MSAL-YOLO: a YOLOv8-based detector for small and densely distributed object detection in UAV aerial imagery

Object detection in UAV aerial imagery is challenging because of small object scales, dense distributions, occlusions, and complex backgrounds. To address these issues, we propose MSAL-YOLO, an improved YOLOv8-based detector for aerial object detection. Specifically, a Spatial–Channel Mixed Convolution (SCMConv) module is introduced to enhance local feature extraction by combining standard and dilated convolutions. A Multi-Branch Enhanced Coordinate Attention (MBECA) mechanism is further designed to strengthen spatial dependency modeling and contextual interaction. In addition, a Region-Adaptive Wise-IoU (RA-WIoU) loss is developed to alleviate scale imbalance by increasing the contribution of small objects during training. Experiments on the VisDrone2019 dataset show that MSAL-YOLO improves mAP@0.5 and mAP@0.5:0.95 from 30.0% and 17.0% to 35.7% and 20.5%, respectively, while maintaining competitive model complexity. Cross-dataset experiments on DOTAv1 further demonstrate the robustness and generalization capability of the proposed method. These results indicate that MSAL-YOLO provides an effective solution for UAV aerial object detection in complex scenes.