COTTON LEAF DISEASE DETECTION USING DEEP LEARNING

Abstract

Cotton is the most important crop for making textiles around the world. To protect the fiber, food, and environment, farmers need to use sustainable farming methods. Cotton crops are very important to farming economies, but sometimes diseases can hurt production. Among the major contributors to the global agricultural revolution, the cotton industry is one of the most influential. Timely and accurate detection of crop conditions is essential in this field, and deep learning has proven to be a powerful tool in addressing this challenge. This study provides a clear and efficient framework of Deep Learning with respect to automated detection of Multi-Class Cotton Leaf Disease Classification in response to a major challenge that determines the yield of crops, farmer productivity and the sustainability of agriculture as a whole. This thesis focuses on four advanced CNN architectures, including DenseNet201, ResNet50, InceptionV3, and MobileNetV3 to a comprehensive design of evaluation based on a systematic strategy, comprising data preprocessing, data augmentation, stratified splitting, model training, and multi-metric performance evaluation, using the SAR-CLD-2024 dataset. The DenseNet201 model was the best in the assessed models with the highest validation of 99.78% and exhibited good learning throughout training. However, MobileNetV3 showed the highest overall test results, with the test accuracy of 99.85% and equally high precision, recall, and F1-score (all equal to 99.86%), which is why it is the most trustworthy and can be easily deployed in the real-world environment. InceptionV3 and ResNet50 generated less competitive yet similar-in-quality results. The results indicate clearly that deep learning, particularly Convolutional neural networks (CNNs) models, can be used to give effective and precise answers to the problem of plant leaf disease classification. Despite being limited by computational resources and diversity of datasets, the study provides a strong basis on which it can improve in the future by investigating emerging architectures, increasing datasets in different regions, and creating mobile or edge-based diagnostic tools Altogether, this research presents a tested and proven system and emphasizes the high potential of AI in enhancing agricultural disease detection.