Cotton Cure : Smart Detection of Cotton Plant
Diseases
Suresh Reddy*,Vasudha J Rao*,Uzma Amber*,Hanirooth K B*
(Department of Computer Science and Engineering(Artificial Intelligence), Dayananda
Sagar Academy of technology and Management, Bangalore – 560028)
December 24,2024
Abstract
Cotton is a cornerstone of the global textile industry and an essential crop for farmers worldwide.
Unfortunately, diseases affecting cotton plants can severely impact their growth, leading to significant
economic losses. Detecting these diseases early is crucial, but traditional methods are often expensive,
slow, and rely on expert help, making them impractical for large farms.
This study introduces an easy-to-use system that helps farmers identify cotton diseases through photos of
their plants. By using advanced deep learning techniques, the system can quickly analyze leaf images and
detect common issues like bacterial blight and leaf spot. This helps farmers save both time and money,
while also allowing them to quickly address any issues and protect their crops before it’s too late.
Looking ahead, we aim to make this technology even more accessible by integrating it with smart
devices, enabling large-scale monitoring of fields. By bringing innovation to agriculture, we hope to
support farmers and boost cotton production for a better future.
Keywords: Cotton Diseases, Deep Learning, Agriculture, Disease Detection, Farmers, Innovation.
Introduction
India is well known as one of the earliest agricultural nations, but so many farmers
are still using unscientific methods of farming. These old methods often show low yields of
crops and negligible economic returns. There are quite a number of difficulties in growing crops,
though the first challenge is making the right choice of crops for planting. The emergence of
various crop-related diseases further complicates matters, leading to substantial losses in
agricultural productivity. Infections can devastate crops, hindering the production process and
raising concerns for farmers whose livelihoods depend on healthy plants.
There are new technologies that promise solutions for these problems. For example, image
processing techniques have gained increased usage in plant diseases caused by pathogens,
including bacteria, fungi, and microorganisms. Among such developments, the application of
CNNs is a vital stride towards the development of new efficient methods for identifying crop
diseases, which would call for early detection and management.
This paper discusses the efficiencies of CNN models and processing images in disease
recognition as well as classification, thus focusing on cotton leaf diseases such as Alternaria
Macrospora and Bacterial Blight. The proposed system is user-friendly, offering farmers the
opportunity to submit images of diseased leaves for analysis. The system processes the digital
images, extracting key features at each layer of the CNN to diagnose the health of the crops
accurately. At the end, this application not only identifies crop diseases but also gives farmers
corrective and preventive measures to reduce the effects of these diseases on their yield and
financial stability.
Related Work
The identification and classification of plant diseases have long been a focal point of agricultural
research, with numerous advancements aimed at improving the accuracy and efficiency of
disease detection. Early approaches included traditional image-processing techniques, but more
recently, machine learning and deep learning models have played a pivotal role in achieving
higher accuracy and scalability.
Siddharth Singh Chouhan et al. [1] introduced an innovative approach that combines Bacterial
Foraging Optimization (BFO) with a Radial Basis Function Neural Network (RBFNN) to
enhance plant leaf disease detection and classification. Their method uses a region-growing
algorithm to improve the feature extraction process, grouping similar seed points to optimize the
network design. This combination significantly improved detection efficiency, particularly in
complex plant disease scenarios.
Similarly, Muhammad Waseem Tahir et al. [2] developed a specialized dataset for fungal disease
detection using Convolutional Neural Networks (CNNs). Their custom CNN architecture
achieved an impressive 94.8% accuracy using a five-fold validation method, demonstrating the
potential of deep learning in identifying diverse fungal species on plant leaves. This approach
has become a significant benchmark in plant disease detection, emphasizing the power of CNNs
for high-accuracy results.
Building on this, Sukhvir Kaur et al. [3] proposed a semi-automated framework for diagnosing
diseases in soybean leaves. By integrating k-means clustering for image segmentation and
Support Vector Machine (SVM) classifiers for disease categorization, their system effectively
identified conditions such as downy mildew, frog-eye leaf spot, and sectorial leaf blight. Their
system achieved an accuracy of approximately 90%, showcasing the effectiveness of combining
clustering algorithms with traditional classifiers in plant disease diagnostics.
