Classification of Fruit Images as Fresh and Rotten Using Convolutional Neural Networks

Abstract

Many fruits are produced all over the world, and the fruits produced are sent abroad and sold in a relatively short time in many countries. During the period between collection and sale, the fruits may undergo various types of spoilage. In many cases, the fruits go through a number of fabrication processes for sale. If the fruits comply with certain standards after passing the quality tests at the factory, for example, if the fruits have not started to rot, shipment can begin. In this study, we consider the problem of detecting any type of rot/mold as a binary classification problem. To train and test different convolutional neuralnetwork models, we have collected a dataset using images of rotting /spoiled and fresh fruits. Different deep-learning models have been implemented and tested to solve this problem. In this study, the ResNet50 network architecture with real-time data augmentation obtained the best results. Our experimental results show that the ResNet50 model can solve this problem with an accuracy of around 90%.