A Target Detection Method for UAV Aerial Images Based on Bidirectional Weighted Feature Fusion YOLOv8s

With the rapid advancement of deep learning technologies, target detection in Unmanned Aerial Vehicle (UAV) aerial images has found extensive applications in fields such as transportation and urban management. However, in practical applications, the high altitude and unique shooting angles of UAVs often result in targets in aerial images being small in size and prone to mutual occlusion, leading to a decline in the performance of detection models and a need for improved accuracy. To address these issues, this paper proposes an improved model based on YOLOv8s: the BAC-YOLOv8 model. Firstly, the PANet structure in the feature fusion stage of YOLOv8 is replaced with a Bidirectional Feature Pyramid Network (BiFPN) structure. By assigning reasonable weight distributions to input features of different scales during the feature fusion stage, multi-scale features are successfully integrated. Secondly, in the C2f part of the backbone, a lightweight convolution and attention mechanism hybrid module (ACMix) is incorporated, which introduces minimal computational overhead while significantly enhancing the detection capability for small targets. Finally, experiments conducted on the VisDrone2019 dataset show that, compared to YOLOv8s, the BAC-YOLOv8 model achieves a 2.5% improvement in the precision metric, demonstrating its superior performance.