A Robust Integrated DenseNet201-SVM Approach for Wheat Leaf Disease Detection

Wheat leaf diseases significantly affect agricultural productivity, crop quality, and global food security. Manual disease inspection is time-consuming, subjective, and less reliable for large-scale field monitoring. This study proposes a robust DenseNet201-SVM integrated framework for automated wheat leaf disease classification. “DenseNet201 is used as a deep feature extractor, while a Support Vector Machine (SVM) performs the final classification. The dataset contains 54,306 images from four classes: Healthy, Yellow Rust, Brown Rust, and Powdery Mildew. The proposed DenseNet201-SVM model achieved 98.64% accuracy, 98.5% precision, 98.7% recall, and 98.6% F1-score. Confusion matrix analysis showed strong class-wise performance, with most errors occurring between visually similar rust categories. The ensemble model further improved accuracy to 99.1%. Statistical validation using a paired t-test produced a p-value of 0.001, confirming that the improvement was statistically significant. The results demonstrate that the proposed framework is accurate, robust, and suitable for intelligent wheat disease monitoring systems.