Ranjith et al. [4] took a more integrated approach by developing a smart irrigation system that
incorporated automated plant disease detection. Using images captured from plant leaves, the
system processed these images through a cloud-based server and compared them to a database of
known diseased leaf images. The system’s mobile application then provided users with disease
diagnoses, enabling informed and timely decisions about irrigation and treatment.
Another notable contribution came from Adhao Asmita Sarangdhar et al. [5], who developed an
SVM-based regression system designed specifically for cotton plant diseases. This system not
only identifies diseases like bacterial blight, Alternaria, gray mildew, Cercospora, and Fusarium
wilt but also offers pesticide recommendations. It integrates environmental monitoring, including
soil moisture, temperature, and humidity, and presents this information through a user-friendly
Android app. This system exemplifies the growing trend of combining disease detection with
practical agricultural tools, making it a valuable resource for farmers.
Our work builds upon these foundational studies, aiming to create a scalable and precise solution
for cotton disease identification. By leveraging advanced CNN models, we seek to improve the
accuracy of cotton disease classification, while addressing challenges such as dataset diversity
and usability. Our goal is to provide farmers with actionable insights that help them safeguard
their crops, increase productivity, and reduce the environmental impact of pesticide use.
Literature Survey
There are constantly evolving agricultural landscapes along with various innovative methods
surfacing to combat the devastating effects of plant diseases against crop yields. This paper
outlines several contemporary approaches focusing on the application of CNN within the
PyTorch framework toward the effective identification of plant diseases. One recent example is
the study presented at the International Conference on Internet of Things and Intelligence
Systems (IEEE, 2018), which successfully detected the fungal disease caused by fungi in
sugarcane using leaf area analysis, although this method has high costs when implemented and is
computationally complex.
Similarly, research by the International Conference for Convergence in Technology (IEEE, 2018)
detected diseases based on image processing techniques by taking images of the leaves and
matching them against a database. The method though effective only measures leaf area which is
prone to errors because of its lower parameters concerning disease detection, and even suggested
pesticide usage, thus posing a threat to soil health over time.
Other research has been done on different ways of detecting plant diseases. For instance, a 2016
paper on rice disease detection used Bhattacharya's Similarity Calculation method. This method
compared the diseased plants to a database of healthy images. However, this method could
identify some diseases but was not effective due to its training data being non-linearly separable.
Another method focused on tomatoes used a combination of thresholding algorithms and
K-means clustering for disease detection. Even though these methods enhanced their ability, they
could not distinguish the ripe from the unripe tomato.
Recent improvements in machine learning have further led to promising techniques for detecting
diseases in jute plants. Through image capturing and then enhancing the image qualities with
hue-based segmentation, systems could be designed to identify the specific disease on the stems.
Accurate plant classification remains important and is useful in several areas, such as enhanced
productivity and quality in products developed from agricultural activities.
Artificial intelligence (AI) has increasingly played a role in diagnosing plant illnesses. For
example, scientists successfully used AI to classify plant diseases and identify their defensive
features. The work on maize diseases was undertaken through a proposed strategy that yielded a
mean accuracy of 92% and thus holds a very good potential for crop preservation in developing
regions.
Detection and management of leaf diseases in countries like India become a great concern since
agriculture is the backbone of the economy. The disease can have a drastic impact on crop yields.
Diagnosis and treatment must, therefore, be timely for purposes of sustainable agriculture. A
trained system that recognizes plant diseases can empower farmers to make correct decisions,
reduce crop losses, and promote environmentally friendly farming practices.
Methodology
The proposed cotton disease detection system uses deep learning techniques to classify and
diagnose common diseases affecting cotton plants. First, the dataset preparation phase involves
curating and preprocessing a comprehensive collection of cotton plant images categorized into
Healthy, Bacterial Blight, Powdery Mildew, and Target Spot. The dataset is divided into training,
validation, and testing subsets to ensure effective training and model evaluation. The model that
has been pre-trained is EfficientNet-B0, which is used for transfer learning. In this case, the
classifier layer is replaced with a four-output class classifier. The model is trained on the training
dataset and validated on the validation dataset. Then, it is evaluated on unseen test data. The best
accuracy is saved, and the model is used for inference.
