A Classification and Detection of Cotton Leaf Disease Using

Lightweight CNN Architecture

Swetha G

, Veena V

, Kruthi T C

, Shwetha A B

, Selvarani S

and Jhansi K

Do CSE, R R Institute of Technology, Bangalore, Karnataka, India

Do CSE, Sapthagiri College of Engineering, Bangalore, Karnataka, India

Do ISE, R R Institute of Technology, Bangalore, Karnataka, India

Do EEE, R R Institute of Technology, Bangalore, Karnataka, India

Keywords: Lightweight Convolutional Neural Network (CNN), Depth Wise Separable Convolutions, Global Average

Pooling.

Abstract: The rapid and accurate detection of diseases in cotton leaves is crucial for ensuring high agricultural

productivity and minimizing economic losses. Traditional methods of disease detection, relying heavily on

manual inspection, are time-consuming and prone to errors. To address these challenges, we propose a

lightweight Convolutional Neural Network (CNN) architecture designed specifically for the classification and

detection of cotton leaf diseases. Our model focuses on optimizing computational efficiency and accuracy,

making it suitable for deployment on edge devices with limited processing power. The proposed CNN

architecture employs a series of convolutional layers with reduced parameters, leveraging techniques such as

depth-wise separable convolutions and global average pooling to maintain high performance while

minimizing computational costs. We trained and validated our model on a comprehensive dataset comprising

various cotton leaf diseases, including bacterial blight, leaf spot, and mildew. Experimental results

demonstrate that our lightweight CNN achieves a high classification accuracy, outperforming several existing

models in terms of both speed and precision. Furthermore, our model's ability to generalize across different

disease types highlights its potential for real-world applications. By enabling early and accurate detection of

cotton leaf diseases, our approach promises to significantly enhance crop management practices, reduce the

reliance on chemical treatments, and contribute to sustainable agricultural practices.

1 INTRODUCTION

Cotton is a critical cash crop and a staple in the textile

industry worldwide. However, cotton cultivation is

often threatened by various leaf diseases, which can

significantly impact crop yield and quality. Timely

detection and accurate classification of these diseases

are essential for effective disease management and

prevention. Traditional methods of disease detection

rely heavily on visual inspection by experts, which is

labour-intensive, time-consuming, and prone to

human error (Chatfield, Simonyan et al. 2014).

Recent advancements in computer vision and

machine learning offer promising solutions to

automate this process. Among these, CNNs have

demonstrated remarkable performance in image

classification tasks, including disease detection in

crops.

This study focuses on employing a lightweight

CNN architecture for the classification and detection

of cotton leaf diseases. The term "lightweight" refers

to neural networks designed to have fewer parameters

and computational requirements, making them

suitable for deployment on devices with limited

resources, such as

smartphones or edge devices. This approach not

only aims to provide accurate disease classification

but also to ensure efficiency and practicality in real-

world applications (Tijare, Khade, et al. 2019). The

lightweight CNN architecture proposed in this study

is designed to balance performance and

computational efficiency, enabling real-time analysis

of cotton leaf images. By leveraging transfer learning

and optimization techniques, the model achieves high

120

G, S., V, V., T C, K., A B, S., S, S. and K, J.

A Classiﬁcation and Detection of Cotton Leaf Disease Using Lightweight CNN Architecture.

DOI: 10.5220/0013609700004664

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

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 3, pages 120-125

ISBN: 978-989-758-763-4

classification accuracy while maintaining a low

computational footprint (Zhang, Shang, 2019). This

enables rapid and scalable deployment in agricultural

settings, potentially transforming the way farmers

and agricultural professionals manage cotton crops.

In summary, this research aims to enhance the

precision and accessibility of cotton leaf disease

detection through the development of a lightweight

CNN model, offering a practical solution for disease

management in cotton cultivation.

2 LITERATURE SURVEY

In their paper titled (Barbedo, and Arnal, 2019), the

authors delve into the realm of cotton leaf disease

detection through innovative image processing

techniques. This research is poised to offer insights

into the application of advanced technologies for

identifying and analyzing diseases affecting cotton

plants. Leveraging the capabilities of image

processing, the study likely explores methodologies

for efficient disease detection and analysis, aiming to

contribute valuable knowledge to the intersection of

agriculture and technology. In the paper (Bhong,

Vijay, et al. 2018), the authors introduce an

innovative approach to cotton leaf disease detection.

