Abstract:To standardize cherry quality assessments for improved commercialization and increased market value，cherry grading has become an indispensable tool.This paper proposes a key point detection method based on deep learning to classify cherries by size and stalk characteristics.A convolutional neural network was used to automatically extract key characteristics of each cherry.A regression network model was then built to coordinate key points at both ends of the fruit stem and both sides of the fruit calyx，to achieve the goal of cherry grading.The accuracy of cherry size detection was 93.14%，and the accuracy of cherry stalk determination was 90.57%.The key point regression detection method based on deep learning effectively detected cherry size and stalk characteristics with high accuracy，and a detection speed of 33 fps，which could meet real-time requirements.