Explainable Artificial Intelligence-Based Deep Learning for Detecting Cotton Leaf Diseases

Abstract

This study aims to address the increasing prevalence of diseases in cotton plants, particularly affecting their leaves, by developing an efficient computer-assisted diagnostic system. The primary objective is to use deep learning techniques to accurately diagnose cotton leaf diseases via image analysis and recommend appropriate treatments to minimize their impact on quality and productivity. The research utilized a dataset comprising images of cotton leaves categorized into four classes: Fusarium wilt, curl virus, bacterial blight, and healthy leaves. Six Convolutional Neural Network (CNN) architectures were employed for model development, including Xception, InceptionResNetV2, EfficientNetV2S, MobileNetV2, DenseNet169, and ResNet101V2. The models were trained and tested on the prepared dataset to assess their performance in multiclass classification. Each model processes input images of cotton leaves and extracts relevant features to classify the leaves into one of four categories, with accuracy used to evaluate the models' effectiveness. The proposed models achieved high accuracy rates on test data: 100% for InceptionResNetV2, EfficientNetV2S, and Resnet101v2; 99.49% for Xception; 78.39% for Mobilenetv2; and 69.34% for DenseNet169. The findings demonstrate the capacity of DL (deep learning) models for accurate and efficient disease detection in cotton leaves. Future work could focus on expanding the dataset by including additional disease types and exploring hybrid architectures or ensemble methods further to improve classification accuracy and generalizability across diverse agricultural conditions.