<p>In UAV scenes, dense small objects often suffer from heavy occlusion, invisible keypoints, and scale variation, which hinder recognition performance. To address these challenges, we propose a novel approach for dense small object detection that leverages multi-scale fusion and context information enhancement. Firstly, we propose the enhanced spatial pyramid pooling-fast (ESPPF) module to integrate multi-scale feature information, enabling the network to better understand and classify densely arranged objects. Secondly, we design an adaptive scale feature fusion (ASFF) module that adjusts feature responses in each channel using global semantic and local detail information. Additionally, we propose a context information enhancement module (CIEM) at the head of the region proposal network (RPN) to capture long-distance dependencies, improving the model’s ability to understand relationships and contextual meanings within the scene. Finally, we replace ROI Pooling with ROI Align to preserve important details for small objects. Experimental results demonstrate that our method significantly enhances detection accuracy, achieving a 3.1% increase on the Pascal VOC2012 dataset, a 6.4% increase on the VisDrone2019 dataset and a 8.4% increase on DOTAv1.0 dataset, effectively improving the detection of dense small objects.</p>