Cherry detection algorithm based on improved YOLOv5s network
Rongli Gai, Mengke Li, Na Chen
Abstract
In order to realize the automatic fruit picking by robot in orchard environment, cherry target detection is carried out by studying the algorithm based on deep learning, so as to identify cherry and carry out picking operation. However, most of the existing methods are aimed at relatively sparse fruits and cannot solve the detection problem of small and dense fruits. In this paper, a YOLOv5s-cherry model based on YOLOv5s was proposed for cherry detection. By adjusting the CSP structure of BackBone and Neck, deepening the depth and width of the network, improving the ability of network feature extraction, feature fusion and learning ability of network feature extraction, so as to improve the detection ability of small and dense cherry fruit. The model is trained on the self-made cherry maturity classification data set. The trained YOLOv5s cherry network is used to extract the feature and location information of cherry images with different maturity, so as to realize the classification and detection of cherry. We tested the performance of the model, compared YOLOv5s-cherry with YOLOv5s and YOLOv4, and used F1 value as the evaluation value. The test results show that the F1 of YOLOv5s cherry is 0.08 higher than that of YOLOv4 algorithm and 0.03 higher than that of YOLOv5s algorithm. The improved YOLOv5s cherry model provides better detection performance for small and dense cherry detection, which is conducive to intelligent picking.