See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/362722166

Leaf Disease Detection of Multiple Plants Using Deep Learning

Conference Paper · May 2022

DOI: 10.1109/COM-IT-CON54601.2022.9850899

CITATIONS READS
19 60

6 authors, including:

Shital Pawar
Bharati Vidyapeeth's College of Engineering for Women Pune
31 PUBLICATIONS 67 CITATIONS

SEE PROFILE

All content following this page was uploaded by Shital Pawar on 05 February 2024.

The user has requested enhancement of the downloaded file.

 2022 International Conference on Machine Learning, Big Data, Cloud and Parallel Computing (COM-IT-CON), 26-27 May 2022
2022 International Conference on Machine Learning, Big Data, Cloud and Parallel Computing (COM-IT-CON) | 978-1-6654-9602-5/22/$31.00 ©2022 IEEE | DOI: 10.1109/COM-IT-CON54601.2022.9850899

Leaf Disease Detection of Multiple Plants Using

Deep Learning
Shital Pawar1, Sakshi Shedge2, Nibedita Panigrahi3, Jyoti A P4, Pradnya Thorave5, Samina Sayyad6
1,2,5,6
Computer Engineering Department, Bharati Vidyapeeth’s College of Engineering for Women, Pune, India
3
Information Science and Engineering Department, RV Institute of Technology and Management, Bengaluru, India
4
Computer Science and Engineering Department, Ramaiah University of Applied Sciences, Bengaluru, India
1
[email protected]
2
[email protected]
5
[email protected]
6
[email protected]

Abstract : The great advancement in deep learning has definitely required for appearance of diseases in plants and can
increased the interest in technologies of image recognition. be transmitted to all over the plant tissues such as leaves,
Different deep learning techniques help big data analysis with shoots, trunks, canopy, roots, berries, beans, arteries etc.
large potential and finding the disease of plant in agriculture Therefore, diagnosis and disease detection is very important
sector. The deep learning model extract the features from an
input and uses its features for classification. In consideration of
task [5]. A specialist eye examination is a majorly used method
climate change, plants are affected, which impacts agricultural in the diagnosis and treatment of plant illnesses. Still, this
yield. When the condition worsens, the plants become more requires professional attention, which can be very expensive in
susceptible to many diseases caused by fungi, bacteria, viruses, large farms [6]. Using an auto-diagnostic method is beneficial
etc. Diseases can affect the plant growth. In the proposed in diagnosing plant disease early. One can examine the picture
system, CNN algorithm is used to detect the disease from leaf of the leaf with disease by using monitor image processing
images and pesticides are suggested based on the disease found. research and extracting the features from leaf image like color,
The proposed method uses 15-layer CNN. The proposed system appearance, and other features to detect the leaf diseases [7- 8].
will provide pesticide supplier name according to area as well as
it provides additional information about the leaf disease of the As a result, the traditional method of diagnosing leaf diseases
affected plant in the area, disease name, complete accuracy, by a knowledgeable person is more costly and time consuming.
time, and weather forecasting information also. The dataset In such a situation, the proposed process seems to favor large
contains 10 plants - apple, potato, corn, tomato, grape, rice, plantations. The automatic fetching of diseases just by looking
sugarcane, cucumber, pepper, soyabean. at the indications on the leaves of the plant is easy and cheap.
Keywords— Pretrained Deep Models, feature, CNN, Visual verification of plant illness is very difficult and at the
extraction, , image processing, leaf image, disease same time a little precise process and can only be done in
detection, pesticide suggestion. restricted areas. Although when the automatic detection
method is used, even it takes less effort, it estimates time and
I. INTRODUCTION greater accuracy [9]. In plants, other common infections are
brown and like yellow spots, or premature or late extinction, or
In Indian economy the agricultural field plays an essential role. other fungus diseases, affected by microbes and viruses .
The aim of this agricultural field is to increase the agricultural Image processing is a method used to measure an infected area,
yield without considering surrounding effects which causes as well as to determine the color differences of the infected
environmental degradation. In today's world, agricultural land area [10].
is far more valuable than just the food industry. The Indian
economy is heavily dependent on agricultural products. In earlier days the farmers were relying on the expert’s
Whenever the leaf of any plant gets infected, this infection gets inspection for detection of plant diseases. But this method is
spread in all parts of that plant which will results into too expensive and time consuming. Researchers have started
decreasing the farmer’s income. Therefore, the detection of using the techniques of Artificial Intelligence and Machine
plant leaf disease is very essential in agricultural sector [1- Learning to detect the plant leaf diseases faster and more
2].On a global scale, the various diseases in plant are not just intelligently. The techniques of Machine learning are
warnings for food security but, it has catastrophic influences successful in case of specific setups only, certain variation in
for emerging farmers whose daily bread depends on healthy some conditions may results into performance degradation of
crops. More than 80 percent of agricultural production in the Machine Learning models. So, now the researchers are using
developing world is produced by smallholder farmers and the Deep learning with CNN in leaf disease detection as with
reports of more than 50% of crop losses due to pests and CNN they can use huge amount of image data and can find out
diseases [3-4].Plant germs includes fungus, organisms, the distinguishing features from those images, avoids image
pathogens, germs, microbes , viruses, , etc. Three nutrients are processing and reducing the memory requirement. CNN

