Abstract: In view of the large number of small targets in remote sensing images, the limited performance of existing detection algorithms, and the inherent difficulty of small-target detection, this study proposes a novel target detection algorithm based on spatial feature fusion down-sampling convolution. First, to mitigate the loss of small-target information caused by conventional down-sampling convolution modules, a spatial feature fusion down-sampling convolution module is designed to replace traditional down-sampling convolution, enabling the model to better focus on small targets during both training and inference. Second, a lightweight neck feature fusion module is introduced, incorporating a parameter-free attention mechanism to maintain detection accuracy while reducing model complexity and parameter count. Finally, WIoU is employed as a regression loss function to improve the accuracy of the regression frame. Experimental results on the DIOR remote sensing image dataset show that, compared with YOLOv8n, the proposed method achieves improvements of 2.3% in mAP50 and 2.1% in mAP50-90 while reducing the number of parameters by 0.12 M.