The methodology centres around a super pixel-based

roughness measure, showcasing a nuanced and

sophisticated method for capturing intricate details

associated with diseases affecting cotton plants. By

leveraging this novel approach, the research aims to

enhance disease detection and classification accuracy

as well as efficiency, showcasing the potential for

advanced information technology applications in the

agricultural domain. In the article authored by Jayme

Garcia Arnal and Barbedo, titled (Barbedo, and

Arnal, 2019) "A new automatic method for disease

symptom segmentation in digital photographs of

plant leaves," the author introduces an innovative and

automated approach to segmenting disease symptoms

in plant leaves from digital photographs. This method

is designed to streamline and enhance the process of

identifying and isolating specific symptoms

associated with plant diseases, contributing to the

field of plant pathology (Swetha, et al. 2021). The

research likely delves into sophisticated image

processing techniques to automate the segmentation

process, potentially providing a more efficient and

accurate means for researchers and practitioners to

analyze and understand the manifestation of diseases

in plant leaves.

In their collaborative work, (Schmidhuber, 2019)

The authors undertake a detailed investigation,

emphasizing the importance of addressing subtle yet

crucial aspects in the design and implementation of

CNNs. This work likely delves into specific

architectural choices, optimization techniques, or

challenges encountered during deep neural network

training, shedding light on nuanced factors that

substantially influence these models’ performance.

Current article (Khan, et al. 2018), the authors

address the critical task of plant disease recognition

using leaf images. The study likely explores

methodologies and techniques for leveraging image

processing along with machine learning for disease

identification & classification of affected plant

leaves. By focusing on plant leaf images as a key

input, research contributes to field of agricultural

science & technology, aiming to enhance early

disease detection and ultimately improve crop

management. In the collaborative work (Lumb, Sethi,

et al. 2017), the authors conduct a comprehensive

survey on methods and technologies employed in the

identification of diseases affecting cotton leaves. In

the research paper (Islam, Talukder, et al. 2023), the

authors explore innovative methods for texture

feature extraction from various image

representations. The study likely investigates the use

of diverse color spaces, including RGB, HSV, and

YIQ, along with dithered images, employing

techniques such as Wavelet Decomposition &

GLCM(Grey Level Co-occurrence Matrix). In

(Russakovsky, Deng, et al. 2020) author offers

comprehensive survey along with an analysis of

fundamental principles and advancements in deep

learning(DL) field. The work is likely to offer insights

into the historical development, key concepts, and

theoretical underpinnings of deep neural networks.

Given the title's emphasis on an overview, the paper

may serve as an introductory guide for researchers,

practitioners, and enthusiasts seeking a broad

understanding of the principles that form DL basis in

neural networks. In the paper (Patki, Sable, et al.

2018) authors contribute to field of computer vision

by presenting ImageNet Large Scale Visual

Recognition Challenge(ILSVRC). This influential

challenge, documented in the paper, has performed

vital function in advancing algorithm development

and evaluation for large-scale image classification

tasks.

3 METHODOLOGY

Proposed system processes foliage diseases’ input

images, expressed as three-dimensional arrays. These

images undergo several steps to ensure accurate

detection and classification:

A Classiﬁcation and Detection of Cotton Leaf Disease Using Lightweight CNN Architecture

121

Data Pre-processing: Photographs are rescaled

and labelled during data pre-processing phase. These

prepared images subsequently inputted into model.

Model Training: CNN is trained utilizing cotton

leaves images. The training process involves multiple

iterations where the model learns to identify patterns

associated with various foliage diseases.

Evaluation and Adjustment: After training, the

model's performance is evaluated for accuracy.

Parameters are adjusted based on evaluation results,

and the training process is repeated until satisfactory

accuracy is achieved.

Deployment: Once training & evaluation are

complete, model is deployed for use. The trained

classifier labels new, unseen images of leaves,

identifying and categorizing them based on learned

patterns.

By accurately identifying and mapping disease-

affected areas, farmers can develop targeted

treatment plans. This supports the concept of

precision agriculture, optimizing the use of resources

such as pesticides and fertilizers.

