Classification of apples using artificial vision and artificial neural networks

Abstract

The added value in a fruit can be increased with a good postharvest handling. The classification in different parameters is one of the most important operations. In small companies it is done manually, obtaining deficiencies in the quality of the product. These problems could be solved or reduced with the implementation of intelligent algorithms that in this case include artificial vision and artificial neural networks. In this project is presented the classification of apples through an intelligent algorithm, using a convolutional neural network (CNN), which is developed using Open Source libraries (OpenCV, Tensorflow and Keras) in Python with a structure of different convolutional layers and MaxPooling, for a dataset of 2,800 images of 128x128 pixels, of which 80% were used for training and 20% for test of the network, obtaining an accuracy of 98.3% and 95.36%, respectively. After the training a classification was made with a video in real time, obtaining an accuracy of 92.25%. Likewise, the possibility of using it in the industry is explored with the classification by other visual characteristics of the fruit such as size, color, shape, etc.