Complex scene green walnut target detection method based on improved YOLOv11n

To meet the efficiency and accuracy requirements of agricultural product inspection, deep learning-based object detection technology has been widely adopted. However, factors such as varying lighting conditions and foliage occlusion cause green walnuts in complex scenes to exhibit low texture contrast and a tendency to lose edge information. To address these issues, this paper proposes GPL-YOLO, an object detection algorithm for green walnuts based on an improved YOLOv11n architecture. First, the GhostModuleV2 module is introduced into the backbone network to enhance global modeling capabilities and enrich the detailed information of walnut targets within high-level feature maps. Second, a PPA module is integrated into the multi-scale feature fusion stage of the neck; this module effectively filters and enhances fused features across spatial and channel dimensions, thereby highlighting the target regions of green walnuts while suppressing background noise. Finally, a DW-LAE feature enhancement module is embedded in the downsampling fusion branch to adaptively preserve and enhance shallow-level detailed features during downsampling, thereby minimizing the loss of edge, texture, and shape information and improving the effectiveness of deep semantic features. Experimental results demonstrate that, compared to the baseline model, this algorithm achieves improvements of 8.6%, 5.7%, and 8.7% in Precision, Recall, and mAP50, respectively. The improved model outperforms the original YOLOv11n in detection accuracy and effectively mitigates missed detections and false positives in complex backgrounds, providing technical support for intelligent harvesting, yield estimation, and precision management in green walnut orchards.