TARG-YOLO: an efficient small target detection framework for UAV

In recent years, the rapid proliferation of small unmanned aerial vehicles (UAVs) has transformed various industries, making accurate and efficient detection of small-target UAVs increasingly critical for security monitoring, environmental protection, and traffic management. Detecting these small UAVs poses significant challenges due to their agile flight patterns, small visual footprint, and the complex environments in which they operate. To address these challenges, this paper presents TARG-YOLO, an innovative detection framework based on the YOLOv9 architecture, specifically tailored for small UAV detection. TARG-YOLO integrates several novel components, including the RepMS module that enhances feature extraction by dynamically adjusting focus areas, and the DDetHead detection head designed to minimize background redundancy, thereby improving detection accuracy of small UAVs. Moreover, the TARG-IoU loss function is introduced, which optimally balances localization precision and target sensitivity, catering specifically to the requirements of small UAV detection. Experimental results on the DUT Anti-UAV dataset demonstrate that TARG-YOLO achieves a mean average precision (AP $$^{val}_{50}$$ ) of 89.93%, surpassing the baseline YOLOv9 by 4.22% and the latest YOLOv12 by 3.54%. The source code is available at https://github.com/hkhldzf/TARG-YOLO .