YOLOv11-SV: Small Target Detection in Visible Light via Multi-Scale Feature Extraction and Frequency domain enhancement

Small target detection in visible light plays a crucial role in applications such as traffic monitoring, security surveillance, and autonomous driving. However, traditional YOLO models often struggle with feature loss, high false detection rates, and unstable detection performance in low-light, complex backgrounds, and low-contrast environments. To address these challenges, this paper proposes an improved model, YOLOv11-SV, which enhances small target detection accuracy and stability by optimizing multi-scale feature extraction and high-frequency information retention strategies. Specifically, our model designs two key modules. The first module replaces the feature extraction component in the YOLOv11 backbone with a wavelet transform-based structure, which Enhance high-frequency information. The second module proposes an advanced feature aggregation framework to perform multi-scale pooling, enhancing feature representation across different scales. Experimental results demonstrate that YOLOv11-SV outperforms other YOLO models in key metrics such as mAP, precision, and inference speed, particularly excelling in low-contrast environments.