UAV-DETR+: more tiny pixels and frequency information help small object detection

We present UAV-DETR+, an advanced framework for small object detection in unmanned aerial vehicle (UAV) imagery that addresses key limitations in existing methods. Our model integrates multiscale focus, wavelet transform down sampling, and SE+Cat with feature fusion convolution. This hybrid architecture preserves fine-grained details while optimizing frequency-domain extraction through the “WT + FT + WT” cascade. Experiments on VisDrone and UAVVaste datasets show that UAV-DETR+r18 achieves SOTA performance (AP50.1/AP76.6), with +1.3/+1.9 AP gains over UAV-DETR. Concurrently, the model’s transferability and restoration capability have been experimentally validated. The source code for this project is released on https://github.com/zepher-kk/UAV-DETRimp