PDA-DETR: Progressive Dimension Adaptation for UAV-Based Infrared Small Target Detection

Object detection in UAV-acquired infrared imagery faces significant challenges due to complex urban environments, low signal-to-noise ratio caused by external light interference, and computational constraints of edge devices. To address these fundamental limitations, this work introduces PDA-DETR (Progressive Dimension Adaptation DETR), a novel lightweight detection framework for drone-based remote sensing. The proposed architecture synthesizes three core innovations: (1) A Cross Stage Partial-Polynomial Kernel Inception (CSP-PKI) module that reduces parameters by 35% while improving precision by 11.01%; (2) A Dynamic Attention mechanism optimized for small target detection; and (3) A Dynamic Sample Attentional Scale Sequence Fusion (DS-ASF) module for multi-scale feature integration. Comprehensive evaluation on the HIT-UAV benchmark demonstrates that PDA-DETR significantly outperforms existing approaches, achieving improvements of 12.39% in AP<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0.5:0.95</inf> and 17.03% in AP<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</inf>.<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">75</inf>while reducing computational overhead by 15.93% in parameters. The results demonstrate the effectiveness of our approach for efficient small target detection in UAV applications.