Cherry Size and Shape Classification Detection Based On Deep Convolutional Neural Network

Abstract

In order to enhance the post-production value of cherries, to improve the efficiency of cherry sorting, and to standardize and commercialize the industry, cherry grading detection becomes extreamly important. In this paper, we proposed a deep learning-based key point detection algorithm to identify the size and shape of cherries, key point features were extracted based on the fruit body through a feature extraction network, and a heat map regression method was used to construct a model to obtain the key point coordinates of the cherry fruit body, and the purpose of grading detection was achieved. The test results show that the accuracy of cherry size detection is 95.18%, and the accuracy of deformity detection is 94.50%. The network detection method proposed in this paper can effectively detect the size and deformity of cherries with high accuracy, and the average speed of detection is about 59 pieces/s, which meets the demand of real-time.