Scholarlink

<p>Traditional firefighting robots require operators to manually connect to a water source or rely on the robot’s onboard water supply. We plan to develop a robot capable of autonomously connecting to a fire hydrant to obtain a water source. This requires the implementation of computer vision technology to identify and locate the hydrant's outlet and valve. However, in fire scenarios, the presence of smoke and flames interferes with traditional computer vision algorithms, leading to suboptimal performance. We propose an improved model based on YOLOv8n, named YOLO-ESIDE. The key innovation is the introduction of the ESIDE module as a preprocessing component. The core idea of this module is to preserve edge features of the image using the Sobel operator, and then dynamically fuse these edge features with spatial features. Additionally, we have improved the C2f module based on the principles of this approach. Finally, we incorporate the LSKA attention mechanism to further enhance performance. Experiments demonstrate that YOLO-ESIDE ensures effective detection of fire hydrant outlets and valves in fire environments while maintaining low computational overhead. Compared to the YOLOv10 in the YOLO series, mAP50 increased by 1.24%, mAP50-95 improved by 2.19%, and accuracy enhanced by 2.24%.</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

YOLO-ESIDE: fire hydrant detection under fire environment

  • 期刊论文

  • 发表日期：

    2025年1月17日

查看全文
  • Haidong Xu

摘要

Traditional firefighting robots require operators to manually connect to a water source or rely on the robot’s onboard water supply. We plan to develop a robot capable of autonomously connecting to a fire hydrant to obtain a water source. This requires the implementation of computer vision technology to identify and locate the hydrant's outlet and valve. However, in fire scenarios, the presence of smoke and flames interferes with traditional computer vision algorithms, leading to suboptimal performance. We propose an improved model based on YOLOv8n, named YOLO-ESIDE. The key innovation is the introduction of the ESIDE module as a preprocessing component. The core idea of this module is to preserve edge features of the image using the Sobel operator, and then dynamically fuse these edge features with spatial features. Additionally, we have improved the C2f module based on the principles of this approach. Finally, we incorporate the LSKA attention mechanism to further enhance performance. Experiments demonstrate that YOLO-ESIDE ensures effective detection of fire hydrant outlets and valves in fire environments while maintaining low computational overhead. Compared to the YOLOv10 in the YOLO series, mAP50 increased by 1.24%, mAP50-95 improved by 2.19%, and accuracy enhanced by 2.24%.