Scholarlink

<p>Illegal emission of black smoke from trucks is the focus and difficulty of road traffic environmental protection law enforcement. In order to solve the problem of insufficient accuracy and speed of black smoke detection in complex environments, we propose a lightweight real-time black smoke detection model EC-RTDETR based on edge intelligence. EC-RTDETR is deployed on intelligent hardware. It realizes real-time detection of illegal emission of black smoke from trucks in complex traffic environments. Firstly, a multi-scale feature extraction module is designed in backbone by using different types of convolutional modules to reduce information loss during feature extraction and memory access during inference. Secondly, we introduce an improved multipath coordinate attention mechanism. It enhances the model feature fusion by utilizing the information of spatial and channel dimensions. Finally, minimum point distance intersection over union is used to replace complete IoU to optimize the model loss function, which reduced the redundancy of the predicted bounding box. The mean average precision achieves 94.6%. It is improved by 2.8% compared with RTDETR. We have deployed the model at the detection points in Xuchang City, Henan Province. The results show that EC-RTDETR meets the requirement of real-time detection. The results are also transmitted to the cloud directly. This facilitates the rapid enforcement by the environmental protection department.</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

Traffic environmental protection edge computing: a monitoring algorithm and system of truck black smoke emission in complex scene

  • 期刊论文

  • 发表日期：

    2025年4月19日

查看全文
  • Kang Yang,
  • Lili Zhang,
  • Yucheng Han,
  • Ke Zhang,
  • Jing Li,
  • Wei Wei,
  • Hongxin Tan,
  • Pei Yu

摘要

Illegal emission of black smoke from trucks is the focus and difficulty of road traffic environmental protection law enforcement. In order to solve the problem of insufficient accuracy and speed of black smoke detection in complex environments, we propose a lightweight real-time black smoke detection model EC-RTDETR based on edge intelligence. EC-RTDETR is deployed on intelligent hardware. It realizes real-time detection of illegal emission of black smoke from trucks in complex traffic environments. Firstly, a multi-scale feature extraction module is designed in backbone by using different types of convolutional modules to reduce information loss during feature extraction and memory access during inference. Secondly, we introduce an improved multipath coordinate attention mechanism. It enhances the model feature fusion by utilizing the information of spatial and channel dimensions. Finally, minimum point distance intersection over union is used to replace complete IoU to optimize the model loss function, which reduced the redundancy of the predicted bounding box. The mean average precision achieves 94.6%. It is improved by 2.8% compared with RTDETR. We have deployed the model at the detection points in Xuchang City, Henan Province. The results show that EC-RTDETR meets the requirement of real-time detection. The results are also transmitted to the cloud directly. This facilitates the rapid enforcement by the environmental protection department.