LS-YOLO: A Lightweight, Real-Time YOLO-Based Target Detection Algorithm for Autonomous Driving Under Adverse Environmental Conditions

Autonomous driving faces significant object detection challenges under complex backgrounds characterized by dense scenes, object occlusion, long-range targets, and extreme weather conditions. These challenges are further exacerbated in adverse weather such as rain, snow, and fog, leading to decreased detection accuracy and increased missed detection rates. To address these issues, a lightweight real-time object detection algorithm, LS-YOLO, is proposed. The LS-YOLO incorporates a MACA module to capture both global and local features, an SPDD module to reduce computational complexity, and a DR-Concat module to optimize feature fusion. Additionally, an improved ATFL-Wasserstein loss function is employed to enhance the learning capability for small objects and hard samples. Experimental results on public datasets demonstrate that LS-YOLO significantly outperforms existing algorithms in terms of accuracy, robustness, and real-time performance. Notably, under adverse weather and complex backgrounds, LS-YOLO achieves lower missed detection rates and higher object detection accuracy.