UAV small target detection model based on an improved YOLOv8s

In UAV target detection, the main challenge is usually detecting small targets in complex backgrounds. This paper proposed an improved YOLOv8s model to solve the problem of low detection performance and frequent missed detections in UAV small target detection. The improved model integrates the C2f_UniRepLKNetBlock module to enhance context feature extraction capabilities while reducing the number of model parameters. In addition, the SPPF_C2fUniRep structure is used to aggregate feature information of multiple observation views, and the WIoU loss function is applied in the YOLOv8 loss function to balance the differences in center offset and aspect ratio. The improved model was evaluated on the VisDrone2019 public data set, and compared with the baseline model, it improved mAP50 by 6.0% and mAP95 by 4.0%, Effectively improves the performance of UAVs in small target detection tasks.