Scholarlink

<p>Low-light conditions significantly impede the performance of object detection systems, a challenge we address with En-YOLO, a novel framework integrating dual attention mechanisms with implicit feature learning. Here we show that En-YOLO substantially enhances detection accuracy under low-light conditions, offering a robust solution for applications in surveillance, autonomous driving, and beyond. En-YOLO employs a joint learning framework and utilizes end-to-end training strategies. Following the joint learning scheme, the object detection module can share the features obtained from ENet restoration for learning the potential features in the image and enhance the detection ability of the detector. Moreover, we designed a Dual Attention-guided Feature Filtering Module to enable En-YOLO to focus more on the exploitation and learning of critical features. Experiments are conducted on ExDark and MS-COCO datasets to show that our proposed method exhibits robust detection ability and performs well in different low-light as well as normal-light environments. Our code is available at <a href="https://github.com/Qibear/EnYOLO">https://github.com/Qibear/EnYOLO</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 with En-YOLO: leveraging dual attention and implicit feature learning

  • 期刊论文

  • 发表日期：

    2025年5月13日

查看全文
  • Xu Liu,
  • Chenhua Liu,
  • Xianye Zhou,
  • Guodong Fan

摘要

Low-light conditions significantly impede the performance of object detection systems, a challenge we address with En-YOLO, a novel framework integrating dual attention mechanisms with implicit feature learning. Here we show that En-YOLO substantially enhances detection accuracy under low-light conditions, offering a robust solution for applications in surveillance, autonomous driving, and beyond. En-YOLO employs a joint learning framework and utilizes end-to-end training strategies. Following the joint learning scheme, the object detection module can share the features obtained from ENet restoration for learning the potential features in the image and enhance the detection ability of the detector. Moreover, we designed a Dual Attention-guided Feature Filtering Module to enable En-YOLO to focus more on the exploitation and learning of critical features. Experiments are conducted on ExDark and MS-COCO datasets to show that our proposed method exhibits robust detection ability and performs well in different low-light as well as normal-light environments. Our code is available at https://github.com/Qibear/EnYOLO.