Convolutional Neural Networks-Based Potato Leaf Disease

Classification

Ravinder Kaur

and Sonam Khattar

Department of Computer Science and Engineering, Thapar Institute of Engineering and Technology, Patiala, India

Department of Computer Science and Engineering, Chandigarh University, Mohali, India

Keywords: Classification, Leaf Disease, Deep Learning, Potato Plant, CNN

Abstract: The potato has risen to the position of fourth most eaten staple food in the world, among many others.

Additionally, the world's population is the primary driver of the dramatic increase in potato consumption. But

the main reason why the crop isn't as good as it might be because of potato illnesses. Things will become

much worse for the plants if the illness is misclassified and is discovered too late. Fortunately, leaf conditions

may be used to identify a number of illnesses in potato plants. Thus, this study introduces a method that uses

deep learning convolutional neural network architectural model to accurately diagnose the four kinds of potato

plant illnesses based on leaf conditions. The experiment has shown that the deep neural network based VGG19

technique is able to produce significant results, with an average accuracy of 99.07%.

1 INTRODUCTION

Numerous countries all over the globe consider

potatoes to be a fundamental component of their

cuisine, and they are well-known all over the world.

The term "the root of all vegetables" refers to potatoes

in another context. In light of the fact that India's

economy is mostly based on agriculture and that the

country cultivates a wide range of crops, potatoes

play an important part in our society. India follows

the United States as the globe's second-biggest

supplier of potatoes (Wasalwar, Bagga, et al. , 2023).

Expanding potato production is the most important

issue since the demand for potatoes throughout the

world is always growing, and our area has to start

exporting as much as it can. It is a fact that several

severe illnesses that affect the leaves of potatoes, such

as early blight, late blight, and others, have led to a

decline in export and production throughout the

course of the previous few years. The majority of

these issues are brought on by the delayed detection

of infections in potato plants as well as errors in the

recognition of illnesses. In order to lessen the

negative effects that diseases have on potato plants, it

is of the utmost importance to properly and promptly

identify any infections that may be present in potato

https://orcid.org/0000-0002-8535-6617

https://orcid.org/0000-0002-5444-4358

plants. The activities of manual monitoring that are

carried out by farmers become challenging and

impracticable due to the fact that they need a

significant amount of time and in-depth expertise.

The identification of plant illnesses that are slower

will cause the spread of diseases in plants to occur in

an uncontrolled manner. Because of this, farmers are

also subjected to adversity. Farmers utilize the

findings of personal identification as a reference in

the protection of plants that are affected by the

disease. This is done without the assistance of an

expert in the area of plant diseases who is guiding

them. As a consequence of this, the preventative

actions that farmers take may not be successful and

may cause harm to their crops owing to a lack of

information combined with an incorrect perception of

the severity of the illness, an excessive dose, or an

insufficient dosage (Sholihati, Sulistijono, et al. ,

2020). The proposed study aims to provide farmers

with a method that is both quick and accurate for

recognizing and categorizing illnesses that affect

potato plants. In this research work, CNN-based

VGG19 classification technique is being proposed for

the categorization of potato leaves. To summarize,

the remaining parts of the paper are organized as

follows: Section 2 makes reference to the body of

Kaur, R. and Khattar, S.

Convolutional Neural Networks-Based Potato Leaf Disease Classiﬁcation.

DOI: 10.5220/0013595500004664

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 2, pages 493-497

ISBN: 978-989-758-763-4

493

work that is associated with the classification of

disease detection in potato leaf cases. In the third

section, the recommended approach is discussed, and

then in the fourth section, the results of the

experiments and discussions are presented. In the

fifth and final section, the conclusion is presented

2 LITERATURE SURVEY

Distinct agricultural issues that CNN-based deep

learning models have been able to solve include the

detection of weeds, the diagnosis of microbiological

diseases, the identification of crop bugs, and the

classification of fruits. Through the process of

training the network using images of both good and

sick plants, the CNN model can accurately detect

plant illnesses. This section discusses the technique

of diagnosing unhealthy plant leaves by analyzing

their images using a CNN model (Bommala, Babu, et

al. , 2023) Figure 1 depicts three distinct images of

potato leaves. Joseph et al. (Joseph, Ashraf, et al. ,

2022) proposed a lightweight CNN-based model for

the classification of tomato leaves taken from the

plant village dataset available at Kaggle. The authors

have employed CNN-based architecture made by

using three convolution layers and one max pooling

layer. The experimental outcomes of the proposed

lightweight model outperform both cutting-edge

machine learning methods and pre-trained models in

terms of accuracy when applied to the publicly

accessible Plant Village dataset. Hylmi et al. (Hylmi

and Suryani, 2022) employed an image segmentation

technique and a Multi-Class Support Vector Machine

to construct a system for detecting potato leaf

diseases. Image segmentation output is derived from

a Red Green Blue (RGB) color histogram, a Gray

Level Co-occurrence Matrix (GLCM) for texture

features and computations of leaf spot contours for

shape features. The subsequent phase is to use a linear

kernel Multi-Class Support Vector Machine (SVM)

to perform the categorization. The technique

developed in this work has a detection accuracy of

97.56% when applied to potato leaves.

