SRW-YOLOv8n: a high-precision method for main-stem detection and clamping-point positioning of plug pepper seedlings

Precise positioning of clamping-points is the core and difficulty of realizing fully automated grafting of plug pepper seedlings. Traditional mechanical positioning methods often struggle to accommodate the morphological variations of pepper seedlings across an entire plug tray, resulting in large positioning errors and high clamping failure rates. To address this problem, this study develops an improved YOLOv8n-based framework for accurate detection and spatial positioning of seedling clamping points. The baseline YOLOv8n is optimized by integrating the SimAM, RFAConv and WIoU loss function to establish an enhanced SRW-YOLOv8n model. Moreover, a shielding-supporting mechanism and structured image processing strategy are adopted to suppress dense seedling interference, and depth camera calibration is applied to convert pixel coordinates into 3D spatial coordinates. Experimental results show that the SRW-YOLOv8n achieves 96.6% precision, 98.4% recall, 97.5% F1 and 97.4% mAP@0.5, outperforming the original YOLOv8n. The proposed system delivers average absolute positioning errors of 2.49 mm, 2.39 mm and 1.83 mm in the x, y and z axes, fully satisfying high-precision grafting requirements. This method provides robust spatial positioning guidance for automated pepper seedling grafting operations.