A method for classifying oranges based on image processing and neural networks

In recent days, there have been many recommendations on social media about eating healthy fruits to strengthen the immune system and corona resistance. Therefore, it is very important to identify spoiled fruits at this time when human society is concerned about coronavirus and the human body needs healthy fruits in case of this disease. This paper proposes a method to identify the type of defects found in orange fruits. We used a machine vision system to capture sample images, which includes a charge-coupled device camera, black box, lighting system, and personal computer. The citrus fruits are classified into eight classes, including Wind scar, Stem-end breakdown, Snail bites, Thrips scar, Scale injury, Medfly, Rings, and Calyx, depending on the type and model of the defects. In the proposed method, classification by the neural network with the help of co-occurrence matrix for four angles θ=0°, 45°, 90°, and 135°, were extracted to identify various defects and 24 features related to the areas with defect in citrus. For the final classification of defects in citrus, after evaluating many classification tools from various tools available, Feed-forward Back Propagation Neural Network (FFBPNN) is used. The result of the neural network classifier was obtained with the help of the co-occurrence matrix by taking four angles (horizontal, right diagonal, vertical, and left diagonal) with an accuracy of 89.65%. The evaluation shows acceptable results compared with similar studies. It is a reliable method in the food classification industry with reasonable accuracy.