Anomaly Detection and Classification in Agricultural Produce Using Image Processing and CNN Assisted by a Robotic Arm

TheGlobal Hunger Crisis has long been one of the most pressing problems of the modern world. Surveys have shown that globally, around 14 percent of food produced is wasted between harvest and retail. This project aims to develop a mechanism that uses image processing and deep learning to classify agricultural produce and perform anomaly detection. The system performs two kinds of evaluations; a mass-evaluation and a singular evaluation. The mass evaluation of produce is done by angling a camera at an angle theta (.), that is pre calculated through an optimal angle calculation algorithm. In addition, the system provides controls to a supervisor to specifically evaluate individual items based on the factor of "intuitive inquiry". In this process, a robotic arm picks the target item and takes it to the camera physically for end-to-end coverage. The data obtained from both mass analysis and individual analysis is fed into a program containing metrics for evaluation. Based on the degree of adherence/divergence from standards, the system also recommends a further progression by classifying the item into sets-i.e., if the item is anomaly-free, if it is fully defective and must be discarded, if it can be corrected through further processing, or if it has been under processed. With each iteration of item evaluation, the system intelligently learns from its decisions for improved accuracy and speed.