Deep learning technique for rice leaf disease classification and severity level identification via Hybrid ResConvolutional Neural Network

Rice leaf diseases significantly reduce agricultural yield and pose a major challenge for sustainable food production, particularly owing to the limitations associated with manual and visual inspection methods that are subjective and often ineffective in early-stage detection. This investigation proposes a new Hybrid ResConvolutional Neural Network (HyResCN-Net) for effective classification and severity identification of rice plant leaf diseases. The proposed framework integrates an Internet of Things (IoT) -based data acquisition and routing simulation using CrowWhale Energy Trust Routing (CrowWhale-ETR) for efficient data handling. Initially, preprocessing is done by an averaging filter to reduce noise. Then, plant leaves are segmented using the Eff-UNet++ method. Augmentation techniques like rotation, scaling, and color change are applied to expand the dataset. Features, like entropy with Gradient Directional Pattern (GDP), Complete Local Binary Pattern (CLBP), and histogram features, are extracted to enhance feature representation. These features are then used within the proposed HyResCN-Net model, which integrates Parallel Convolutional Neural Network (PCNN) and ResNeXt to improve discriminative learning for disease classification and severity estimation. Experimental evaluation is conducted on the Rice Leaf Bacterial and Fungal Disease Dataset. Considering a k-value of 8, the HyResCN-Net gains an accuracy of 94.258%, a True Positive Rate (TPR) of 96.479%, a True Negative Rate (TNR) of 92.898%, a precision of 91.312% and an F1-score of 93.824% compared to existing methods. The HyResCN-Net efficiently enhances rice leaf disease identification and severity analysis, supporting its applicability in precision agriculture applications.