Object detection algorithm for UAV perspectives based on YOLOv8s

Aiming at the problems such as small target scale, occlusion and blurring existing in unmanned aerial vehicle (UAV) target detection, a lightweight improved algorithm based on the YOLOv8s model is proposed to improve the detection performance and efficiency. Firstly, the multi-branch C2f-DBB module is integrated into the YOLOv8s backbone network. The re-parameterization technique is utilized to enhance the diversity and robustness of feature extraction while maintaining the inference time. Secondly, upgrade the neck network to a weighted bidirectional feature pyramid network to achieve efficient cross-scale information interaction and enhance the detection capability. Finally, a lightweight shared convolution detection head is adopted to further reduce the model parameters and complexity. Experiments on the VisDrone2019 dataset have demonstrated the effectiveness of this method. Compared with YOLOv8s, the improved model improved accuracy by 1.8%, recall by 2.1%,[email protected] by 2.7%,[email protected] by 2.4%:0.95, while parameters and computational costs were reduced by 41.7% and 17.5%, respectively. The results show that this method achieves a dual improvement in model lightweighting and detection accuracy.