A Detection Method for Lion-Head Goose in Complex Scenes Based on Rcd-Yolo

To address the challenges of detecting Lion-head goose in complex scenes such as multi-scale targets and occlusion in farming environments, we propose a method based on improved YOLOv8n, named RCD-YOLO. Firstly, the standard convolution module CBS of the backbone is replaced with the Receptive-Field Attention Convolution (RFAConv) module to enhance the network's ability to extract features of Lion-head geese. Secondly, the neck is replaced with the improved CNN-based Cross-scale Feature Fusion (CCFM) network to enhance the network's multi-scale feature fusion capability for feature of Lion-head geese at different scales. Finally, the head is replaced with DynamicHead (DyHead) component to enhance the detection ability for occluded Lion-head geese in farming environment by employing scale-aware, spatial-aware, and task-aware attention mechanisms. Experimental results show that the precision (P), recall (R), mAP@0.5 and mAP@0.5:0.95 of RCD-YOLO are 90.6%, 84.6%, 91.5% and 59.9%, respectively. Compared to YOLOv8n, these metrics have improved by 1.6%, 2.9%, 1.7%, and 2.5%, respectively. This method can achieve visual detection of Lion-head geese in complex scenes in actual farming environments and provide a reference for the Lion-head Goose farming industry, contributing to its intelligence.