AI-Based Crop Monitoring System: Cotton Disease Detection Using CNN

Abstract

Agricultural production depends on the diseases that affect plants and pests; early recognition is essential to crop management. Traditional visual analysis is labor-intensive and subjective because it might depend on the expertise of an analyst; this paper presents an AI-driven cotton disease identification system using Convolutional Neural Networks (CNNs). This model is supported by a web dashboard and a mobile app. The model was trained using 9,495 images of cotton leaves from nine categories: Alternaria Leaf Spot, Aphids, Army Worm, Bacterial Blight, Fusarium Wilt, Healthy, Powdery Mildew, Target Spot, and Verticillium Wilt. All images were, in other words, scaled to 224 × 224 pixels and normalized to zero-to-one in the real-life scenario. Augmentation such as rotation, shifting, zooming, horizontal flipping, and shearing is implemented during the training process. What I mean to say is that the dataset was split into 7,594 training images, 947 validation images, and 954 test images. Custom CNN reached the accuracy of 95.18%, with the macro F1-score of 93.49% and the weighted F1-score of 95.15%.Flask-based backend supports the application, while the web dashboard provides the information and the model's analytics. Further evaluation using images collected under the mixed real-field conditions is needed to establish concrete reliability and generalization.

Introduction

Cotton is mainly one of the most significant commercial crops and weighs heavily in the agricultural economy, yet cotton plants are typically vulnerable to different diseases and pest-related conditions that can negatively alter plant health and crop productivity, in many respects; indeed, early identification of these conditions is vital for supporting timely monitoring and suitable agricultural management, in actual application. More often than not, plant diseases are, more accurately, spotted through the visual inspection by farmers or agricultural experts, and while, this method is widely employed, it can be time-consuming and may rest heavily on the observer's experience, particularly when multiple disease symptoms appear comparable, so an automated and consistent strategy to plant disease identification is highly desirable.

Recent advances in AI technology, especially Deep Learning, have provided new possibilities for automated plant disease diagnosis, to an extent. Convolutional neural networks (CNNs) have demonstrated the potential of classifying images through automatic learning of relevant features including edges, textures, patterns and colour variations from input images. In particular, unlike classical image processing techniques that generally involve hand-crafted features, CNN learns layered representation from images automatically. Deep learning algorithms have shown their potential in detecting and classifying plant diseases based on leaf images, as illustrated in some earlier studies [2], [3], [12], [13]. The advancement in the development of deep CNN models has also improved recognition accuracy on image data [11].

Nevertheless, creating such a disease detection system entails more than reaching high classification accuracy; on the contrary, the correct approach requires the development of an easily accessible interface where people can upload images of leaves and get predictions and information about potential diseases in general. There has been quite some research conducted in agricultural monitoring which is concerned with using computer vision, mobile applications, and drones in order to help recognize the state of crops [5], [6].

To deal with the above problem, This paper discusses an Artificial Intelligence-based Cotton Disease Detection System Using Convolutional Neural Network (CNN). In this connection, the proposed system makes use of a tailored Sequential CNN model that has classified the cotton leaf images into nine different classes including Alternaria Leaf Spot, Aphids, Army Worm, Bacterial Blight, Fusarium Wilt, Healthy, Powdery Mildew, Target Spot, and Verticillium Wilt, etc.; the dataset has 9,495 images and contains training, validation, and testing portions. Image preprocessing, normalization, and data augmentation are used to preprocess the dataset for the training process; the trained model achieves a test accuracy of 95.18%.

In addition, for the classification model, the project incorporates a backend built using Flask, and the web dashboard designed using React, besides a mobile application interface; all of this setup is aimed at the moment towards making the process of disease classification easier through integration of predictions based on images with an application interface.

The primary insight of this work is the development of a cotton leaf disease classification approach employing a custom CNN architecture, combined with an application-oriented monitoring framework, broadly speaking; the suggested method, system architecture, experimental results, and conclusion are being outlined by the following sections, to a significant degree.

1. Related Work
2. CNN-Based Plant Disease Detection

Tejaswini et al. [1] studied CNN-based plant disease detection and reported an accuracy of 86.21%, on the whole; what their study showed was the potential of convolutional neural networks to spot disease-related visual patterns from plant images, broadly speaking. Sutaji and Rosyid [2] evaluated pretrained CNN architectures for crop disease classification, reporting 98.97curacy using DenseNet121. Their findings pointed out the significance of taking into consideration the trade-off between accuracy and computational needs.