4 SYSTEM REQUIREMENTS

Cotton leaf disease recognition, utilizing a

lightweight CNN structure, represents a sophisticated

yet resource-efficient approach. This system

optimizing image acquisition and processing,

leveraging the efficiency of a streamlined CNN for

feature extraction and disease classification. A

lightweight design allowing swift real-time detection,

ensuring timely intervention in the field. Despite its

efficiency, the system maintained high accuracy and

precision, critical for reliable disease identification.

The user interface tailored for simplicity, facilitating

easy use, specifically in mobile applications for on-

the-field diagnoses. Architecture's scalability

enhances, enabling seamless integration with

diversified devices and accommodating a growing

user base. Moreover, the system adheres to strict

security mechanisms, ensuring protection for

sensitive agricultural data.

Functional requirements of the system defining

the behaviour, responses, and functionality of the

system, while the non-functional requirements focus

on features like reliability and user-friendly design

(Rothe, Kshirsagar et al. 2019).A feasibility study

conducted under the non-functional requirements.

The system requires a laptop with an Intel i3 or above

processor with satisfactory storage and RAM.

Software needs include Windows 10, Python for

coding, and libraries like Kera’s and TensorFlow.

4.1 System Architecture

System architecture defines the larger overall

structure that covers both software hardware. This

illustrating diagram demonstrating the components

and their interactions. The entire architecture

describing how input, processing data, and displaying

the output

Figure 1. System Architecture

5 IMPLEMENTATION

To detect cotton leaf diseases applying CNN, follow

these steps:

Collect Data: Gather many images of healthy

cotton leaves and leaves with diseases like blight or

curl.

Prepare Images: Resize and adjust the colors of

these images to ensure consistency.

Build CNN: Create a CNN model with layers that

can recognize patterns in images, differentiating

between healthy and diseased leaves.

Train the Model: Show the CNN the prepared

images, labelling them as healthy or diseased so it can

learn to recognize these patterns.

Evaluate Performance: Test the CNN on new,

unseen images to check its accuracy.

Deploy: Use the trained CNN to analyze new

images of cotton leaves, identifying if they are

healthy or diseased, allowing for early intervention.

Data Collection: Gather images dataset

containing healthy cotton leaves & leaves infected

with various diseases such as bacterial blight, leaf

curl, etc. You can find datasets online or collect and

label your own.

Data Pre-processing: Pre-process images by

resizing them to fixed size, normalizing pixel values,

INCOFT 2025 - International Conference on Futuristic Technology

122

along with augmenting data (e.g., rotating, flipping,

zooming) for increasing dataset diversity &

modifying model's generalization.

Model Architecture: CNN architecture designed

for image classification. You can start with a simple

architecture like LeNet or go for more complex

architectures like VGG, ResNet, or DenseNet

(Sarangdhar, Pawar, et al. 2024), depending on the

size and complexity of your dataset.

Model Training: Divide your dataset into

training, validation, & test sets. Train your CNN

model on the training set using techniques like

RMSprop, stochastic gradient descent (SGD), or

Adam, & tune hyperparameters such as batch size,

learning rate, along with number of epochs

employing validation set.

Finally, with a trained and validated model in

hand, we can deploy it for practical use in detecting

cotton leaf diseases. This could involve integrating

the model into a software application or system that

can analyze images of cotton leaves and provide

insights or recommendations based on the detected

disease conditions. Ongoing monitoring and

refinement may be necessary to ensure the model

remains effective as new data becomes available or as

environmental conditions change.

6 RESULTS

The results chapter outlines the outcomes of the

project, which focused on cotton leaf disease

classification & detection employing lightweight

CNN architecture.

These results highlight effectiveness of using a

lightweight CNN architecture for disease detection in

cotton crops, showcasing its potential to advance

agricultural practices and support sustainable cotton

farming.

Figure 2: Login Image and Register Image

The login page for Cotton Leaf Disease Detection

System serves as the secure entry point for users,

including farmers, agronomists, and researchers, to

access the system's functionalities. Designed with

user-friendliness and security in mind, the page

features fields for entering username or email &

password, along with "Login" button for submitting

these credentials for authentication. To accommodate

new users, a sign-up link directs them to a registration

page, while a "Forgot Password" link assists users in

recovering their account access. Security is bolstered

through HTTPS encryption, CAPTCHA verification,

and an account lockout mechanism after multiple

failed login attempts.

