SHUFFLED SHEPHERD SOCIAL POLITICAL OPTIMIZED DEEP LEARNING FOR RICE LEAF DISEASE CLASSIFICATION AND SEVERITY PERCENTAGE PREDICTION

Abstract

Rice is the most commonly consumed food in the world and several diseases affect the rice plants easily resulting in huge economic losses and decreased yield. Thus, the early stage of identification is necessary to control and alleviate the influences of pest attacks. The common disease affecting in rice is brown spot (BS). Most of the previous methods used image recognition techniques and machine-driven disease diagnosis systems to detect the crop diseases. However, these techniques are not feasible to process lots of images, time-consuming, inaccurate, and expensive. Hence, an effective approach, named shuffled Shepherd social political optimization algorithm (SSSPOA) based deep learning is developed for rice leaf infection categorization and severity percentage detection. The developed SSSPOA is the merging of shuffled shepherd social optimization (SSSO) and political optimizer (PO). Here, the input image is pre-processed by using the RoI extraction method to eliminate the unwanted noise from the image. Then, the segmentation process is done by using the DFC technique. Deep maxout network (DMN) is adopted for grading the leaf diseases into blast, bacterial blight, tungro, and BS where the training step of DMN is conducted utilizing designed SSSPOA. In addition, forecasting of severity percentage takes place using deep long short-term memory (LSTM) by taking segmented values such that the tuning mechanism of deep LSTM is done utilizing the same SSSPOA. Therefore, the presented strategy outperformed different conventional models and achieved efficient performance with a higher testing accuracy of 0.954, a sensitivity of 0.987, a specificity of 0.965, a lower mean square error (MSE) of 0.076, and a lower root mean square error (RMSE) of 0.275, respectively.