B. Deep Learning Approaches for Crop Disease Classification

Shoaib et al. [3] have proposed novel deep learning methodologies for plant diseases as well as problems related to availability of labelled data, computational performance and generalization capability to some extent. Similarly, Salka et al. [4] have reviewed the classification of plant leaf diseases using CNNs and highlighted the benefits of feature extraction, as well as the problems encountered in the field environment; this research has been helpful in understanding the merits and demerits of deep learning methodologies for plant disease detection to some extent.

C. Drone-Assisted Agricultural Monitoring

The methodology that incorporated drone-based agricultural surveillance with CNN-based image recognition techniques along with application interfaces was put forward by Shah et al. [5]. 98.80curacy level was achieved with their EfficientNet-B3 model in the experimental setup described; yang et al. [6] came up with an intelligent crop health surveillance mechanism which made use of drone remote sensing along with CNN-based image recognition system. The modified MobileNetV2 model of theirs achieved 97.14curacy with an estimated size of 2.01 MB; these examples practically demonstrate the scope of integration between aerial imaging and AI-based crop monitoring systems.

D. Advanced Grouping and Object Detection Techniques

When it comes to the practical application in disease detection in tomatoes, Sowmya and Guruprasad [7] have employed Inception-v4 for category detection while utilizing YOLOv8 for object detection. They obtained an accuracy rate of 96% when it came to categorizing disease symptoms, whereas the mAP@0.5 value was 86%. On the other hand, Manoj et al. [8] have studied the implementation of drones for disease detection in crops.

E. Explainable AI and Attention-Based Disease Detection

Chihaoui et al. [9] provided a review on the use of AI for crop disease detection via leaf images and explainable AI. Some of the drawbacks identified by the authors include biased datasets, comments, cross-domain validation, and computational constraints. Gani et al. [10] performed research on the use of attention-based CNN with explainable AI for papaya leaf diseases.

F. Foundational Research on CNN and Plant Disease Classification

The efficiency of using deep convolutional neural networks in large scale image classification was proved by Krizhevsky et al. [11]. The application of deep learning in image based plant disease detection was studied by Mohanty et al. [12]. In reality, deep learning models for detecting and diagnosing plant diseases were further experimented with by ferentinos [13].

Fig.1. Category of Existing Research Areas

G. Research Gap and Motivation for the Proposed System

In this sense, it can be seen from the studies discussed above that CNNs and pre-trained models have the ability to reach very good classification results on different plant-disease datasets; however, important problems still exist in terms of field testing, computational efficiency, and disease-classification models’ integration with monitoring systems. Still, performance depends on the dataset used, crops, and classes as well as evaluation methods.

Fig.2. Image Preprocessing Pipeline

Motivated by these problems, the present work is constructing a custom CNN model for classifying nine cotton leaf categories and connects it with a Flask-based backend, a React web dashboard, and a mobile application interface, in actual application, and the model hit 95.18curacy on its held-out cotton test dataset. The proposed framework focuses on linking disease classification using images with the digital interface. Going a step further validation employing images collected under the real agricultural conditions is needed to establish its field-level reliability, in actual application.

Conclusion

The contribution of this paper was an AI-based monitoring system of cotton crops for automatic classification of diseases of the leaves of cotton plant with a specialized CNN architecture. This model was developed to categorize images of cotton leaves into nine classes, comprising eight classes related to diseases and one class of healthy plants. The dataset comprised of 9,495 images that were pre-processed, augmented, and partitioned into training, validation, and test subsets, to be more specific. The use of convolutional layers, batch normalization, pooling and dropout helped formulate a classification model capable of learning relevant visual features from leaf images, on the whole. Proposed CNN architecture reached 95.18curacy in testing with macro averaged precision being 93.94%, recall 93.25% and F1-score of 93.49%. The above mentioned results illustrate the potential of proposed model in identifying various cotton leaf states in the tested data set. In addition to modeling part of the project, it includes Flask API server with a user-oriented web dashboard developed with React which represents a suggestion for an approach to access disease classification results. In addition to that, a mobile application interface was designed in the proposed system as an example of the potential for integration of deep learning image classification results with convenient software interfaces in agriculture, in practice. Before, the experimental results are promising, on the whole. More often than not, a foundation for AI-assisted cotton leaf disease classification is being started by the project, in actual application. More accurately, going a step further validation under the real agricultural field conditions is needed before the framework can be considered trustworthy for employed large-scale deployment. Broadly speaking. Several improvements may be considered to further enhance the accuracy, robustness, scalability, and practicality of the proposed system. Real-World Image Dataset Enlargement: The future direction should aim at obtaining a larger set of cotton leaf images in real-world scenarios. Images captured under different lighting, background, development stage, and environmental variations can improve the model’s generalization capability in the practical deployment process. Acceleration of the Model: State-of-the-art Convolutional Neural Network architectures. On the other hand, transfer learning methodologies can also be explored in depth to increase classification efficiency and reduce computational requirements. Indeed, model pruning and quantization can, for sure, also be tested to ensure deployment in resource-restricted settings. Explainable AI Integration: Explaining AI methods like Grad-CAM may be integrated for highlighting the areas of the images which contribute to the decision-making of the model. This means that to some degree, transparency may be increased due to which the classification process becomes easy for the user to understand. Mobile App Development: The mobile app may be further developed to enable the capture of images, their classification, and to provide the results in an easy manner. It is important to perform integration and testing under varying network conditions. Image Acquisition from Drones: In future versions, drone-based image acquisition may be considered for capturing images of crops over large agricultural areas.