The "Register Image" page for the Cotton Leaf

Disease Detection System employs CNN architecture

designed to facilitate the seamless upload of cotton

leaf images for disease analysis. This page provides

an intuitive interface where users, including farmers

and agronomists, can easily submit images for

processing.

Figure 3: Disease classification.and uploading image

The "Cotton Leaf Detecting Image" page of the

Cotton Leaf Disease Detection System provides users

with an efficient and user-friendly interface to upload

images of cotton leaves for disease diagnosis. Upon

accessing this page, users can easily select images

from their devices or use drag-and-drop functionality

for quick uploads. The system guides users through

the upload process with clear instructions and real-

time feedback, ensuring images are correctly

formatted and of adequate quality for analysis.

The "Cotton Leaf Uploading Image" feature

within the Cotton Leaf Disease Detection System

streamlines the process of submitting images for

analysis, offering users a straightforward and

intuitive interface. Upon accessing this functionality,

users are prompted to upload images of cotton leaves

from their devices, supported by drag-and-drop

functionality or traditional file selection methods.

The system ensures ease of use by providing clear

instructions and visual cues throughout the upload

process. This step ensures that images are

appropriately formatted and prepared for accurate

disease detection. By simplifying the image upload

process and incorporating pre-processing

A Classiﬁcation and Detection of Cotton Leaf Disease Using Lightweight CNN Architecture

123

functionalities, the system enhances user experience

and empowers users to efficiently contribute to the

disease detection process, ultimately fostering more

effective management strategies for cotton crop

health.

6.1 Cotton Leaf Variant Identified as

Below

Figure 4: Healthy Cotton and Army Worm

Healthy cotton leaves are vital indicators of a

well-maintained and productive cotton crop. These

leaves are essential for photosynthesis, providing the

energy needed for plant growth and cotton fibre

production. Healthy cotton leaves are typically a rich,

vibrant green, indicating adequate chlorophyll and

proper nutrient levels.

Armyworm disease, caused by the larvae of

various moth species, is a major global threat to

agriculture. The fall armyworm (Spodoptera

frugiperda) is particularly notorious for its voracious

feeding, causing extensive damage to crops such as

maize, rice, cotton, and sugarcane.

Figure 5: Bacterial Blight and Powdery Mildew.

Destructive disease in plants caused by various

bacterial pathogens, most notably Xanthomonas spp,

is termed a Bacterial blight. It affects a wide range of

crops, including rice, cotton, and soybeans, leading to

significant yield losses and economic impact.

A prevalent fungal disease affecting variety of

plants, encompassing cereals, vegetables, &

ornamentals termed as Powdery mildew. It is caused

by different species of fungi in the order Erysiphales,

with each species typically specific to a particular

host.

Figure 6: Target Spot and Aphids

Target spot disease is a fungal infection caused

primarily by Corynespora cassiicola, affecting a

variety of crops, including cotton, soybean, tomato,

and cucumber. The disease is named for the

characteristic concentric rings or "targets" that appear

on infected plant tissues.

Aphid infestation is frequently impacting wide

range of crops, consisting cotton, cereals, vegetables,

& fruit trees. Aphids are small, soft-bodied insects

feeding on plant sap, causing direct damage &

potentially transmitting plant viruses.

Figure 7: Cotton leaf disease classification

The results encapsulate the outcomes and

achievements derived from the implementation of the

project. Through a comprehensive analysis of user

engagement, transaction processes, transparency

measures, resource management practices, and data-

driven decision-making, the project has demonstrated

significant advancements in the agricultural domain.

Key highlights include enhanced user engagement

facilitated by a user-friendly interface, streamlined

transactions leading to improved efficiency,

heightened transparency through blockchain

integration, optimized resource management

practices, and data-driven insights empowering

informed decision-making.

INCOFT 2025 - International Conference on Futuristic Technology

124

7 CONCLUSIONS

Our study demonstrates the efficacy of Convolutional

Neural Networks (CNNs) in detecting cotton leaf

diseases with remarkable accuracy and reliability. By

leveraging techniques such as data augmentation and

transfer learning, we achieved commendable results

in distinguishing between healthy and diseased cotton

leaves. The use of pre-trained CNN architectures

facilitated rapid convergence during training and

enabled the extraction of crucial features necessary

for accurate disease identification.

