基于改进YOLO11n-seg模型的人参幼苗分拣算法OA

To improve the automation and intelligence of ginseng seedling sorting prior to transplan-tation,this study proposed a DBE-YOLO11-seg ginseng seedling recognition model to address the challenges of small seedling targets,slender roots,severe occlusion and adhesion,and insufficient segmentation accuracy under complex illumination backgrounds.The proposed model used YOLO11n-seg as the baseline.A DCNv4-C3k2 module was constructed in the backbone network to enhance the geometric modeling capability for slender and curved targets.A P2 feature layer and an HFFE module were introduced into the neck to improve the preservation of shallow texture details and the effectiveness of multi-scale feature fusion.In the segmentation head,CARAFE upsampling and a boundary-assisted supervision strategy were adopted to construct a CARAFE-Edge-Head seg-mentation and detection head,thereby improving mask boundary recovery and separation of adherent targets.Training and testing were conducted on a self-constructed ginseng seedling dataset,followed by validation on a sorting system test bench.The results showed that the proposed model achieved an mAP@0.5 of 92.7%,a segmentation mAP@0.5 of 87.2%,a BIoU of 80.9%,and a sorting accuracy of 96.1%,demonstrating superior overall performance compared with those of Mask R-CNN,YOLACT++,YOLOv8n-seg,and YOLO11n-seg.Therefore,the proposed method can effectively im-prove the recognition and segmentation performance of ginseng seedlings in complex sorting environ-ments,and possesses promising potential for engineering applications.