Classification of Peach Fruits in Ripe, Unripe and Damaged Towards Automated Harvest

Abstract

Using computer vision technology, specifically convolutional neural networks, a solution was proposed to perform the recognition of ripe peach fruits, as well as the identification of damaged fruits. The purpose is to obtain fruits with the appropriate level of quality for their commercialization. To achieve this purpose, images of peaches were obtained in an uncontrolled environment. Digital images were cropped until the area of interest was obtained. Three data sets were configured: the first, for ripe and unripe peaches; the second, also of ripe and unripe peaches but only focused on a textural area, and the third, of healthy and damaged peaches. A convolutional neural network was applied, which was programmed in the Python language, the Keras and TensorFlow libraries. During the tests, a precision of 95.31 % was obtained when choosing between mature and immature. While when classifying healthy and damaged peaches, 92.18 % accuracy was obtained. Finally, when classifying the three categories (damaged, immature and mature), 83.33 % precision was obtained. The previous results indicate that with artificial intelligence embedded in a physical device, the classification of the peach fruit can be done.