An improved YOLOv11n dense object detection algorithm

To address the issue of low detection accuracy, missed detection, and false positives in safety helmet target detection under complex backgrounds and dense targets, an improved YOLOv11n algorithm for dense target detection (denoted as Y11-TP1245) is proposed. Firstly, a Reduced Channel Spatial Object Attention (RCS-OSA) module is introduced to enhance the model's feature extraction ability by reducing spatial object complexity. Next, the network structure is modified by removing the P3 detection layer and adding P1 and P2 detection layers for small targets, while concatenating the features from the backbone and neck to enable more efficient feature fusion across different levels. Experimental results show that the Y11-TP1245 algorithm improves the average precision by 3.7% compared to the original YOLOv11n network. Finally, real-time detection is performed using the D435i depth camera, demonstrating that the improved algorithm is more suitable for dense target detection tasks.