MSCSM-YOLOv11s: multi-scale feature extraction and CSM attention for small target detection

Small‑target detection in aerial images remains challenging owing to low resolution, extremely small pixel footprints, large scale variation, occlusion, complex backgrounds, and the loss of fine‑grained features resulting from repeated downsampling. To address these issues, MSCSM‑YOLOv11s is proposed-an improved YOLOv11s‑based detection framework specifically designed for small‑target detection in UAV imagery. The proposed model comprises three main components. First, a Multi‑Scale Dilated Convolution (MSDC) module is developed to enlarge the receptive field and enhance multi‑scale feature extraction while preserving local details of small objects. Second, a Channel‑Spatial Module (CSM) is incorporated to strengthen discriminative feature representation by jointly modeling channel‑wise and spatial information, thereby enhancing detection robustness in dense and partially occluded scenes. Third, an additional small‑object detection layer is added to preserve shallow spatial features and to reduce information loss during downsampling. Experiments were conducted on the VisDrone‑2019 validation set. The results demonstrate that MSCSM‑YOLOv11s achieves a precision of 56.1%, a recall of 42.7%, an mAP@0.5 of 45.5%, and an mAP@0.5:0.95 of 27.8%, outperforming the baseline YOLOv11s by 6.8, 5.1, 7.0, and 5.0% points, respectively. Ablation studies further demonstrate the complementary contributions of the MSDC, the CSM, and the small‑object detection layer. The proposed model thus achieves a more competitive small‑target detection performance while maintaining acceptable computational efficiency.