Despite these promising results, several

challenges persist. Notably, there is a need for larger

annotated datasets and the exploration of advanced

imaging techniques additional improvement in

model's performance. However, the implications of

our research are significant, providing farmers with

the tools for early and precise disease diagnosis. This

capability can lead to optimized crop management

practices and reduced yield losses.

Looking ahead, further efforts are essential in

expanding datasets, refining algorithms, and

validating the technology in real-world agricultural

settings. Such advancements will be pivotal in

transitioning this technology from research to

practical, widespread use in agricultural systems. Our

study underscores the effectiveness of CNNs in

automated cotton leaf disease detection and

highlights the importance of continued research to

enhance and implement these solutions in farming

practices.

REFERENCES

Barbedo, Jayme Garcia Arnal, A new automatic method for

disease symptom segmentation in digital photographs

of plant leaves, European Journal of Plant Pathology,

2019, 1-16.

Bhong, Vijay S and Pawar B.V, Study and Analysis of

Cotton Leaf Disease Detection Using Image

Processing, International Journal of Advanced

Research in Science, Engineering and Technology, 3

(2),2018.

Dubey YK., et al. “Super pixel-based roughness measure

for cotton leaf diseases detection and classification”.

2018 4th International Conference on Recent Advances

in Information Technology (RAIT). IEEE (2018): 1-5.

J. Schmidhuber, “Deep Learning in neural networks: an

overview,” Neural Networks, vol. 61, pp. 85–117,

2019.

K. Chatfield, K. Simonyan, A. Vedaldi, and A. Zisserman,

“Return of the devil in the details: delving deep into

convolutional nets

Khan RA., et al. “Prevalence of the major pests in

transgenic and non-transgenic cotton genotype in

response to climatic factors and their impact on yield”.

Archives of Phytopathology and Plant Protection51.17-

18 (2018): 927-938.

Lumb, Manisha, and Poonam Sethi, Texture Feature

Extraction of RGB, HSV, YIQ and Dithered Images

using GLCM, Wavelet Decomposition Techniques,

International Journal of Computer Applications, 68

(11), 2017.

Md. Manowarul Islam, Md. Alamin Talukder, Md. Ruhul

Amin Sarker, Md Ashraf Uddin, Arnisha Akhter, Selina

Sharmin, Md. Selim Al Mamun, Sumon Kumar

Debnath, A deep learning model for cotton disease

prediction using fine-tuning with smart web application

in agriculture, Intelligent Systems with

Applications,Volume 20,2023,200278,ISSN 2667-

3053,

O. Russakovsky, J. Deng, H. Su et al., “ImageNet large

scale visual recognition challenge,” International

Journal of Computer Vision, vol.115, no. 3, pp. 211–

252, 2020.

Patki SS and Sable GS. “Cotton leaf disease detection &

classification using multi SVM”. International

Journal of Advanced Research in Computer and

Communication Engineering 5.10 (2018): 165-

168.

Reena Tijare, Pawan Khade, Rashmi Jain, The Survey of

Disease Identification of Cotton Leaf, International

Journal of Innovative Research in Computer and

Communication Engineering, 2019

Rothe PR and Kshirsagar RV. “Cotton leaf disease

identification using pattern recognition techniques”.

2020 International conference on pervasive computing

(ICPC). IEEE (2019): 1-6.

Sarangdhar AA and Pawar VR. “Machine learning

regression technique for cotton leaf disease detection

and controlling using IoT”.2017 International

conference of Electronics, Communication and

Aerospace Technology (ICECA). IEEE 2 (2017): 449-

454

Zhang S. W, Shang Y.J, and Wang L, Plant Disease

Recognition Based on Plant Leaf Image, Journal of

Animal and Plant Sciences, 25 (1),2019, 42-45.

G. Swetha, R. et al., "Smart Survey on Power Quality

Improvement based on Distribution Network

Reconstruction and Reconfiguration," 2021 5th

International Conference on Electrical, Electronics,

Communication, Computer Technologies and

Optimization Techniques (ICEECCOT), Mysuru,

India, 2021, pp. 253-261

A Classiﬁcation and Detection of Cotton Leaf Disease Using Lightweight CNN Architecture

125