Scholarlink

To address the issues of large model size, high parameter count, and excessive computational complexity in urban vehicle detection algorithms for complex traffic scenarios, this paper proposes a lightweight vehicle detection method based on an improved YOLOv8. The backbone network is optimized by replacing the original Conv, C2f, and SPPF modules with the lightweight PPLCNetPlus module. A Multi-Scale Channel Attention (MSCA) mechanism is added to enhance feature extraction capability. An improved Dual_C2f structure is introduced in the neck to further simplify the network. Additionally, a Lightweight Asymmetric Detection Head (LADH) is adopted to boost detection efficiency, and the Inner-EIoU loss function is used to improve bounding box regression. By effectively reducing model complexity while preserving detection accuracy, the proposed approach demonstrates strong applicability to real-world vehicle detection tasks in complex traffic scenarios.

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

Traffic Vehicle Detection Method Based on Lightweight Deep Learning Networks

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Shuaishuai Shao,
  • Huanyu Zhao,
  • Guanyu Zhang,
  • Shiyu Jiao,
  • Wei Liu

摘要

To address the issues of large model size, high parameter count, and excessive computational complexity in urban vehicle detection algorithms for complex traffic scenarios, this paper proposes a lightweight vehicle detection method based on an improved YOLOv8. The backbone network is optimized by replacing the original Conv, C2f, and SPPF modules with the lightweight PPLCNetPlus module. A Multi-Scale Channel Attention (MSCA) mechanism is added to enhance feature extraction capability. An improved Dual_C2f structure is introduced in the neck to further simplify the network. Additionally, a Lightweight Asymmetric Detection Head (LADH) is adopted to boost detection efficiency, and the Inner-EIoU loss function is used to improve bounding box regression. By effectively reducing model complexity while preserving detection accuracy, the proposed approach demonstrates strong applicability to real-world vehicle detection tasks in complex traffic scenarios.