YOLOv10-CS: A Model for Enhancing Recognition of Small Targets

Small target detection remains a challenging task for YOLOv10, primarily due to insufficient pixel coverage and degraded feature representation caused by low-resolution feature maps. To address these limitations, we propose YOLOv10-CS, an enhanced architecture with two key innovations: First, a dedicated small-target detection layer is added to the original neck structure, enabling fine-grained feature extraction through high-resolution feature maps; Second, a dual-attention enhanced C2F module integrates channel-spatial attention mechanisms, which selectively amplifies small-target features while suppressing background noise. Extensive experiments on VisDrone dataset demonstrate that YOLOv10-CS achieves 19.4% higher <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{AP}_{\text{small}}$</tex> (vs. baseline YOLOv10) while maintaining real-time performance (160 FPS on RTX 3090). The proposed modifications advance the state-of-the-art in small-target detection without compromising computational efficiency.