Small Target Detection Algorithm for UAV aerial images based on YOLOv8s-SSB

Aiming at the problems of low detection accuracy and missed or false detection of small targets in UAV (Unmanned Aerial Vehicle) aerial images by existing target detection methods, this paper proposes a small target detection method YOLOv8s-SSB for small UAV aerial images based on YOLOv8s from the perspective of feature extraction and feature fusion. Firstly, the SPD (space-to-depth) layer and the non-strided convolution layer are used to replace the pooling and step convolution in the original model to suppress the loss of fine-grained information in the feature extraction process. Secondly, a small target detection layer is added to make full use of the location information of shallow features and the semantic information of deep features in the feature fusion stage to improve the sensitivity of the model to small target objects. Finally, the dynamic sparse attention mechanism Bi-Level Routing Attention is introduced into the process of feature fusion, so that the network pays more attention to the features that are beneficial to learning and improves the detection ability of the model for dense small targets. Experiments on the public dataset VisDrone2019 show that the mAP@50 of YOLOv8s-SSB reaches 44.2%, which is 4.9% higher than that of the original YOLOv8s model.