Rusli et al. (Rusli, Meng, et al. , 2022) proposed a

technique based on the K-Means clustering algorithm

for performing image segmentation followed by

feature extraction using GLCM. The extracted

features are further given as input to the Artificial

Neural Network (ANN) classifier for performing the

classification of healthy and diseased potato leaves

with the reported accuracy equivalent to 94%. Bonik

et al. (Bonik, Akter, et al. , 2023) proposed a method

that uses a sequential model based on convolutional

neural networks to foresee potato leaf diseases. The

experimental results revealed an accuracy equivalent

to 94%. However, Cross-validation and hyper

parameter optimization techniques, on the other hand,

were prominently missing from a significant number

of the research papers that were already available. As

a result of examining the studies that have already

been conducted, it is clear that there are a number of

research gaps and limits that need to be addressed in

order to improve the predictive capabilities of plant

leaf diseases.

Figure 1: Sample Images from the dataset (from left to

right) Potato Early Blight, Potato Healthy, Potato Late

Blight.

3 METHODOLOGY

Figure 2, provides an illustration of the processes that

are involved in the methodology that has been

proposed. Data collecting, preprocessing,

categorization, and performance evaluation using

different metrics are among the various steps that are

involved in this process.

Figure 2: llustration steps involved in the methodology

3.1 Data Acquisition

The dataset that was employed in this investigation

consisted of photos of potato leaves that were labeled

according to three distinct categories: healthy leaves,

early blight, and late blight. The authors have

acquired the "PlantVillage Dataset" from the kaggle

repository to conduct this work (Khattar and Verma,

2023).

INCOFT 2025 - International Conference on Futuristic Technology

494

3.2 Pre-processing

Here, the two primary pre-processing operations,

resizing and rescaling, are put into action. Due to the

large number of photographs with varying pixel sizes,

all of them were resized to 256 × 256. This pre-

processing step improves the model's performance by

ensuring that all the input photos are of the same size.

A normalization approach was used for rescaling,

which separates the image's pixel values—which may

be anywhere from 0 to 255 in color pictures—into the

0 to 1 range (Kaur and Juneja, 2020), (Borawar and

Kaur, 2023). The suggested CNN model includes

Layers including Convolution, Max or Average

Pooling, and Regularization (Dropout) that must

perform some computation at every level. In addition,

there is a ratio of 80:10:10 between the train,

validation, and test sets in the

dataset.

3.3 Classification

While performing the classification task, several

methods for enhancing data were applied to the

training set to enhance the size of the dataset and

increase its diversity, including distinct rescaling and

flipping techniques. The classification model

underwent training by utilizing data from the input to

the output layers, which led to the formulation of

predicted results and then the recognition of any

mistakes or results. Back-propagation technique was

employed here to adjust model weights if the

predicted results came out to be inadequate and the

Adam optimization technique was also utilized to

enhance the classification results. During

convolutional procedure, the input volume and

weights were both convolutioned. Because of the

stride and padding, the size of the convolved matrix

can either be decreased or expanded, depending on

the situation at hand. In spite of the fact that the depth

grew, the spatial height and breadth decreased.

Following the application of the ReLu nonlinear

action function to each convolutional layer, the

likelihood of the gradient vanishing was reduced, and

negative values were brought down to zero. Further

with the use of max pooling, the photographs were

down sampled, which resulted in a reduction in

overfitting, an improvement in the performance of the

activation function, and a speeding up of

convergence. The ultimate output layer, which was

responsible for identifying a category of potato leaf

pictures, was dense or completely intertwined.

4 EXPERIMENTAL RESULTS &

DISCUSSION

The identification of plant leaf diseases by deep learning

algorithms leads to a considerable improvement in

crop output and quality. This is accomplished by

reducing the biotic factors that are responsible for

significant levels of agricultural production losses.

The purpose of this research is to offer CNN based

deep learning model that is both quick and easy, to

diagnose diseases that harm potato leaves. Following

the training of the models with eighty percent of the

data, the Training Accuracy and Loss are computed

for suggested model. After that, ten percent of the

dataset was validated. A graph was created to

represent the training and validation accuracy as well

as the loss is shown in Figure 3.

Figure 3: Training and validation Accuracy

Figure 4: Training and validation loss

Convolutional Neural Networks-Based Potato Leaf Disease Classiﬁcation

495

The ImageNet was used to train the CNN-based

VGG-19, that was trained on more than one million

images. With a total of nineteen layers—six

convolutional, three fully connected, five MaxPool,

and one SoftMax—this model is an adaptation of the

VGG-16 model. A library of high-quality image

representations has been built by the network. In

order to classify potato leaf diseases, the

experimental findings show that the suggested CNN

based on the VGG19 model has low testing loss and

good testing accuracy. The inference function, which

was developed, received the VGG16 Trained Model

as a parameter.