978-1-6654-9602-5/22/$31.00 ©2022 IEEE 241

Authorized licensed use limited to: VIT University. Downloaded on January 26,2024 at 08:56:58 UTC from IEEE Xplore. Restrictions apply.
detects the plant leaf diseases more accurately. So, in smart II. BACKGROUND AND RELATED WORK
agriculture CNN is now popularly used for detection of plant
leaf diseases [11]. T. Vijaykant Reddy et. al. has proposed the Deep Leaf
Disease Prediction Framework (DLDPF) model which is
compared with various models of deep learning like AlexNet In this paper, the CNN's design approach is used to detect the
and GoogLeNet. Researchers have used the dataset of apple leaf diseases of 10 plant species. CNN's main idea is to build a
leaves and they have achieved better accuracy than existing leadership network and provide the good accuracy for leaf
models. The training time and memory requirements of this disease detection.
framework are less. In future researchers can assess the other III. MOTIVATION
deep learning model like RNN with upgradation [1].
For identification of plant diseases, the traditional approach
Arunabha M. Roy and Jayabrata Bhaduri have done the used by experts is observation of plants and leaves through
diagnosis of apple leaf disease using Advanced Reading, CNN. naked eyes. But this approach of disease detection needs
The current work allows an effectful and efficient way to continuous watch on plants by experts which is too expensive
diagnose a variety of plant diseases below complex conditions
for large farm fields. Furthermore, the farmers from the
and could be enlarged to the acquisition of a variety of fruits
developing countries must have to travel the long distance for
and vegetables , the diagnosis of common diseases, and
automatic agricultural diagnostic procedures [2]. meeting with experts which makes the consultation time
consuming as well as expensive. So automatic detection of
Melike Sardogan et.al. has presented CNN model and LVQ plant diseases is an important research issue which is
algorithm for detection of leaf diseases of tomato plant. beneficial for automatic detection of symptoms of plant
Researchers have used 500 images of tomato leaves and diseases once the diseases start appearing on plant leaves.
identified the improvement of disease recognition rate. Various Therefore, for detection of plant diseases there is a need of
filters and distinct convolution sizes can be applied in fast, accurate, less costly, and automatic method [7].
classification phase [3].
Farmers often use the naked eye conservation method for
Nikita Goel et.al has presented a method for automatic identifying the diseases. But sometimes this may be the wrong
detection of diseases in plants. The main target of this work is approach. Most of the time, farmer has to call the specialists to
detection of diseases in pomegranate leaf. Initially the initial diagnose the diseases in large farms which is time-consuming
processing was done. The second stage occurs when the K- process. Use of digital imaging processing techniques, many
Means algorithm is used for all images available in image applications have been accepted in various fields like industrial
database. The third category involves the removal of a features testing, medical imaging, remote sensing, agricultural
which combines color and shape features. In future, specific purifying etc. Thus, leaf segregation is necessary to evaluate
application can be developed which provides facility to the agricultural production, increase market value and meet quality
researcher to upload the real time plant image dataset and get standards. The disease identification is helpful to stop
the disease name and solutions for curing the diseases [4]. spreading the diseases in plants [8].
Sandeep Kumar et. al. has proposed the differentiation method
such as k-methods is used to exclude various gray matter IV. RESEARCH GAP
characteristics and SVM for classification of diseases. This Most of the researchers have used the Deep learning approach
method helps in detection of various plant leaf diseases for leaf disease detection but they have done the disease
accurately. Researchers have compared their results with detection for one plant. The research can be extended to cover
existing methods and identified how much percentage of leaf various other multiple plants. Most of the researchers have
area is affected by the predicted disease [5]. focused on just improving the accuracy of classification and
Peng Jiang et.al has proposed the model which uses deep achieving high accuracy for disease detection. Many
learning with CNN for leaf disease detection of Apple plant in researchers have got success in getting the maximum accuracy
real time. Author have used around 26000 images of apple for disease detection, but this accuracy drastically reduced
leaves and detected 5 leaf diseases with great accuracy [6]. when the Deep learning models were tested for the datasets
The deep learning models have become another attractive and which were not the part of same database. This is still
effective way to detect leaf diseases compared to the traditional unresolved issue. Sometimes the visual appearance of some
models/techniques. The main reason behind this is the deep symptoms of different diseases are similar. So, it’s
learning models are able to handle the huge amount of data challenging to detect such diseases for which the advanced
with the support of previously trained deep learning models [1- technique is required [1-5]. There is a requirement to develop
3]. For input leaf images are taken from dataset and a model which can be applied to multiple datasets [11-12].
classification by using leaves like inserts, in depth research
models examining different crops in agriculture. Accuracy V. SCOPE AND ADVANTAGES
rates in deep learning methods are higher than the traditional a. Scope: In proposed system for detection of leaf diseases
methods. And the traditional method is dependable on the multiple plants are considered. The proposed system can detect
character of the database, hence it shall not be the norm. The diseases for 10 plants. These 10 plants are categorized in two
deep learning is very effective in classifying images [11-12]. different categories like, fruit plants and crops. The List of 10

