基于ResNet模型的聚乙烯燃气管道接头缺陷识别算法  

Defect recognition algorithm of polyethylene gas pipeline joint based on ResNet model

作　　者：凌晓[1] 程凌宇 郭凯[1] 杨凯 孙宝财 LING Xiao;CHENG Lingyu;GUO Kai;YANG Kai;SUN Baocai(College of Petroleum and Chemical Engineering,Lanzhou University of Technology,Lanzhou 730050,China;Gansu Province Special Equipment Inspection and Testing Institute,Lanzhou 730050,China)

机构地区：[1]兰州理工大学石油化工学院,兰州730050 [2]甘肃省特种设备检验检测研究院,兰州730050

出　　处：《压力容器》2023年第7期73-80,共8页Pressure Vessel Technology

基　　金：国家自然科学基金项目(52204074);甘肃省重点研发计划项目(22YF11GA316);甘肃省自然科学基金项目(21JR7RA221);甘肃省教育厅青年博士基金项目(2022QB-046)。

摘　　要：接头是聚乙烯燃气管道容易出现危害性缺陷的薄弱环节,需对此薄弱区域进行定期检测,以确保聚乙烯燃气管道的安全运行。为提高聚乙烯燃气管道接头缺陷图像识别能力,提出了一种基于ResNet网络模型的改进型卷积神经网络识别算法。首先运用Laplacian算子、中值滤波等方式实现对PE燃气管道接头缺陷图像的预处理;然后,将dropout层和ELU函数加入在ResNet34网络模型中完成图像识别模型的构建;最后,采用改进的ResNet34网络模型通过试验对包含6种热熔缺陷类型的数据集进行训练和测试。试验结果表明,改进后的ResNet34网络模型对缺陷图像的训练正确率可达到97.3%,且拥有比原始的ResNet34网络模型和DenseNet网络模型更高的正确率,验证了此模型对于热熔接头缺陷图像识别的有效性。Joint is a weak link prone to harmful defects which are easy to exist on polyethylene gas pipeline.Therefore,it is necessary to carry out regular testing in this weak area to ensure the safe operation of polyethylene gas pipelines.In order to improve the image recognition ability of polyethylene gas pipeline joint defects,an improved convolutional neural network recognition algorithm based on the ResNet network model was proposed.Firstly,the method of the Laplacian operator and median filter was used to preprocess the defect image of the PE gas pipeline joint;The dropout layer and the ELU function were then added to the ResNet34 network model to build the image recognition model.Finally,an improved ResNet34 network model was used to train and test the data set containing six types of hot melt defects through experiments.The experimental results show that the improved ResNet34 network model can achieve 97.3%accuracy in training defect images,and has a higher accuracy than the original ResNet34 network model and DenseNet network model,which verifies the effectiveness of this model for the recognition of hot-melt joint defect images.

关 键 词：聚乙烯燃气管道 缺陷检测 图像识别 卷积神经网络 ResNet34模型 

分 类 号：TH49[机械工程—机械制造及自动化] TQ055.8[化学工程] TP183[自动化与计算机技术—控制理论与控制工程]