore,
the Yolo V5 performs best for real-time detection tasks. The Yolo V5 algorithm can

28
complete the detection task of tomato diseases and insect pests well on the premise of
considering both the detection accuracy and detection speed.

Table 4.2 Comparison of Experimental Results

Algorithm Name Accuracy (%) Time/(ms)

SSD 84.32 25.69

Faster R-CNN 90.67 2868.94

YOLOv3 88.31 21.18

YOLOv5 97.39 20.39

mAP (%)

SSD Faster R-CNN YOLOv3 YOLOv5

Algorithm

Fig.4.2 Object size sensitivity analysis of Four Algorithms

The above figure (Fig.4.2) explains about the object size sensitivity SSD is a
popular object detection algorithm that uses a single deep neural network for both

29
object classification and detection. While SSD is fast and efficient, it struggles with
small objects, leading to decreased accuracy as object size decreases.

Faster R-CNN, a state-of-the-art object detection algorithm, employs a region

proposal network (RPN) to generate object proposals and a separate network to
classify and refine the proposals. While Faster R-CNN performs well across a wide
range of object sizes, it is not as fast as SSD.

YOLOv3 is a single neural network-based object detection algorithm that

detects objects in an image. YOLOv3 performs well on small and medium-sized
objects but experiences difficulty in detecting very large objects.

YOLOv5 is an advanced version of YOLOv3 that uses a more potent neural

network architecture and other enhancements to achieve better accuracy and speed.
YOLOv5 can detect small objects accurately and efficiently, making it a suitable
choice for object size sensitivity analysis across a wide range of object sizes.

Overall, all four algorithms have their strengths and limitations, but YOLOv5
is the most suitable algorithm for object size sensitivity analysis because of its high
accuracy and efficiency across various object sizes.

mAP (%)

SSD Faster R-CNN YOLOv3 YOLOv5

Algorithm

Fig.4.3 Object Image Resolution Sensitivity analysis of four algorithms

The above figure (Fig.4.3) explains about Object Image Resolution Sensitivity
while all four algorithms have their strengths and weaknesses, YOLOv5 appears to be
30
the best choice for object image resolution sensitivity analysis due to its ability to
maintain high accuracy in object detection even at lower image resolutions. Faster R-
CNN also performs well across a range of image resolutions, but it requires higher
image resolutions for accurate object detection. SSD and YOLOv3 may struggle with
accurate object detection at lower image resolutions.

Chapter-5

31
 CONCLUSION
5.1 CONCLUSION

The Proposed having an algorithm based on YOLOv5 to detect leaf diseases

and for automatic spraying. The YOLOv5 network is enhanced using multi-scale
feature detection, object bounding box dimension clustering, and multi-scale training
based on image pyramid. The experimental results indicate that the proposed
algorithm has a detection accuracy of 97.39% and a detection time of only 20.39 ms.

Consequently, the YOLOv5 algorithm proposed in this paper can achieve a

high detection rate and real-time detection requirements, while accurately and quickly
identifying the location and category of leaf diseases. Moreover, the YOLOv5 CNN
outperforms SSD, Faster R-CNN, and the original YOLOv3 by achieving higher
detection accuracy and shorter detection time, which meets the requirements of real-
time detection accuracy and speed of leaf diseases and pests.

Furthermore, performance analysis conducted in a small object scenario and

different resolution images confirms the robustness of the improved YOLOv5
network in detecting objects of different sizes and resolution images in complex
environments. The algorithm also demonstrates high detection and positioning
accuracy, which makes it ideal for detecting leaf diseases and pests in complex
environments.

5.2 FUTURE SCOPE

The future of leaf illness detection and automatic spraying system using
YOLOv5 algorithm is extensive and full of possibilities. Here are some potential
areas of development:

1. Enhanced accuracy: Although the YOLOv5 algorithm already exhibits

high accuracy in detecting leaf diseases and pests, there is always room for
improvement. Researchers can explore ways to further improve the accuracy of the
model by refining its training and testing procedures, employing additional data
augmentation methods, or creating better feature extraction techniques.

2. Real-time application: A real-time application of the YOLOv5 algorithm

for leaf illness detection and automatic spraying system is highly desirable. With the

32
increasing use of automation in agriculture, a real-time system would enable the
detection and treatment of leaf diseases and pests promptly and efficiently, resulting
in increased crop yield and reduced losses.

3. Multi-class detection: Currently, the YOLOv5 algorithm is designed to

detect only leaf diseases and pests. In the future, researchers can work on developing
a multi-class detection system that can identify other plant diseases or abnormalities,
such as nutrient deficiencies or drought stress.

4. Mobile deployment: The YOLOv5 algorithm can be deployed on a

mobile device, allowing farmers to conduct on-site detection of leaf diseases and
pests. By deploying the system on a mobile device, farmers can quickly detect and
treat affected crops, minimizing the spread of diseases and reducing crop losses.

5. Integration with spraying system: Integrating the YOLOv5 algorithm

with an automatic spraying system can enable the automatic treatment of crops after
detecting the presence of leaf diseases and pests. This integration can result in a fully
automated system that detects, diagnoses, and treats crops without any human
intervention.

Overall, the future of leaf illness detection and automatic spraying system
using YOLOv5 algorithm is promising, and its development can have a significant
impact on agriculture, leading to increased crop yield, reduced losses, and sustainable
farming practices.

