Performance comparison of different convolutional neural networks for vegetable and fruit recognition

Abstract

Image classification, a significant research problem in the computer vision community, aims to assign different types of images to a certain category in a fixed category set according to different information features reflected in images. With the continuous improvement of living standards, people's daily demand for accurate identification of food categories (such as vegetable, fruit, etc.) is growing. Early vegetable and fruit recognition mostly relied on manual features and machine learning algorithms, with low recognition accuracy and weak generalization ability. Thanks to the rapid evolution of the convolutional neural network, vegetable and fruit recognition based on depth learning has made breakthroughs in accuracy and speed. In this paper, three representative convolutional neural networks are introduced around vegetable and fruit recognition, and their performance differences are quantitatively compared on different data sets to explore the boundaries of their applications. In addition, we summarized the research problems in vegetable and fruit recognition and discussed its future development direction.