242
Authorized licensed use limited to: VIT University. Downloaded on January 26,2024 at 08:56:58 UTC from IEEE Xplore. Restrictions apply.
plants is shown in table 1. The system provides pesticide pesticide, address and phone number of pesticide supplier
suggestion and pesticide supplier information like name of through text SMS.

datasets. In Proposed system instead of using models which

Instead of considering the specific plant only, in this paper the
multiple plants are considered in one system. So, it’s necessary are already developed by other universities or researchers,
to train the CNN model according to the different plant customized model is developed from scratch. It uses three
categories. Forecasting the weather can support with the layers for feature extraction these are Convolutional, Relu,
farmer's business commitment. The weather forecasts can Pooling, three layers for classification namely flatten, fully
support them to plan numerous daily decisions. connected and softmax and it uses three stages of feature
extraction layers (3*3 =9) and two stages of classification
Table 1: List of plants for leaf disease detection
layers (3*2 = 6) so that the method makes the use of 15-layer
CNN. The proposed system provides additional things which
FRUIT PLANTS CROP PLANTS the existing system doesn’t are system can detect the diseases
in 10 different plants very accurately, the proposed
methodology uses customized model with the intent of speed
APPLE POTATO
efficiency and accuracy, provides the pesticide suggestion and
pesticide supplier information to the farmer through text SMS,
GRAPES RICE it also provides the weather forecasting report which includes
the information like Date, day, Time, temperature of the day.
CUCUMBER CORN Convolutional Neural Network (CNN) Architecture:
Formula for convolution layer:
SUGARCANE SOYABEAN
Win – F + 2p
TOMATO PEPPER Wout = +1 --- eq. (1)
S
The existing models have not provided the pesticide proposal
and supplier information. In addition to this the existing Where,
systems are limited to detect the diseases of only the specific Win is size of input
plant. The proposed system will detect the leaf diseases for Wout is width of output
multiple plants. The proposed system will also provide the F is spatial size
pesticide suggestion details as well as contact details of the S is stride
pesticide provider. Weather forecast information will also be p is padding
provided.
Following layers are included in CNN [7-8]:
b. Advantages: The proposed system has following
advantages. a. Convolutional layer: This layer is structured block of CNN
which produces dot production between two matrices, where
1. Deep learning with CNN is the most accurate technique first matrix contains a set of learnable variables and the other
for detection of leaf diseases of various plants in one matrix is a finite part of the reception field.
agricultural field.
b. Relu Layer: Relu Represents a Fixed Line Unit that
2. The proposed system can detect the leaf diseases
combines function f(k) = magnitude (0, k). It means that the
automatically.
Relu Layer converts a negative value into zero or greater
3. The proposed system provides the information like
value.
disease name, date and time, weather forecasting, name
of pesticide and pesticide supplier contact information to c. Pooling Layer: Pooling layer minimizes size of the
the farmers. representation area, which reduces the required amount of
4. The proposed system can detect the diseases for multiple calculation and weight as well.
plants. d. Flatten Layer: The output from the last compound and
convolutional layer is a 3-dimensional matrix, to be considered
VI. PROPOSED METHODOLOGY which releases all its values into a vector. Then that flat vector
is connected to the Fully connected layer.
The system detects diseases using CNN with Deep Learning
and suggests pesticide according to detected disease and e. Fully Connected layer: The output of convolutional layer is
provide pesticide supplier information too. Deep learning can provided as an input to this layer which will be flattened and
extract features from images automatically and CNN is deep then provided as an input to a fully integrated layer.
learning model which is capable to do the efficient
classification as well as can handle large and complex

