基于多尺度特征融合和SAM引导的无人机小尺度目标检测OA

In UAV aerial images,the target objects to be detected are often only dozens of pixels in size due to long shooting distances and low target occupancy ratios,resulting in severe feature scarcity and a significant degradation in small object detection performance.Existing approa-ches primarily fall into two categories:sample augmentation and multi-scale perception.The former tends to introduce semantic conflicts in dense aerial scenarios,while the latter remains inadequate in deep feature perception and global modeling.To address these limitations,this paper proposes a small object detection network based on multi-scale feature fusion and SAM-guided learning.Specifically,we design a multi-scale architecture incorporating dedicated detection layers for small objects to enhance feature representation;integrate dilated convolutions with Transformers to enlarge the receptive field and model long-range dependencies;and leverage the prior knowledge of the Segment Anything Mod-el(SAM)foundation model to guide network training,thereby improving the extraction of discriminative features for small objects.Experimen-tal results demonstrate that our method significantly improves small object detection accuracy on the VisDrone-DET2019 benchmark.