Dual Attention Residual Network(DARN) for Multi crop leaf disease classification

Effective classification of crop leaf diseases are paramount for ensuring agricultural productivity and sustainability. The exploration present a Dual Attention Residual Network (DARN), designed to address the challenge of multiple crop leaf disease detection and classification across three different crop types: Tomato, Potato, and Corn. The proposed DARN architecture integrates channel wise spatial attention mechanisms within residual blocks, optimizing feature extraction and boosting model performance. For Tomato leaf disease detection, our model attained a comprehensive training accuracy of 99.49%, with an average training accuracy of 94.13% and an overall validation accuracy of 98.62%, with an average validation accuracy of 94.74%, respectively, over 100 epochs. For Potato leaf disease detection, DARN yielded an overall training accuracy of 98.84%, with an average training accuracy of 95.08%, and an impressive overall validation accuracy of 99.53%, with an average validation accuracy of 95.89%. In the case of Corn leaf disease detection, the model demonstrated an overall training accuracy of 99.66%, with an average training accuracy of 92.89%, and an overall validation accuracy of 95.83%. These results highlights the effectiveness of the DARN framework in correctly identifying and classifying leaf diseases across multiple crop types, showcasing its prospective for deployment in practical agricultural scenarios.