Journal of South China University of Technology(Natural Science Edition)

• Special Topic on Digital-Intelligent Transportation • Previous Articles     Next Articles

Vehicle Detection Algorithm for Foggy Highways Based on Detail Enhancement and Feature Preservation

TANG Yonghua1 DUAN Xiaoteng1 LIN Sen2 BAI Tianyu1 LIU Hongpu1   

  1. 1．School of Information Science and Engineering，Shenyang University of Technology/Key Laboratory of Machine Vision of Liaoning Province，Shenyang 110870，Liaoning，China；

    2．School of Automation and Electrical Engineering，Shenyang Ligong University，Shenyang 110159，Liaoning，China

  • Published:2026-03-06

Abstract:

To address the issues of low recognition accuracy in intelligent vehicle environmental perception under foggy highway conditions, along with frequent missed detections of multi-scale and distant small targets, a foggy highway vehicle detection algorithm based on detail enhancement and feature preservation is proposed. Firstly, the DEA-Net framework is adopted for image defogging preprocessing. A Gated Adaptive Fusion (GAF) module is incorporated into the Content-Guided Attention (CGA) mechanism, where a gated generator dynamically weights channel and spatial attention. This enhancement achieves precise defogging and strengthens vehicle texture details, effectively improving the quality of defogged images. Secondly, in the object detection stage, a Split Dual-Path Downsampling (SDPDown) module is introduced into the backbone and neck networks of YOLOv13n, enabling the collaborative preservation of both local contour details and global contextual features of vehicles. Finally, the Detail Enhancement Highway Vehicle Detection Head (DEHV) is introduced to capture the multi-scale edge and texture details of vehicles, improve the recognition accuracy of vehicle targets in complex traffic scenarios, and the improved model is named SD-YOLO. The results show that compared with the original algorithm, the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) of the improved DEA-Net algorithm are increased by 0.71dB and 0.46 percentage points respectively. Compared with the baseline method YOLOv13n, SD-YOLO has improved the precision rate, recall rate and mAP@0.5 by 2.7, 7.8 and 5.0 percentage points respectively, reduced the parameter quantity and floating-point calculation amount to 1.9M and 5.5G respectively, and detected 127 frames of images per second. The experimental results show that the proposed algorithm is superior to the current mainstream algorithms in terms of vehicle detection accuracy in foggy weather, providing a reliable vehicle detection technical solution for intelligent vehicle environmental perception on foggy expressways.

Key words: intelligent vehicles, environment awareness, highways, DEA-Net, YOLOv13