Abstract

The agriculture industry's production and food quality are significantly impacted by plant diseases. It is vital to automatically identify and diagnose diseases at an early stage to improve the quality of agricultural output and prevent widespread plant extinction. State-of-the-art methods do not provide accurate identification when there is similarity among different diseases or when input leaf images have complex backgrounds. In this work, a Transfer Learning-based automated crop disease recognition system is proposed. The proposed Cotton Leaf Disease Detection framework consists of two key stages: (1) Optimal Feature Extraction: A pre-trained Convolutional Neural Network (CNN) model from the Visual Geometry Group (VGG16) is used to extract deep features from cotton leaf images with complex backgrounds. (2) Cotton Leaf Disease Detection: The extracted optimal features are classified using traditional classifiers such as Random Forest (RF) and non-linear SVM for the final classification of diseases. To achieve optimal performance in detecting cotton crop diseases, Kaggle’s four-class cotton disease dataset with a complex background was chosen. Upon analysis, the proposed hybrid approach—combining VGG with RF (VGG-RF) and VGG with SVM (VGG-SVM)—achieved average accuracies of approximately 98.29% and 99.31%, respectively. This demonstrates that the proposed hybrid VGG-SVM model outperforms state-of-the-art traditional classifiers.

Authors

Pandiyaraju V; Anusha B; Senthil Kumar A M; Shravan Venkatraman; Diya Sapra; Kannan A