DFF-Yolo11n: a lightweight model for efficient vehicle detection

Abstract This study proposes DFF-Yolo11n, a lightweight and high-performance model for road vehicle detection. It aims to improve detection accuracy and inference speed while maintaining model efficiency. First, the enhanced C3k2-ADBB module is integrated into Yolo11n to better detect small and distant objects. Second, the ASDF module refines multi-scale features through cross-scale fusion and alignment, improving robustness and accuracy. Finally, a novel Focal-AEIOU loss is introduced, combining dynamic weight α and Focal Loss to address sample imbalance and improve localization. Experimental results show a 5.8% increase in detection accuracy over the baseline, with an inference speed of 88.5 FPS, demonstrating strong performance with minimal computational cost.