Scholarlink

Object detection in low-light environments presents significant challenges, as existing algorithms often struggle under such conditions. To address this, we propose ADHI-YOLO, a novel method optimized for low-light images, improving both detection accuracy and robustness. ADHI-YOLO integrates the large-kernel Attention Feature Pyramid Module (AFAM), an improved Dynamic Attention Head Module (DPH), and the innovative I-MPDIOU loss function, which effectively tackles image noise and detail loss. Additionally, mixed-precision training enhances efficiency. Experimental results on the ExDark dataset show that ADHI-YOLO achieves a mean Average Precision (mAP) of 74.9%, outperforming existing algorithms, making it a promising solution for low-light object detection.

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

ADHI-YOLO: Attentive SPPF with Dynamic Head Integration for Object Detection in Low-Light Environments

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yaoyang Zhang,
  • Songyang Li,
  • Ya Zhou,
  • Bohan Li,
  • Jianping Shuai

摘要

Object detection in low-light environments presents significant challenges, as existing algorithms often struggle under such conditions. To address this, we propose ADHI-YOLO, a novel method optimized for low-light images, improving both detection accuracy and robustness. ADHI-YOLO integrates the large-kernel Attention Feature Pyramid Module (AFAM), an improved Dynamic Attention Head Module (DPH), and the innovative I-MPDIOU loss function, which effectively tackles image noise and detail loss. Additionally, mixed-precision training enhances efficiency. Experimental results on the ExDark dataset show that ADHI-YOLO achieves a mean Average Precision (mAP) of 74.9%, outperforming existing algorithms, making it a promising solution for low-light object detection.