Scholarlink

<p>In order to solve the problems of multiple parameters and large volume of existing safety hat wearing state detection algorithms in complex construction scenarios, we propose a GE-YOLO model based on YOLOv5s. Firstly, the Ghost convolution is used in place of the traditional convolution to reduce the computational cost of the model. Secondly, by integrating CSPNet and GhostNet, a more efficient feature extraction module, CGMS, is designed using the idea of gradient shunt to reduce the parameters and complexity of the network. Additionally, the SENet attention mechanism enables the network to learn and utilize the relationships between feature channels more efficiently, thereby improving the model performance. At last, EIoU loss function is used to replace CIoU loss function to calculate the position loss, which improves the convergence speed and positioning effect of the model. Experimental results validate the effectiveness of the proposed model. The improved algorithm achieves only 38.6% of the parameter count and 36.7% of the GFLOPs of the YOLOv5s model, effectively balancing detection accuracy and computational cost. Furthermore, the model exhibits a fast inference speed, demonstrating significant potential for applications in complex construction scenarios.</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

GE-YOLO: a lightweight object detection method for helmet wearing state

  • 期刊论文

  • 发表日期：

    2025年3月19日

查看全文
  • Chenxue Zhang,
  • Yanping Du,
  • Xianghui Xiao,
  • Yang Gu

摘要

In order to solve the problems of multiple parameters and large volume of existing safety hat wearing state detection algorithms in complex construction scenarios, we propose a GE-YOLO model based on YOLOv5s. Firstly, the Ghost convolution is used in place of the traditional convolution to reduce the computational cost of the model. Secondly, by integrating CSPNet and GhostNet, a more efficient feature extraction module, CGMS, is designed using the idea of gradient shunt to reduce the parameters and complexity of the network. Additionally, the SENet attention mechanism enables the network to learn and utilize the relationships between feature channels more efficiently, thereby improving the model performance. At last, EIoU loss function is used to replace CIoU loss function to calculate the position loss, which improves the convergence speed and positioning effect of the model. Experimental results validate the effectiveness of the proposed model. The improved algorithm achieves only 38.6% of the parameter count and 36.7% of the GFLOPs of the YOLOv5s model, effectively balancing detection accuracy and computational cost. Furthermore, the model exhibits a fast inference speed, demonstrating significant potential for applications in complex construction scenarios.