Fusion of Region Extraction and Cross-Entropy SVM Models for Wheat Rust Diseases Classification

Wheat rust diseases are one of the major types of fungal diseases that cause substantial yield quality losses of 15%–20% every year. The wheat rust diseases are identified either through experienced evaluators or computerassisted techniques. The experienced evaluators take time to identify the disease which is highly laborious and too costly. If wheat rust diseases are predicted at the development stages, then fungicides are sprayed earlier which helps to increase wheat yield quality. To solve the experienced evaluator issues, a combined region extraction and cross-entropy support vector machine (CE-SVM) model is proposed for wheat rust disease identification. In the proposed system, a total of 2300 secondary source images were augmented through flipping, cropping, and rotation techniques. The augmented images are preprocessed by histogram equalization. As a result, preprocessed images have been applied to region extraction convolutional neural networks (RCNN); Fast-RCNN, Faster-RCNN, and Mask-RCNN models for wheat plant patch extraction. Different layers of region extraction models construct a feature vector that is later passed to the CE-SVM model. As a result, the Gaussian kernel function in CE-SVM achieves high F1-score (88.43%) and accuracy (93.60%) for wheat stripe rust disease classification.