首页|基于改进的Swin Transformer的输油管道微小缺陷检测模型

基于改进的Swin Transformer的输油管道微小缺陷检测模型

扫码查看
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-Trans-former)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.
Detection Model of Oil Pipeline Tiny Defects Based on Improved Swin Transformer
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.

gate parallel dilated convolutionprobabilistic modelsSwin Transformermagnetic flux

姚睿、郎宪明、袁开欣、侍殿超、张欣冉

展开 >

辽宁石油化工大学信息与控制工程学院

辽宁石油化工大学环境与安全工程学院

门控并行膨胀卷积 概率模型 移位窗口变换器 磁通量

国家自然科学基金中国博士后科学基金辽宁省博士科研启动基金辽宁省教育厅一般项目辽宁石油化工大学引进人才科研启动基金辽宁省自然科学基金面上项目辽宁省教育厅基本科研项目

620731582020M6601252019-BS-158L20200172019XHHL-0082023-MS-289JYTMS20231441

2024

管道技术与设备
沈阳仪表科学研究院

管道技术与设备

影响因子:0.504
ISSN:1004-9614
年,卷(期):2024.(3)
  • 1