References

T. Tejaswini et al., “Early Disease Detection in Plants using CNN,” Procedia Computer Science, 2024. DOI: 10.1016/j.procs.2024.04.327. D. Sutaji and H. Rosyid, “Convolutional Neural Network (CNN) Models for Crop Diseases Classification,” KINETIK, vol. 7, no. 2, 2022. DOI: 10.22219/kinetik.v7i2.1443. M. Shoaib et al., “An Advanced Deep Learning Models-Based Plant Disease Detection: A Review of Recent Research,” Frontiers in Plant Science, vol. 14, 2023. DOI: 10.3389/fpls.2023.1158933 T. D. Salka et al., “Plant Leaf Disease Detection and Classification Using Convolution Neural Networks Model: A Review,” Artificial Intelligence Review, 2025. DOI: 10.1007/s10462-025-11234-6. S. A. Shah et al., “Application of Drone Surveillance for Advance Agriculture Monitoring by Android Application Using Convolution Neural Network,” Agronomy, vol. 13, article 1764, 2023. DOI: 10.3390/agronomy13071764. H. Yang et al., "Construction of an Intelligent Analysis System for Crop Health Status Based on Drone Remote Sensing Data and CNN," IEEE Access, 2025. DOI: 10.1109/ACCESS.2025.3540279. B. Sowmya and S. Guruprasad, “Deep learning based plant health disease detection in tomatoes using inception v4 convolutional neural network and YOLO V8,” Discover Artificial Intelligence, 2025. DOI: 10.1007/s44163-025-00540-1. H. M. Manoj et al., "AI-driven drone technology and computer vision for early detection of crop disease in large agricultural areas," Scientific Reports, 2026. DOI: 10.1038/s41598-025-32384-1. N. Chihaoui et al., “A comprehensive review on AI-based crop disease detection using leaf image classification and explainable AI,” Discover Applied Sciences, vol. 8, article 663, 2026. DOI: 10.1007/s42452-026-08684-0. R. Gani et al., "Attention guided convolutional neural network with explainable AI for papaya leaf disease detection in edge and drone agricultural systems," Scientific Reports, 2025. DOI: 10.1038/s41598-025-25374-w. A. Krizhevsky, I. Sutskever and G. E. Hinton, “ImageNet Classification with Deep Convolutional Neural Networks,” Advances in Neural Information Processing Systems, vol. 25, 2012. S. P. Mohanty, D. P. Hughes and M. Salathé, “Using Deep Learning for Image-Based Plant Disease Detection,” Frontiers in Plant Science, vol. 7, article 1419, 2016. DOI: 10.3389/fpls.2016.01419. [13] K. P. Ferentinos, “Deep learning models for plant disease detection and diagnosis,” Computers and Electronics in Agriculture, vol. 145, pp. 311-318, 2018. DOI: 10.1016/j.compag.2018.01.009. J. G. A. Barbedo, “Factors influencing the use of deep learning for plant disease recognition,” Biosystems Engineering, vol. 172, pp. 84-91, 2018. DOI: 10.1016/j.biosystemseng.2018.05.013. D. P. Hughes and M. Salathé, “An open access repository of images on plant health to enable the development of mobile disease diagnostics,” arXiv:1511.08060, 2015. TensorFlow, “TensorFlow Documentation.” Available: https://www.tensorflow.org/ Keras, “Keras Documentation.” Available: https://keras.io/ Pallets Projects, “Flask Documentation.” Available: https://flask.palletsprojects.com/ React, “React Documentation.” Available: https://react.dev/ Google, “Google Antigravity Documentation.” Available: https://antigravity.google/docs/ide/overview/