Scholarlink

<p>Volatile organic compound (VOC) leakage during production is difficult to avoid, posing a series of safety hazards to the people. However, traditional gas leakage detection methods and mainstream optical gas imaging (OGI) techniques are unable to efficiently and accurately identify leaking gas clouds. To address these challenges, a novel infrared imaging gas leakage detection model named Gas Leak Detector (GLD) is proposed in this study. Firstly, the model incorporates an edge-aware multi-scale module (EAMS) in the backbone to enhance the extraction of high-frequency edge features from infrared images. Secondly, based on the channel-space attention mechanism, the superficial detail fusion module (SDFM) is designed to fuse image details further to improve detection accuracy. Thirdly, the group normalization (GN) technique is employed to optimize the performance of the detection head. In ablation experiments and horizontal comparisons on the created infrared image dataset of gas leakage containing multiple scenes, the average accuracy (mAP50) of the GLD surpasses that of other mainstream models. In practical applications, the GLD effectively addresses the issues of false and missed detections in the baseline model, demonstrating superior adaptability to complex scenarios. Additionally, when deployed on edge devices, the GLD achieves a better balance between frames per second (FPS) and detection accuracy to meet real-time inference requirements.</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

Infrared Gas Leak Detection via Edge-Aware and Frequency-Enhanced Lightweight Detector

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Yong Fang,
  • Ao Zhang,
  • Yanliang Li,
  • Jingjing Gao,
  • Zhaoqing Qie,
  • Juntao Hu

摘要

Volatile organic compound (VOC) leakage during production is difficult to avoid, posing a series of safety hazards to the people. However, traditional gas leakage detection methods and mainstream optical gas imaging (OGI) techniques are unable to efficiently and accurately identify leaking gas clouds. To address these challenges, a novel infrared imaging gas leakage detection model named Gas Leak Detector (GLD) is proposed in this study. Firstly, the model incorporates an edge-aware multi-scale module (EAMS) in the backbone to enhance the extraction of high-frequency edge features from infrared images. Secondly, based on the channel-space attention mechanism, the superficial detail fusion module (SDFM) is designed to fuse image details further to improve detection accuracy. Thirdly, the group normalization (GN) technique is employed to optimize the performance of the detection head. In ablation experiments and horizontal comparisons on the created infrared image dataset of gas leakage containing multiple scenes, the average accuracy (mAP50) of the GLD surpasses that of other mainstream models. In practical applications, the GLD effectively addresses the issues of false and missed detections in the baseline model, demonstrating superior adaptability to complex scenarios. Additionally, when deployed on edge devices, the GLD achieves a better balance between frames per second (FPS) and detection accuracy to meet real-time inference requirements.