33
 REFERENCES

[1] K. Gurung, S. N. Vink, J. F. Salles, and B. Wertheim, "More persistent bacterial

than fungal associations in the microbiota of a pest insect," 2022.

[2] Y. Diller, A. Shamsian, B. Shaked, Y. Altman, B.-C. Danziger, A. Manrakhan, L.

Serfontein, E. Bali, M. Wernicke, A. T. Papadopoulos, and D. Nestel, "A real-time
remote surveillance system for fruit flies of economic importance: sensitivity and
image analysis," Journal of Pest Science, 2022.

[3] R. Brodschneider, J. Schlagbauer, I. Arakelyan, J. Brus, V. Brusbardis, M.

Fedoriak, and I. Forsythe, "Spatial clusters of Varroa destructor control strategies in
Europe," Journal of Pest Science, 2021.

[4] L. Zapponi, R. Nieri, V. Zaffaroni Caorsi, N. M. Pugno, and V. Mazzoni,

"Vibrational calling signals improve the efficacy of pheromone traps to capture the
brown marmorated stink bug," 2022.

[5] J.-W. Chen, W.-J. Lin, H.-J. Cheng, C.-L. Hung, C.-Y. Lin, and S.-P. Chen, "A
smartphone-based application for scale pest detection using multiple-object detection
methods," vol. 10, no. 4, 2021.

[6] Lena Durocher, Granger, Tibonge Mfune, Monde Musesha, Alyssa Lowry,
Kathryn Reynolds, Alan Buddie, Giovanni Cafà, Lisa Oxford, Gilson Chipabika,
Marcel Dicke, and Marc Kenis, "Factors influencing the occurrence of fall armyworm
parasitoids in Zambia," 2021.

[7] Channamallikarjuna Mattihalli, Edemialem Gedefaye, Fasil Endalamaw, and

Adugna Necho, "Real time Automation of Agriculture Land, by Automatically
Detecting Plant Leaf Diseases and Auto Medicine," 32nd International Conference on
Advanced Information Networking and Applications Workshops, 2021.

[8] Glenn Jocher, Bochkovskiy, A., Wang, C.-Y., Mark Liao, H.-Y., "YOLOv5:
Optimal Speed and Accuracy of Object Detection," arXiv, arXiv:2004.10934, 2020.

[9] V. Wang, H. Zhang, S. Zhao, S. Wang, Q. Zhu, Li, D., and Zhao, R., "YOLO-V5
FIRI:Improved YOLOv5 for Infrared Image Object Detection," IEEE Access, 2020.

34
[10] Apeksha Thorat, Sangeeta Kumari, and Nandakishor D. Valakunde, "An IoT
Based Smart Solution for Leaf Disease Detection," International Conference on Big
Data, IoT and Data Science (BID), Dec 20-22, 2017.

[11] Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi, "You Only
Look Once: Unified, Real-Time Object Detection," Proceedings of the IEEE
conference on computer vision and pattern recognition, Las Vegas, NV, 2016.

[12] Vijai Singh, Varsha, and A. K. Misra, "Detection of unhealthy region of plant
leaves using Image Processing and Genetic Algorithm," International Journal of
Advanced Research in Computer and Communication Engineering, vol. 5, no. 6, pp.
1-5, June 2016.

[13] Amreen Abbas, Sweta Jain, Mahesh Gour, and Swetha Vankudothu, "Plant
disease Detection using transfer learning with C-GAN synthetic images," vol. 1, pp.
8186, 2016.

[14] Sanjay B. Dhaygude and Nitin P. Kumbhar, "Agricultural plant Leaf Disease
Detction Using Image Processing," International Journal of Advanced Research in
Electrical, Electronics and Instrumentation Engineering, vol. 2, no. 1, pp. 253-259,
January 2013.

[15] Arti N. Rathod, Bhavesh Tanawal, and Vatsal Shah, "Image Processing
Techniques for Detection of Leaf Disease," International Journal of Advanced
Research in Computer Science and Software Engineering, vol. 3, no. 11, pp. 616-619,
November 2013.

[16] Smita Naikwadi and Niket Amoda, "Advances in Image Processing for Detection
of Plant Diseases," International Journal of Application or Innovation in Engineering
& Management, vol. 2, no. 11, pp. 68-73, November 2013.

[17] S. Arivazhagan, R. Newlin Shebiah, S. Ananthi, and S. Vishnu Varthini,

"Detection of unhealthy region of plant leaves and classification of plant leaf diseases
using texture features," Agric Eng Int: CIGR Journal, vol. 15, no. 1, pp. 211-217,
2013.

35
[18] S. Bashir and N. Sharma, "Remote Area Plant Disease Detection Using Image
Processing," IOSR Journal of Electronics and Communication Engineering (IOSR
JECE), vol. 2, pp. 31-34, 2012.

[19] L. Granger, T. Mfune, M. Musesha, A. Lowry, and K. Reyolds, "Factors

influencing the occurrence of fall armyworm parasitoids in Zambia," 2012.

[20] J. Zhu, J. Wang, H. Liu, and H. Mi, "Insect Identification and Counting in Stored
Grain: Image Processing Approach and Application Embedded in Smartphones," vol.
54, no. 2, pp. 211-217, 2011.

36