The system integrates the trained deep learning model into a Flask-based web application that
provides an easy-to-use interface for disease detection. Users can upload images or capture them
in real time using a camera. When an image is received, the backend processes it, runs inference
with the trained model, and returns the prediction results. To make the system more effective,
detailed information about the detected disease, such as its causes, symptoms, and modes of
prevention, is called from the Gemini API. The system's robustness is ensured through thorough
testing; the web application is released on a cloud platform in order to support accessibility as
well as scalability.
System Architecture:
The architecture of this system is divided into three important parts: the frontend part, the
backend part, and the deployment infrastructure. The frontend is developed using HTML, CSS,
and JavaScript so that the user can upload or capture images of cotton plants. The backend uses a
Flask framework that includes the trained EfficientNet-B0 model for real-time disease
prediction. The deep learning model, fine-tuned for cotton disease classification, processes
preprocessed images and gives out the predicted disease class along with its confidence score.
The Gemini API is also utilized to retrieve secondary information concerning the disease, such
as diagnosis and prevention methods.
The architecture is well-designed to allow for proper interaction between the components
involved. Images uploaded from the frontend are processed in the backend where they are
resized and normalized and passed through a deep learning model. When a prediction is needed,
then the additional information regarding the disease is retrieved from the API. This involves
hosting the web application on a cloud infrastructure such as AWS, Google Cloud, or Heroku,
allowing for scalability and accessibility. The architecture allows for easy future enhancements
like adding a database to store the user's data or even historical predictions.
Algorithm:
The core algorithm in the proposed system is such that it operates in step-by-step progression to
assure accurate disease prediction and an easy-to-interact model. It takes a picture of a cotton
plant as an input from either uploading from the user's side or capturing using the camera
attached. This image then gets preprocessed in terms of size, being resized to 224x224 pixels,
and normalization in accordance with some predefined mean and standard deviation values to
suit the model of EfficientNet-B0.
This preprocessed image is then passed into the trained model, and for each of the four classes, it
outputs logits. A Softmax function is applied to calculate the probabilities, and the class with the
highest probability is considered to be the predicted disease. There is a confidence threshold to
handle uncertain predictions; if the confidence score is below this threshold, the system classifies
the image as Unknown. For all diseases other than Healthy, the Gemini API is called to fetch
further information, such as symptoms, diagnosis, and preventive measures.
The system finally returns a structured answer to include the predicted disease and confidence
score along with some additional information. That would be shown on the user interface, giving
insights about action to be performed upon the result. The whole system will ensure real-time
processing with accurate disease detection so that it can stand confidently in assisting farmers or
agriculture experts.
Result
The system proposed, Cotton Cure, applies the CNN-based image processing model to classify
cotton diseases from plant and leaf images that the users upload. On uploading the image, the
system preprocessed it to standardize to 224 × 224 pixels so that the system was compatible with
the model. The system's accuracy was 98%, enabling it to give a user an accurate description of
the disease, suggested treatments, the available pesticides and their cost, and prevention methods.
The dataset consists of images from four categories: Bacterial Blight, Powdery Mildew, Target
Spot and Healthy plants. However, the model is currently optimized to classify four categories:
Bacterial Blight, Powdery Mildew, Target Spot, and Healthy. This categorization is done using a
pre-trained EfficientNet B0 architecture, fine-tuned for high accuracy and robust classification.
Cotton Cure is designed to work under a variety of conditions of images, such as intensity of
light, orientation, and resolution. This will make the prediction robust across all the scenarios.
The system also interfaces with the Gemini API, giving real-time access to the latest diagnostic
insights. This improves disease detection and provides complete management strategies for
farmers.
The Cotton Cure empowers timely and accurate disease diagnoses, supports sustainable
agricultural practices, helps to mitigate crop losses, and empowers farmers to take informed
action for improved cotton yield.
Conclusion
Here is the conclusion adapted to Cotton Cure :
It proved the potential utility of applying a CNN combined with PyTorch, in identifying Cotton
diseases namely Bacterial Blight, Powdery Mildew, and Target Spot by developing an effective,
strong system named as Cotton Cure to be employed for online applications in farm operations.
It helps the farmers understand what diseases it is and gives preventive, corrective steps to be
performed, recommendation of treatment applied, which pesticide should be used along with an
approximation of its cost. Despite challenges related to variability in image conditions, the
system achieves commendable accuracy, which highlights the critical importance of large,
diverse, and high-quality datasets. This scalable approach also opens doors to applications in
other crops, enabling multi-disease detection and adaptation to a variety of agricultural contexts.
