广西科技大学学报2026，Vol.37Issue(3)：12-19,8.DOI:10.16375/j.cnki.cn45-1395/t.2026.03.002
针对复杂场景下荔枝采摘的目标检测算法研究
Object detection algorithms for litchi harvesting in complex scenes
摘要
Abstract
Accurate identification and harvesting of litchi fruits are crucial for achieving mechanized and automated harvesting.Due to the small size and dense growth of litchi fruits,traditional object detection algorithms struggle to achieve efficient recognition.This paper proposed the YOLOv8-SC algorithm based on the YOLOv8 model,which integrated the Swin Transformer into the backbone network to enhance small object detection performance.A shared-parameter detection head was designed to replace the original decoupled head,achieving model lightweighting.Additionally,the density-based spatial clustering of applications with noise(DBSCAN)clustering algorithm was employed to segment litchi growth patterns,providing guidance for harvesting strategies.Experimental results show that YOLOv8-SC achieves a precision of 88.5%,recall of 76.3%,F1 score of 0.819,mean average precision(mAP@0.5)of 82.1%,and mAP@0.5:0.95 of 58.2%in complex scenes,with frame rates increased to 119 frames/s.On the Jetson Xavier NX,the real-time frame rate was improved to 26 frames/s,demonstrating significant performance optimization.In field tests conducted at a litchi orchard in Lianjiang,Guangdong,the DBSCAN algorithm exhibited excellent target clustering performance,providing a reliable basis for harvesting planning.关键词
荔枝采摘/YOLOv8/DBSCAN/小目标检测/模型轻量化Key words
litchi harvesting/YOLOv8/DBSCAN/small object detection/model lightweighting分类
信息技术与安全科学引用本文复制引用
李添恒,丛佩超,胥羽涛,梁吉,王昆..针对复杂场景下荔枝采摘的目标检测算法研究[J].广西科技大学学报,2026,37(3):12-19,8.基金项目
中央引导地方科技发展专项资金项目(桂科ZY19183003) （桂科ZY19183003）
广西重点研发计划项目(桂科AB20058001)资助 （桂科AB20058001）
