现代农业装备2025，Vol.46Issue(5)：64-70,7.DOI:10.3969/j.issn.1673-2154.2025.05.009
基于机器视觉的串果和类球果识别技术研究进展
Research Progress in Machine Vision-Based Recognition Technology for Cluster Fruits and Globular Fruits
摘要
Abstract
The global agricultural sector is facing dual challenges of labor shortages and the demand for precise harvesting.Traditional mechanical harvesting methods suffer from significantly high fruit damage rates due to the inadequate adaptability of visual systems,which has accelerated the development of intelligent recognition technologies based on machine vision.This paper systematically reviews the evolution of recognition technologies for cluster fruits(e.g.,grapes,lychees)and globular fruits(e.g.,apples,citrus fruits):early traditional image processing methods relied on color space conversion(HSV,Lab)and morphological feature extraction,achieving detection accuracy of 82％～96％under ideal conditions but exhibiting poor environmental robustness and insufficient real-time performance.Traditional machine learning techniques improved accuracy to 88％～95％by optimizing feature weights with classifiers such as SVM and random forests,yet remained constrained by the semantic understanding limitations of manual feature design.Deep learning methods,represented by YOLO and Mask R-CNN,broke through the restrictions of feature engineering,achieving detection accuracy of 90％～98％in complex orchard environments.After optimization with lightweight networks(e.g.,MobileNet,Ghost modules)and attention mechanisms(e.g.,CBAM,CA),detection speeds reached 15～120 FPS.Current technologies face challenges such as strong data dependency and limited generalization ability with small samples.Future development will focus on multi-modal sensor fusion,self-supervised learning,and edge computing optimization to provide technical support for high-precision,low-loss operations in agricultural harvesting robots.关键词
果实识别技术/深度学习/串果检测/农业机器人Key words
fruit recognition technology/deep learning/cluster fruit detection/agricultural robots分类
农业科技引用本文复制引用
毛顺,朱立学,张世昂,吕秋辉..基于机器视觉的串果和类球果识别技术研究进展[J].现代农业装备,2025,46(5):64-70,7.基金项目
国家自然科学基金项目(32472015) （32472015）
广州市科技计划项目(2023B03J0862) （2023B03J0862）