243
Authorized licensed use limited to: VIT University. Downloaded on January 26,2024 at 08:56:58 UTC from IEEE Xplore. Restrictions apply.
f. Softmax Layer: Softmax is a last layer in CNN. It gives For feature Extraction the first 3 layers of CNN are used 3
output and carry out the classification which is multi class. It times. For classification next 3 layers are used. For
is used to determine the feasibility of a specific category. classification 6 stages are considered. So, in this way in
proposed model the15 layers of CNN are used. This 15- layer
Out of all available architectures in deep learning, CNN is CNN model is used to train the plant leaf image dataset.
popularly applied for recognition of images. Deep learning Further, the trained model will be used to detect the diseases
with CNN has got a great success in leaf disease detection. In from plant leaf images.
CNN there are total 6 layers. In proposed model, these 6
layers of CNN are used as follows:

Fig. 1: CNN Architecture

VII. SYSTEM ARCHITECTURE Convolutional, Relu, Pooling layer of CNN for training the
Initially there is dataset of leaf images in which healthy leaf model. So, by using these 3 layers repeatedly the 15 layers of
images are stored at some place and different types of diseased CNN model will be developed and trained.
images are stored at different place means that is in the Once the model is trained a leaf image will be provided as an
another folder. For the proposed system more than 50000 leaf input to this model through preprocessing and model will
images of 10 plants are used for leaf disease detection. predict the type of disease. The system will send the SMS of
Preprocessing will done on all the leaf images. In the predicted disease to the farmer and based on that disease
preprocessing all images will be converted to a standard size. system will suggest the pesticide or fertilizer to spread on the
After preprocessing Convolutional Neural Network (CNN) plant leaf to cure that disease and pesticide supplier
will be trained, as authors are using the three layers information will also been provided through text SMS.

