煤矿井红外热成像中的多尺度融合目标特征检测  

Multi-scale Fusion Target Feature Detection in Infrared Thermal Imaging of Coal Mines

作　　者：陈湘源 牛青 CHEN Xiangyuan;NIU Qing(Guoneng Yulin Energy Co.,Ltd.,Yulin,Shaanxi 719000,China)

机构地区：[1]国能榆林能源有限责任公司,陕西榆林719000

出　　处：《计算技术与自动化》2025年第1期46-52,共7页Computing Technology and Automation

摘　　要：煤矿井内环境复杂,存在较多的危险因素,单一尺度的特征提取很难实现精准检测目标,无法充分保障工作人员的安全。为此,提出了煤矿井红外热成像中的多尺度融合目标特征检测。通过基于自适应邻域的噪声抑制方法对图像展开去噪处理,并利用线性和非线性灰度变换方法增强红外热成像。采用特征金字塔展开多尺度融合目标特征提取,利用多尺度融合技术,结合YOLOv5网络预测损失函数,实现煤矿井红外热成像目标检测。实验结果表明,所提算法的煤矿井红外热成像效果较好,能够全部检测到红外热成像目标特征,同时P-R曲线图的AP面积较大,说明所提算法提高了目标特征检测的精度、效率和可靠性。The environment in coal mine is complex and there are many risk factors.It is difficult to achieve the accurate detection target by single scale feature extraction,and the safety of staff cannot be fully guaranteed.Therefore,multi-scale fusion target feature detection in infrared thermal imaging of coal mines is proposed.By using adaptive neighborhood based noise suppression methods to denoise the image,and utilizing linear and nonlinear grayscale transformation methods to enhance infrared thermal imaging.The multi-scale fusion target feature extraction is carried out by feature pyramid,and the multi-scale fusion technology is used to predict the loss function combined with YOLOv5 network to realize the target detection of infrared thermal imaging in coal mines.The experimental results show that the proposed algorithm performs well in infrared thermal imaging of coal mines,and can detect all infrared thermal imaging target features.At the same time,the AP area of the P-R curve is larger,indicating that the proposed algorithm improves the accuracy,efficiency,and reliability of target feature detection.

关 键 词：多尺度融合 红外热成像 目标特征检测 特征金字塔 煤矿井 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]