基于YOLO-TM的番茄采摘机器人感知算法OA

To address the issue of insufficient detection accuracy in tomato harvesting robot operations caused by factors such as inter-class and intra-class occlusion,complex lighting variations,and larger scale differences,this paper proposes a tomato maturity detection model based on YOLO-TM(YOLO-transformer for tomatoes maturity detection).First,a multi-head self-attention(MHSA)mechanism is introduced into the backbone network to enhance global feature extrac-tion and suppress background interference.Then,a bidirectional adaptive feature pyramid network(BAFPN)is construct-ed to improve multi-scale feature fusion capability.Furthermore,a Lbox regression loss function is designed to optimize the localization accuracy of small-scale tomato targets.The experimental results show that YOLO-TM achieves a mean aver-age precision(mAP)of 95.3%and an inference speed of 94.6 frames per second(FPS)on a self-collected tomato maturity detection dataset.Compared with the baseline model YOLOv11,YOLO-TM improves mAP by 4.2 percentage points,and achieves a field picking success rate of 94.0%.Compared with Faster R-CNN and other mainstream YOLO series models,YOLO-TM significantly improves detection accuracy while maintaining high real-time performance,demonstrating supe-rior adaptability in complex greenhouse environments.This study provides a strong theoretical basis and technical support for the automated precise picking of tomato and the development of visual perception systems for intelligent agricultural equipment.