Crop Vigil: Automated Wheat Bunt Disease Multi-Classification with a CNN-RNN Hybrid Model and Attention Block

Wheat harvests are seriously threatened by wheat bunt disease, which lowers productivity and grain quality. For efficient illness management, disease intensity levels must be accurately and promptly classified. In this study, the proposed study presents a novel method for categorizing the intensity of wheat bunt disease into six stages by combining recurrent neural networks (RNN) and convolutional neural networks (CNN), together with an attention block. The goal is to create a dependable and understandable model that supports disease management choices and increases crop output. To capture spatial and temporal relationships in illness development, the proposed hybrid model makes use of the strengths of CNNs and RNNs. While the RNN component detects sequential patterns, the CNN component extracts picture features. Furthermore, an attention block method is included to draw attention to instructive areas of the pictures, improving interpretability and classification precision. Several criteria, such as accuracy, precision, recall, and F1 score, were used to evaluate the model. The model's overall accuracy on the testing set was determined to be 0.9225, meaning that it was successful in accurately categorizing 92.25% of the photos of wheat bunt disease across all intensities. A comparative study with cutting-edge methods showed the poposed hybrid model to be superior. Its results outperformed current models, demonstrating its potential for useful application in the categorization of wheat bunt disease. The attention block's inclusion improved interpretability and allowed for the focus on important illness symptoms.