L-YOLO: A Small Object Detection Algorithm for Autonomous Driving Using Multiple Attention Mechanisms

Object detection has become increasingly important in the field of autonomous driving, but the effectiveness of small object detection has always faced challenges such as insufficient feature representation and background noise interference. To address this problem, this paper proposes an L-YOLO detector designed for small object detection in driving scenarios. The detector makes several improvements on the basis of the latest YOLOv11: the Swin Transformer module is introduced into the Backbone network to enhance the local feature extraction capability; the MSCA (Multi-Scale Convolutional Attention) module is integrated into the Neck part to improve multi-scale feature fusion; a detection head is incorporated into the Head part to strengthen the utilization of shallow small object features; meanwhile, the SPPFCSPC (Cross-Stage Partial Fast Spatial Pyramid Pooling) module is introduced to the Backbone part to reduce the computational redundancy and improve the model efficiency. The experimental results show that compared with YOLOv11, the L-YOLO algorithm improves the mAP0.5 values by 4.3% and 3.9% on the SODA10M and SODA_D datasets, respectively. The experimental results on KITTI illustrate that improvements for small objects do not affect other large objects. Furthermore, the increase in parameters and decrease in FPS remain within reasonable limits. Compared to other detectors, L-YOLO strikes a good balance between detection accuracy and computational efficiency, making it well-suited for small object detection in autonomous driving applications.