基于改进的Swin Transformer的输油管道微小缺陷检测模型  

Detection Model of Oil Pipeline Tiny Defects Based on Improved Swin Transformer

作　　者：姚睿 郎宪明[1] 袁开欣 侍殿超 张欣冉 YAO Rui;LANG Xianming;YUAN Kaixin;SHI Dianchao;ZHANG Xinran(School of Information and Control Engineering,Liaoning Petrochemical University;School of Environmental and Safety Engineering,Liaoning Petrochemical University)

机构地区：[1]辽宁石油化工大学信息与控制工程学院 [2]辽宁石油化工大学环境与安全工程学院

出　　处：《管道技术与设备》2024年第3期35-42,47,共9页Pipeline Technique and Equipment

基　　金：国家自然科学基金项目(62073158);中国博士后科学基金资助项目(2020M660125);辽宁省博士科研启动基金计划资助项目(2019-BS-158);辽宁省教育厅一般项目(L2020017);辽宁石油化工大学引进人才科研启动基金资助(2019XHHL-008);辽宁省自然科学基金计划面上项目(2023-MS-289);辽宁省教育厅基本科研项目(JYTMS20231441)。

摘　　要：针对磁漏微小缺陷检测困难、检测数据不足、检测精度低等问题,提出了去噪扩散概率模型(DDPM)-门控并行膨胀卷积的移位窗口变换器(DGPST)。文中引入了DDPM作为数据生成模型,获得高质量的数据样本。门控并行膨胀卷积的移位窗口变换器(GPST)是在移位窗口变换器(Swin-Transformer)的基础上进行改进的。在GPST的主干网络中引入了门控并行膨胀卷积层,门控单元可以减少不必要的计算量,卷积提高了网络提取全局上下文信息的能力,有利于提高模型检测精度。实验结果表明:当IOU=0.5时,该算法的精确度为95.6%,推理时间(Latency)为8.2 ms。Aiming at the problems of detecting minor defects difficulty of magnetic leakage,insufficient detection data and low detection accuracy,the Swin Transformer of DDPM-gated parallel dilated convolution(DGPST)was proposed.This paper introduced a denoised diffusion probabilistic model(DDPM)data generation model to obtain high-quality data samples.The GPST was modified based on the Swin Transformer.The gated parallel dilated convolution layer was introduced into the backbone network of GPST,where the gating unit can reduce unnecessary computation.The convolution improved the ability of the network to extract global context information,which was beneficial to improve the accuracy of model detection.Experimental results show that the accuracy of this algorithm is 95.6%and Latency is 8.2 ms when IOU is 0.5.

关 键 词：门控并行膨胀卷积 概率模型 移位窗口变换器 磁通量 

分 类 号：TE8[石油与天然气工程—油气储运工程]