Enhanced YOLOv11n algorithm for UAV small target detection

Object detection in Unmanned Aerial Vehicle (UAV) aerial imagery encounters several challenges, such as small target sizes, significant scale variations, and limited computational resources. To surmount the difficulties presented by complex and dynamic UAV scenes, substantial target scale variations, and the predominance of small targets, an improved YOLOv11n algorithm is put forward. First, the Receptive Field and Channel Attention Module Convolution (RFCBAMConv) is integrated into the C3k2 module to boost the model's capacity for extracting multi-scale features from complex backgrounds. Next, the SPPF layer is optimized by integrating the LSKA attention mechanism, which reduces computational complexity and memory consumption. Additionally, an enhanced feature pyramid structure, based on modifications to the original PAN-FPN, is suggested to enhance small-target detection. Instead of adding a conventional P2 detection layer, this method uses the feature map from the P2 layer, applies SPDConv to extract small-target-rich features, and fuses them into P3. Subsequently, the CSP (Cross Stage Partial) concept is enhanced using OmniKernel, leading to the development of CSP-OmniKernel for feature integration, ultimately improving small-target detection performance. Experimental results show that the improved model achieves a 6.6% increase in mAP@0.5 and a 4.5% improvement in mAP@0.5:0.95 on the VisDrone dataset, confirming its feasibility.