Table 1: Comparison with existing techniques in terms of

accuracy

Figure 5: Barchart to show comparison with existing

techniques

The system accurately identified the Potato leaf

disease with a high level of certainty, using example

images as input (Kaur and Juneja, 2018), (Kaur and

Juneja, 2016), (Kaur and Juneja, 2021). Table 1

below shows the accuracy of the proposed method

and other existing techniques. Further, Figure 4

shows the comparison of the proposed method with

the existing techniques in graphical form.

5 CONCLUSIONS

This paper presents a classification approach for

recognizing diseased and normal leaf images of

potato plants by utilizing convolution neural network

based method. When it comes to the classification of

potato leaf diseases, it was shown that CNN is an

effective tool. It is possible to forecast with a

validation accuracy of 99% with this strategy. In

general, this kind of initiative will be of utmost

significance for the country's agricultural sector. The

results of this study indicated that CNN gave the

highest level of accuracy in comparison with other

techniques for recognizing and categorizing potato

leaf diseases such as early blight, late blight, and non-

infected. It is possible to draw the conclusion that the

techniques that are currently in use, such as

personally evaluating each and every plant in a

region, are both time-consuming and ineffective in

comparison to the CNN methodology that is used to

diagnose potato leaf diseases. This would be done

with the intention of assisting the agricultural

business, and farmers in particular, in recognizing

and diagnosing potato plant diseases at a cheaper cost

and, most importantly, in a shorter amount of time.

REFERENCES

Wasalwar, Y. P., Bagga, K. S., Joshi, V. K., & Joshi, A.

(2023, April). Potato Leaf Disease Classification using

Convolutional Neural Networks. In 2023 11th

International Conference on Emerging Trends in

Engineering & Technology-Signal and Information

Processing (ICETET-SIP) (pp. 1-5). IEEE.

Sholihati, R. A., Sulistijono, I. A., Risnumawan, A., &

Kusumawati, E. (2020, September). Potato leaf disease

classification using deep learning approach. In 2020

international electronics symposium (IES) (pp. 392-

397). IEEE.

Bommala, H., Babu, N. J., Srikanth, P., Mallidi, S. K. R.,

Sai, T. S. R., & Mounika, R. (2023, September).

Detecting Diseases in Potato Leaves using Deep

Learning and Machine Learning Approaches: A

Review. In 2023 4th International Conference on Smart

Electronics and Communication (ICOSEC) (pp. 788-

792). IEEE.

90%

92%

94%

96%

98%

100%

Accuracy

Existing and proposed study

Accuracy

INCOFT 2025 - International Conference on Futuristic Technology

496

Joseph, S. G., Ashraf, M. S., Srivastava, A. P., Pant, B.,

Rana, A., & Joshi, A. (2022, October). Cnn-based early

blight and late blight disease detection on potato leaves.

In 2022 2nd International Conference on Technological

Advancements in Computational Sciences

(ICTACS) (pp. 923-928). IEEE.

Hylmi, M. S., & Suryani, E. (2022, September). Detection

of potato leaf disease using multi-class support vector

machine based on texture, color, and shape features.

In 2022 International Conference on Electrical and

Information Technology (IEIT) (pp. 20-24). IEEE.

Rusli, A. H. T., Meng, B. C. C., Damanhuri, N. S., Othman,

N. A., Othman, M. H., & Zaidi, W. F. A. W. (2022,

October). Potato leaf disease classification using image

processing and artificial neural network. In 2022 IEEE

12th International Conference on Control System,

Computing and Engineering (ICCSCE) (pp. 107-112).

IEEE.

Bonik, C. C., Akter, F., Rashid, M. H., & Sattar, A. (2023,

January). A convolutional neural network based potato

leaf diseases detection using sequential model. In 2023

International Conference for Advancement in

Technology (ICONAT) (pp. 1-6). IEEE.

Sheenam, S., Khattar, S., & Verma, T. (2023, June).

Automated Wheat Plant Disease Detection using Deep

Learning: A Multi-Class Classification Approach.

In 2023 3rd International Conference on Intelligent

Technologies (CONIT) (pp. 1-5). IEEE.

Dhall, D., Kaur, R., & Juneja, M. (2020). Machine learning:

a review of the algorithms and its

applications. Proceedings of ICRIC 2019: Recent

innovations in computing, 47-63.

Borawar, L., & Kaur, R. (2023, March). ResNet: Solving

vanishing gradient in deep networks. In Proceedings of

International Conference on Recent Trends in

Computing: ICRTC 2022 (pp. 235-247). Singapore:

Springer Nature Singapore.

Kaur, R., & Juneja, M. (2018). Comparison of different

renal imaging modalities: an overview. Progress in

Intelligent Computing Techniques: Theory, Practice,

and Applications: Proceedings of ICACNI 2016,

Volume 1, 47-57.

Kaur, R., & Juneja, M. (2016). A survey of different

imaging modalities for renal cancer. Indian Journal of

Science and Technology.

Kaur, R., Juneja, M., & Mandal, A. K. (2021). Machine

learning based quantitative texture analysis of CT

images for diagnosis of renal lesions. Biomedical

Signal Processing and Control, 64, 102311.

Convolutional Neural Networks-Based Potato Leaf Disease Classiﬁcation

497