The integration of the Gemini API further enhances the practical utility of the system by
allowing real-time access to the latest diagnostic insights. Future enhancements, such as
integrating IoT for real-time image capture and improving dataset diversity and attributes,
could significantly amplify its impact. Moreover, a collaborative platform for farmers to
share experiences and discuss disease trends can further showcase the transformative
potential of technology in modern agriculture.
This work not only contributes to reducing crop losses but also aligns with broader efforts
to leverage artificial intelligence for sustainable agriculture. By empowering farmers to
adopt data-driven, resilient agricultural practices, Cotton Cure is a promising tool for
enhancing
productivity and profitability. The convergence of deep learning, IoT, and collaborative
platforms underscores a future where technology empowers farmers to overcome challenges
and foster sustainable growth in agriculture.
References.
1 Navina Pandhare, Vrunali Panchal, Shivam S.Mishra, Mrs. Darshna Tambe, April
2022, “COTTON PLANT DISEASE DETECTION USING DEEP LEARNING” IRJMTS.
2 Agriplex India Technical Team Agro, management- of-majorcotton-diseases.
AgriplexIndia Technical Team Agro, June 30, 2022”.
3 Ms. Priya Ujawe, Dr. Smita Nirkhi, “Review on Different Types of Tomato Crop Disease
and Detection Using Deep Learning Technique”, 2022, International Journal of Engineering
and
Creative Science.
4 Sriramakavacham Ramacharan, 2021, “A Three-Stage Method for Diseases Detection
of Cotton Leaf using Deep Learning CNN Algorithm”, Research Gate.
5 Muhammad Suleman Memon, Pardeep Kumar, and Rizwan Iqbal, 2022, “Meta Deep
Learn Leaf Disease Identification Model for Cotton Crop”, MDPI
6 Al-bayati, J. S. H., & Üstündağ, B. B. (2020) “Evolutionary Feature Optimization for Plant
Leaf Disease Detection by Deep Neural Networks”, International Journal of
Computational Intelligence Systems. Computational Intelligence Systems.
7 Nikhil Shah, Sarika Jain, “Detection of disease in Cotton leaf using Artificial
Neural Network”, 2019 IEEE
8 A. Jenifa, R. Ramalakshmi, V. Ramachandran, “Classification of cotton leaf Disease
using Multi-support Vector Machine”, 2019 IEEE.
9 J. Ma, K. Du, F. Zheng, L. Zhang, Z. Gong, and Z. Sun, ‘‘A recognition method for
cucumber diseases using leaf symptom images based on deep convolution neural
network,” Computer Electron. Agriculture. Vol 154, pp. 18-24, Nov.2018.
10 S. Chouhan, A. Kaul, U. Singh, S. Jain,” Bacterial foraging optimization based Radial Basis
Function Neural Network (BRBFNN) for identification and classification of plant leaf
diseases: An automatic approach towards Plant Pathology”, 2017.
11 M. W. Tahir, N. A. Zaidi, A. A. Rao, R. Blank, M. J. Vellekoop and W. Lang"A
Fungus Spores Dataset and a Convolutional Neural Networks based Approach for Fungus
Detection",IEEE TRANSACTION ON NANOBIOSCIENCE, MAY 2018.
12 S. Kaur , S. Pandey, S. Goel"Semi-automatic leaf disease detection and classification
system for soybean culture",The Institution of Engineering and Technology 2018.
13 Ranjith, S. Anas, I. Badhusha, Zaheema OT, Faseela K, M. Shelly"Cloud Based Automated
Irrigation And Plant Leaf Disease Detection System Using An Android
Application",International Conference on Electronics, Communication and Aerospace
Technology ICECA 2017.
14 A. A. Sarangdhar, Prof. Dr. V. R. Pawar "Machine Learning Regression Technique for
Cotton Leaf Disease Detection and Controlling using IoT",International Conference on
Electronics, Communication and Aerospace Technology ICECA 2017.
15 Tejonidhi M.R, Nanjesh B.R, Jagadeesh Gujanuru Math, Ashwin GeetD'sa "Plant Disease
Analysis Using Histogram Matching Based on Bhattacharya's Distance Calculation"
International Conference on Electrical, Electronics and Optimization Techniques
(ICEEOT)-20164 Vector Machine”, 2019 IEEE
