A Frequency-Guided Modular Neural Network for Enhanced Rice Plant Disease Classification

The rice plant disease is a major threat to agricultural productivity that lowers crop yield, and early detection lowers the complications of productivity. Hence, inorder to perform early detection of disease, a novel Frequency-Guided Modular Neural Network is proposed, which merges spatial and spectral information to perform accurate diagnosis of plant disease. However, the traditional Deep Learning models are not effective in selecting frequency-specific cues such as high-frequency lesion edges and low-frequency color distortions. These issues are addressed by the proposed Frequency-Guided Modular Neural Network methodology that initiates the process by decomposing the preprocessed rice leaf image into multiple frequency sub-bands using a multilevel frequency transform. The proposed Frequency-Guided Modular Neural Network exhibits the significance of diagnosis, in which each frequency band analyzes spectral energy patterns to extract the diagnosis performance based on extracted signs in the plant. The refined module outputs are fused by compressing the encoded layer, and the final classification effectively detects the disease. The experimental evaluations are performed by diverse benchmark datasets, and the results demonstrated that Frequency-Guided Modular Neural Network model outperforms the performance and achieves a better range of accuracy, precision, recall, F1-score, and specificity at$98.94 \%, 98.85 \%, 98.72 \%, 98.77 {\%}$and 98.40% respectively. Thus, the proposed model signifies that it estimates strong deployment in the agricultural monitoring process.