基于改进PSPnet-MobileNetV2的煤岩界面快速精准识别  

Rapid and Accurate Identification of Coal⁃Rock Interface Based on Improved PSPnet⁃MobileNetV2

作　　者：王海舰 刘丽丽 赵雪梅 张强 WANG Haijian;LIU Lili;ZHAO Xuemei;ZHANG Qiang(School of Mechanical and Electrical Engineering,Guilin University of Electronic Technology Guilin,541004,China;School of Electronic Engineering and Automation,Guilin University of Electronic Technology Guilin,541004,China;College of Mechanical and Electronic Engineering,Shandong University of Science and Technology Qingdao,266590,China)

机构地区：[1]桂林电子科技大学机电工程学院,桂林541004 [2]桂林电子科技大学电子工程与自动化学院,桂林541004 [3]山东科技大学机械电子工程学院,青岛266590

出　　处：《振动．测试与诊断》2024年第4期793-800,832,833,共10页Journal of Vibration,Measurement & Diagnosis

基　　金：国家自然科学基金青年基金资助项目(52204130);广西省自然科学基金青年基金资助项目(2022GXNSFBA035599);桂林电子科技大学研究生教育创新计划资助项目(2022YCXS023)。

摘　　要：针对短时间主动热激励作用下煤岩介质表征差异不明显,不易快速、准确识别煤岩界面的难题,提出一种基于改进金字塔场景解析网络(pyramid scene parsing network,简称PSPnet)模型-MobileNetV2的煤岩界面快速精准识别方法。通过搭建煤岩主动红外试验平台,采集并获取短时主动热激励作用下的煤岩界面红外热图像,构建了煤岩红外图像数据集;对传统PSPnet模型进行改进,采用轻量级网络模型MobileNetV2作为主干网络提取特征,大幅降低了网络模型所占内存和训练时间,同时将注意力机制模块(convolutional block attention module,简称CBAM)与金字塔场景解析(pyramid scene parsing,简称PSP)模块的上采样特征层和PSPnet网络模型的浅层特征层进行融合,有效提升模型对特征的细化能力。试验结果表明:基于改进的PSPnet-MobileNetV2网络模型所占内存仅为9.12 MB,较原始PSPnet模型减少了94.88%;煤和岩的交并比为96.52%和96.87%,分别提升了8.29%和7.7%;像素准确度分别为97.25%和99.15%,较原始网络模型分别提升了7.32%和1.64%;测试时间降低了53.70%。该方法为煤岩界面的快速和预先精准识别提供了一种有效技术手段。Aiming to address the issue that differences of coal-rock medium characterization are not obvious under the action of short-time active thermal excitation,and difficult to quickly and accurately identify the coal-rock interface,a fast and accurate identification method for the coal-rock interface based on the improved PSPnet-MobileNetV2 is proposed.By constructing a coal-rock active infrared testbed,the infrared thermal images of the coal-rock interface under the action of short-term active thermal excitation are collected,and a coal-rock infrared image dataset is constructed.Based on the improvement of the traditional pyramid scene parsing network(PSPnet),the lightweight network model MobileNetV2 is used as the backbone network to extract features,which greatly reduces the memory and training time required by the network model.The convolutional block attention module(CBAM)is added to the upsampling feature layer of the PSP module and the shallow feature layer of the PSPnet to effectively improve the model′s feature refinement capability.The experimental results show that the improved PSPnet-MobileNetV2 network model only requires 9.12 MB of memory,which is a 94.88%reduction compared to the original PSPnet.The intersection-over-union of coal and rock is 96.52%and 96.87%,respective increases of 8.29%and 7.7%over the original values;The pixel accuracy is 97.25%and 99.15%respectively,which are 7.32%and 1.64%higher than the original network model respectively;Furthermore,the test time is reduced by 53.70%,which provides an effective method for the rapid and accurate identification of coal-rock interface.

关 键 词：煤岩识别 主动红外激励 金字塔场景解析网络(PSPnet) MobileNetV2 注意力机制模块 

分 类 号：TD823[矿业工程—煤矿开采] TH744[矿业工程—矿山开采]