<scp>YOLOv8n</scp> ‐ <scp>SLW</scp> : A Lightweight Ripe Yellow Peach Detection Algorithm

ABSTRACT Accurate and efficient detection of maturity‐stage yellow peach is critical for advancing automated harvesting robotics. However, studies on yellow peach detection remain insufficient, with most efforts focused on detecting unripe yellow peach. To address this research gap in agricultural vision systems, a novel lightweight ripe yellow peach detection algorithm based on architectural optimizations of YOLOv8n is proposed to meet the real‐time detection requirements of automated harvesting robots. First, a lightweight attention mechanism module, SimAM (Simple Attention Mechanism), was integrated into the backbone of YOLOv8n to enhance the model's ability to extract features of ripe yellow peach. Subsequently, all standard convolutional layers in the backbone were replaced with LDConv (Linear Deformable Convolution), which accommodates the geometric deformations of ripe yellow peach and reduces the model's parameter count. Finally, the original loss function was substituted with WIoUv3 (Wise Intersection over Union version 3). Through a dynamic non‐monotonic focusing mechanism, the model's generalization performance and detection localization accuracy are improved. Experimental results on the self‐built ripe yellow peach dataset indicated that the mean Average Precision (mAP) and recall of 95.49% and 90.56% were achieved, respectively, which were 4.09 and 2.00 percentage points higher than the baseline model. Compared to other mainstream lightweight models, the highest mean average precision was attained by this model, with only 2.62 M parameters and 7.51G FLOPs (floating‐point operations), and achieved a maximum FPS of 92.7. Efficient performance was demonstrated by YOLOv8n‐SLW, which could provide a reliable technical foundation for the detection system of yellow peach picking robots.