Enhanced wheat leaf disease detection using a modified ResNet-50 architecture with advanced training techniques

Identification and right treatment of wheat foliar diseases are necessary because they open the way for sustainable farming techniques. To classify wheat leaf diseases consisting of StripeRust, Septoria, and healthy leaves, this study introduces a modified ResNet-50. The model, using transfer learning, label smoothing, progressive resize, and an adaptive learning rate scheduler, shows improved performance compared to baseline designs. The combination of such strategies minimizes overfitting and boosts generalization performance. Experiments show a significant increase in both robustness and classification accuracy, making this method fit for practical applications. The proposed method achieves a validation accuracy of 95.4%, exceeding other competitive methods robustness and classification accuracy, making this method fit for practical applications. The proposed method achieves a validation accuracy of 95.4%, exceeding other competitive methods.