LSD-YOLO:Enhanced Small Object Detection via Spatially Adaptive Feature Modulation and Large Separable Kernel Attention

In order to solve the difficulties of complex background, small target and low detection efficiency in remote sensing images, an improved LSD-YOLO11 method is proposed in this paper. The method introduces DSCWWTC, LSKA and SAFM modules, and the model enhances the feature expression capability while reducing the computation amount, focuses on the key region adaptively through the dynamic attention mechanism to alleviate the problem of small-target submergence under the complex background, strengthens the synergy between shallow and deep features, and introduces the spatial decoupling downward sampling strategy to maximize the retention of spatial information of small targets. With the lightweight design, the improved model achieves a better balance between speed and accuracy. The experimental results show that compared with the baseline model, the mAP@50 of the improved model on the NWPU VHR-10 dataset is improved by 1.9%, which verifies the effectiveness and applicability of LSD-YOLO.