Dsca-Yolo11: an Enhanced Algorithm for Small Target Detection in Drone-Captured Imagery

Target detection in drone-captured aerial imagery is crucial for applications such as disaster response and intelligent surveillance. However, this task remains highly challenging due to complex backgrounds, significant scale variations, and severe performance degradation for small objects. To address these challenges, this paper proposes DSCA-YOLO11, a lightweight and robust model featuring two key innovations: (1) a Dynamic Synergistic Contextual Attention (DSCA) module that integrates dynamic pooling operations with multi-scale dilated convolutions and a coordinated channel-spatial attention mechanism, significantly enhancing feature representation capabilities against cluttered backgrounds; and (2) an optimized detection architecture that replaces the P5 layer with a high-resolution P2 layer to amplify small-target feature extraction while reducing computational redundancy. Evaluated on the VisDrone2019 benchmark, DSCA-YOLO11 achieves a 4.58 % improvement in <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{mAP} {@} 50$</tex> and a 3.13 % gain in mAP@ <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$50: 95$</tex> compared to the baseline YOLO11n. This efficient and effective design balances inference speed with accuracy, enabling its real-time deployment in critical scenarios and advancing the practical utility of intelligent drone applications.