Robust Object Detection for Autonomous Driving in Adverse Weather Conditions With Multi-Scale Feature Enhancement
IEEE Open Journal of Intelligent Transportation Systems · 2025 · 6 · 1459-1475
DOI: 10.1109/ojits.2025.3631716 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Open Journal of Intelligent Transportation Systems, 2025. Verfasst von Weiguo Yi und 1 weiteren Autor:innen. Die Arbeit wurde bisher 7-mal zitiert (Crossref) und nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Educational Department of Liaoning Province, Liaoning Provincial Department of Transportation Science and Technology Program.
Abstract (Originalsprache)
In autonomous driving under complex weather conditions, issues such as low detection accuracy and poor model generalization persist due to unstable illumination, target blurring, image occlusion, and enhanced background noise. To address these challenges, this paper proposes an improved multi-scale object detection algorithm for autonomous driving based on the YOLO11s model. Initially, a small-object detection layer is added to the neck structure to enhance detection accuracy for minor targets and reduce the misdetection rate of distant vehicles during autonomous driving. Second, the LightSDI lightweight spatial-dimensional interaction module is introduced to optimize the original feature fusion layers, improving the model’s detection accuracy in harsh environments and strengthening its semantic perception capability. Third, DK_SCDown replaces the conventional downsampling convolution (Conv) to leverage multi-scale feature extraction and dynamic weighted fusion, thereby reinforcing receptive field coverage and dynamic feature selection while achieving parameter lightweighting through depthwise separable convolution. Finally, the Wise-IoU v2 loss function, more suitable for object detection tasks in complex environments, is adopted to mitigate issues like target boundary ambiguity and severe occlusion; it normalizes geometric distances between targets via center-point normalization to guide more rational regression. On the public SODA10M dataset, mAP@0.5 improves by 7.0% compared to the baseline model, while the parameter count decreases by 19%. Moreover, excellent performance on additional public datasets, including BDD100K, KITTI, and DAWN, further demonstrates the generalization capability of the improved model.
Förderung
- Educational Department of Liaoning Province
- Liaoning Provincial Department of Transportation Science and Technology Program
Zitieren
APA
Yi, W., & Wang, L. (2025). Robust Object Detection for Autonomous Driving in Adverse Weather Conditions With Multi-Scale Feature Enhancement. IEEE Open Journal of Intelligent Transportation Systems, 6, 1459-1475. https://doi.org/10.1109/ojits.2025.3631716
Vancouver
Yi W, Wang L. Robust Object Detection for Autonomous Driving in Adverse Weather Conditions With Multi-Scale Feature Enhancement. IEEE Open Journal of Intelligent Transportation Systems. 2025;6:1459-1475. doi:10.1109/ojits.2025.3631716
BibTeX
@article{Yi2025,
author = {Yi, Weiguo and Wang, Longteng},
title = {Robust Object Detection for Autonomous Driving in Adverse Weather Conditions With Multi-Scale Feature Enhancement},
journal = {IEEE Open Journal of Intelligent Transportation Systems},
year = {2025},
volume = {6},
pages = {1459--1475},
doi = {10.1109/ojits.2025.3631716}
}
EUVIMED