Hybrid Method for Training Customized CNN mode for Pomegranate Disease Detection, Classification and Control Mechanisms

Abstract

fruit is both a medicinal and fruit plant cultivated in large quantity in Karnataka. Diseases are the main reasons that impact post-harvest profit and production. Identifying the diseases through automated methods by using real time images will provide information on control mechanisms for farmers. Detecting the diseases on field with simple mobile apps that are loaded on the farmers mobile that connect to background network for data access and inference of disease is an important step to reduce production loss and recommend control mechanisms for spread of disease. In this work, new methods are proposed that combines wavelet features, support vectors, customized CNN model (TinyYOLOV4) and control mechanisms that provide quick information to the farmer once the photograph is uploaded in the app. The classification accuracy of the proposed model is 99.4% and is generates the inferences faster that aids real time assistance. Two methods are combined to provide control mechanism and to improve mean average precisions. The model is validated and demonstrates 4% improvement in precision accuracy compared with existing methods.