MATHEMATICAL MODEL OF HYBRID DEEP LEARNING FOR MULTI-CLASSIFICATION OF PLANT DISEASES

Abstract

 Plant diseases significantly impact on global agricultural productivity and crop yield which is a threat to food security. It is imperative to have effective disease management through timely and accurate identification of these diseases. It reduces crop losses and enabling farmers to take appropriate action. Existing AI models struggle with challenges such as poor generalization to complex leaf patterns, difficulty in distinguishing between visually similar diseases, and limited scalability across multiple crop species. These issues have been addressed and presented in this work where a hybrid model that integrates Mobile Network Version-2 (MNV2) and machine learning such as Support Vector Machine (SVM) learning . The MNV-2 is known for its ability to understand complicated patterns and identify images reliably. The hybrid model leverages the spatial feature extraction capabilities of MNV2 with SVM allowing for more effective learning of both low- and high-level discriminative features. The proposed approach has been evaluated on a multi-class dataset comprising 38 categories, which includes 30 diseased and 8 healthy class of various plant disease from the Plant Village dataset. Simulation results have shown that the hybrid model achieved an accuracy (Ac) of 98.95%, a Precision (Pr) of 99.0%, Recall (Re) of 98.9% and a F-1 Score of 99% representing improvement over standard transfer learning techniques. These results accentuate the model’s possible for real-world solicitation, including amalgamation into mobile-based platforms to support agriculturalists with on-site analysis of the infection and strategies for timely involvement and accomplishment.Hybrid CNN Model; Support Vector Machine Learning, Depthwise Seperable Convolution, Bottelneck Design of MNV2 Architecture, Machine Learning