244
Authorized licensed use limited to: VIT University. Downloaded on January 26,2024 at 08:56:58 UTC from IEEE Xplore. Restrictions apply.
 Fig. 2: Proposed System Architecture
system can be integrated with the specific mobile application
VIII. CONCLUSION through which user can detect the leaf ddiseases just by taking
The aim of this paper is to predict the category of disease photograph of the plant leaf.
appeared on a leaf of a plant. For this prediction, Convolutional
Neural Nnetwork is used which is an unregulated method. This REFERENCES
paper deals with application of Convolution Neural Network
[1] T.Vijaykanth Reddy, et.al. “Deep Leaf Disease Prediction Framework
(CNN) to recognize plant diseases.. One of the most important (DLDPF) with Transfer Learning for Automatic Leaf Disease
image processing tools is image recognition which is an Detection”, 2021 Fifth International Conference on Computing
important tool for diagnosing diseases early in order to dologies and Communication (ICCMC 2021), pp.1408-1415 .
Methodologies
increase crop production. This tool will help to reduce the time [2] Roy, Arunabha M.,et.al. "A Deep Learning Enabled Multi-Class
Multi Plant
and cost of using personal
onal guesses. By taking all the factors Disease Detection Model Based on Computer Vision"
Vision", 2021 AI 2, no.3:
that work into thinking the conclusion is that the Convolution 413-428.
Neural Network provides remarkable accuracy in diagnosing [3] Sardoğan, Melike, et.al. “Plant Leaf Disease Detection and
Classification Based on CNN with LVQ Algorithm.” 2018 3rd
diseases. This work can also be extended to create a real-time
real International Conference on Computer Science and Engineering
system that can detect disease inn multiple plants by simply (UBMK) (2018), pp. 382-385.
adding a set of crop and training data. The system uses [4] N. Goel, et. al. "Prediction Model for Automated Leaf Disease
Gaussian size, limit and filter to pre-scan
scan the image to separate Detection & Analysis," 2018 IEEE 8th International Advance
the leaf area, and finally the proposed system with CNN Computing Conference (IACC),, 2018, pp. 360-365, 360 doi:
sorting process is helpful to determine the categcategory of leaf 10.1109/IADCC.2018.8692116
disease with up to 93% accuracy. Proposed system provides [5] Sandeep Kumar, et al. “Leaf Disease Detection and Classification based
the weather forecasting information to the farmers which can on Machine learning”, International Conference on Smart Technologies
Te
2020), pp.361-365.
in Computing, Electrical and Electronics (ICSTCEE 2020)
help the farmers in making day to day decisions like which
[6] P. Jiang, Y. Chen, B. Liu, D. He and C. Liang, "Real
"Real-Time Detection of
fertilizer is suitable in which whether condition. As by Apple Leaf Diseases Using Deep Learning Approach Based on
applying proper fertilizers according to weather condition can Improved Convolutional Neural Networks," in IEEE Access, vol. 7, pp.
help farmers to protect the crops. This system is very helpful to 59069-59080, 2019.
the farmers as it detects the leaf diseases and provides the [7] K. P. Panigrahi ,et al. “Maize Leaf Disease Detection and Classification
name of pesticide and the pesticide supplier contact Using Machine Learning Algorithm”,, 2020, Springer, pp.659-669.
pp.659
information too according
ing to the detected disease.
disease [8] S. Gayathri, et al. “Image Analysis and Detection of Tea Leaf Disease
using deep Learning, 2020,” International Conference on Electronics
IX. FUTURE SCOPE and Sustainable Communication Systems (ICESC 2020), pp.398-403.
pp.398
[9] Sammy V. Militante et al. “Plant Leaf Detection and Disease
In future additional set of data along with the most accurate Recognition using Deep Learning, 2019,” IEEE Eurasia Conference on
algorithm can be applied to find web-based
based image processing IOT,, Communication and Engineering, 2019, pp.579-582.
pp.579
techniques. For the user two methods can be provided namely [10] Prajwala TM, el al. “Tomato Leaf Disease Detection using
Convolutional Neural Networks,” Eleventh International Conference on
online and offline. In the case of web domain processing Contemporary Computing (IC3), 2018.
remote users can upload an image to the system and get leaf [11] S. Pawar, P. Bagal, P. Shukla and A. Dawkhar,
Dawkh "Detection of Breast
disease detection result by using cloud . Real Real-time data Cancer using Machine Learning Classifier," 2021 Asian Conference on
monitoring can be using the cloud platform. (ASIANCON) 2021, pp. 1-5, doi:
Innovation in Technology (ASIANCON),
10.1109/ASIANCON51346.2021.9544767.
There is a scope for detection of leaff diseases from different [12] Sk Mahmudul Hassan et al., “Identification of Plant Leaf Diseases
geographical locations. Inn addition to just detection of name of Usingg CNN and Transfer Learning Approach, 2021,” Electronics, 2021,
leaf disease, detection of various phases of that disease. The pp.1-19

Authorized licensed use limited to: VIT University. Downloaded on January 26,2024 at 08:56:58 UTC from IEEE Xplore. Restrictions apply.

View publication stats