Recently, unmanned aerial vehicles (UAVs) for aerial target detection has gained significant attention. UAV-based target detection faces challenges like small target regions, limited feature information, complex backgrounds, and frequent occlusions by other objects. Existing detection methods often fail to preserve crucial details in small targets, reducing detection accuracy. To address these issues, we propose the RWD-YOLO model, which enhances the accuracy of small target detection in aerial imagery, while effectively improving detection performance for dense and occluded objects. Specifically, we use RepVGG blocks for downsampling to enhance small target detection by preserving low-level details. We then implement a weighted cross-layer feature fusion module for better feature integration. For the detection head, we design a tiny target detection head and introduce DynamicHead, incorporating three attention mechanisms to improve detection across scales and complex backgrounds. Extensive experiments conducted on four datasets demonstrate the superiority of the proposed UAV-based object detection method, particularly for small target detection. Codes will be released upon publication.