A Soft Computing Framework for Automatic Cotton Leaf Disease Detection via Convolutional Neural Networks

Cotton foliar diseases and pest damage reduce yield and fiber quality, while manual diagnosis is slow and dependent on specialist experience. This paper presents an integrated soft-computing framework for automatic identification, interpretation, and prognosis of cotton leaf conditions. The system classifies images into six categories—healthy leaf, aphids, armyworm, bacterial blight, powdery mildew, and target spot—using a convolutional neural network trained with resized, normalized, contrast-enhanced, and augmented images. A complementary image-segmentation module estimates the infected-area percentage and maps it to severity levels for decision support. The MATLAB implementation combines the Deep Learning Toolbox, Image Processing Toolbox, and a graphical user interface for model training, image upload, prediction, severity visualization, and report generation. The reported confusion matrix contains 1,200 test decisions and yields 98.83% overall accuracy, 98.84% macro-precision, 98.83% macro-recall, and 98.83% macro-F1. The integrated design extends classification-only approaches by linking disease recognition with quantitative severity assessment and interpretable prognosis. The framework therefore offers a practical basis for rapid cotton-health screening and precision-agriculture decision support, while field validation remains necessary before operational deployment.