Rice spikelet’s disease detection using hybrid optimization model and optimized CNN

Abstract

Rice fields all across the world are affected by spikelet sterility, often known as rice spikelet's disease. It is characterized by the improper development of spikelet’s, which lowers grain output and quality. For optimal management and the avoidance of yield losses, this disease must be discovered early. In this study, a deep learning-based approach utilizing optimization techniques was proposed for accurate segmentation and disease detection in rice crops. The research aimed to address the vulnerability of rice crops to disease attacks from seed germination to mature spikelets. The proposed model consisted of three major phases: pre-processing, segmentation, feature extraction and disease detection. Gaussian filtering was employed to preprocess the raw rice crop images, while a new hybrid Whale Customized Gravitational Optimization Algorithm was utilized for segmentation. Disease detection was performed using a hybrid deep learning model called HybridNet, which combines convolutional neural networks (CNN) with an optimized recurrent neural network (RNN) model. The dataset intended for this proposed project is sourced from the CABI PlantwisePlus Knowledge Bank. The proposed model for rice spikelet disease detection achieved high accuracy with Sensitivity (0.9269), Specificity (0.9756), and Accuracy (0.9634). This indicates that the model effectively identifies and detects diseases in rice spikelets, demonstrating its reliable performance in disease management and crop protection.