Scholarlink

<p>Object detection, a core task in computer vision, aims to identify and locate objects in images or videos. Convolutional neural network (CNN)-based methods, such as Faster R-CNN, YOLO, and RetinaNet, have achieved excellent performance under ideal conditions but suffer significant performance degradation in low-light environments. This paper proposes a domain adaptation-based method to enhance low-light object detection by reducing the data distribution difference between normal and low-light images. Image enhancement is performed by decomposing the image using the Laplacian pyramid, designing a global enhancement module for low-frequency components, and optimizing high-frequency components through a cross-layer guidance module. In the joint training stage, a multi-scale feature alignment module based on adversarial training is incorporated into the student model to learn domain-invariant features. Extensive experiments on multiple publicly available datasets demonstrate that our proposed algorithm outperforms the state-of-the-art approach by 1% on the BDD100K dataset and by 1.8% on the SHIFT dataset. This study provides an efficient and robust solution for object detection in low-light environments, with significant application value in scenarios such as autonomous driving and video surveillance. The code can be found at <a href="https://github.com/yanguo-lucky/ELOD">https://github.com/yanguo-lucky/ELOD</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing low-light object detection through domain adaptation and image enhancement

  • 期刊论文

  • 发表日期：

    2025年8月20日

查看全文
  • Gaihua Wang,
  • Yan Guo,
  • Tianshun Hu

摘要

Object detection, a core task in computer vision, aims to identify and locate objects in images or videos. Convolutional neural network (CNN)-based methods, such as Faster R-CNN, YOLO, and RetinaNet, have achieved excellent performance under ideal conditions but suffer significant performance degradation in low-light environments. This paper proposes a domain adaptation-based method to enhance low-light object detection by reducing the data distribution difference between normal and low-light images. Image enhancement is performed by decomposing the image using the Laplacian pyramid, designing a global enhancement module for low-frequency components, and optimizing high-frequency components through a cross-layer guidance module. In the joint training stage, a multi-scale feature alignment module based on adversarial training is incorporated into the student model to learn domain-invariant features. Extensive experiments on multiple publicly available datasets demonstrate that our proposed algorithm outperforms the state-of-the-art approach by 1% on the BDD100K dataset and by 1.8% on the SHIFT dataset. This study provides an efficient and robust solution for object detection in low-light environments, with significant application value in scenarios such as autonomous driving and video surveillance. The code can be found at https://github.com/yanguo-lucky/ELOD.