DD-DETR: Deformable DETR with Dynamic Scale Attention for Infrared Aerial Small Object Detection

With the rapid development of the low-altitude economy and the turbulence in international relations, the threat of unconventional aircraft has intensified the challenges in airspace management. Accurate detection and identification of aircraft have become increasingly important. Infrared imaging technology, with its all-weather sensing advantage, has become the core method for aerial object detection. However, infrared objects in the air are difficult to detect due to characteristics such as small size, low signal-to-noise ratio, and thermal radiation, resulting in high false negative rates and low localization accuracy in existing algorithms. To address these issues, this paper proposes an improved Deformable DETR infrared detection model, which designs an adaptive multi-scale feature weighting module based on object size to enhance the weight of small objects in shallow features. Experimental results show that the proposed model improves the mAP@0.5:0.95 by 3% and a 2.3% gain in in mAP for small objects on a four-class infrared small aerial target dataset, demonstrating an improvement in detection accuracy.