深度残差收缩网络的含噪微泄漏超声识别方法  

An ultrasonic identification method of noised micro-leakage based on deep residual shrinkage networks

作　　者：孟庆旭 沈功田[2] 俞跃[2] 胡斌[2] 王宝轩[2] 李志农[1] MENG Qingxu;SHEN Gongtian;YU Yue;HU Bin;WANG Baoxuan;LI Zhinong(School of Measuring and Optical Engineering,Nanchang Hangkong University,Nanchang 330036,China;China Special Equipment Inspection and Research Institute,Beijing 100029,China)

机构地区：[1]南昌航空大学测试与光电工程学院,南昌330036 [2]中国特种设备检测研究院,北京100029

出　　处：《应用声学》2022年第6期964-972,共9页Journal of Applied Acoustics

基　　金：国家重点研发计划项目(2019YFB1310701),国家自然科学基金项目(52075236),航空科学基金重点项目(20194603001)。

摘　　要：在利用声学信号进行泄漏检测时,复杂的背景噪声往往会淹没微弱的泄漏信号,导致误判率高。针对微小泄漏在含噪环境中识别困难的问题,提出了基于深度残差收缩网络的含噪微泄漏识别方法。在提出的方法中,添加不同强度高斯噪声,建立数据集,使用深度残差收缩网络进行训练,验证深度残差收缩网络对不同泄漏强度、不同噪声含量样本识别的有效性。实验结果表明:深度残差收缩网络对于微弱泄漏可以达到较理想的识别率,即使在高度杂糅数据识别时仍能达到较理想的识别效果,而且噪声含量并不会对深度残差收缩网络迭代次数产生明显的影响。将提出的方法与卷积神经网络识别方法对比,深度残差收缩网络具有明显的优势。Using acoustic signals for leak detection,complex background noise tends to drown out weak leak signals,resulting in a high rate of false positives.Due to the small leakage is difficult to be identified in noisy environments,a noised micro-leakage identification method based on deep residual shrinkage network(DRSN)is proposed.In the proposed method,the data set is built by superimposing Gaussian noise of different intensities.The DRSN network is used for training to verify the effectiveness of DRSN for sample identification with different leakage intensities and different noise contents.The experimental results show that DRSN can achieve a better recognition rate for weak leakage,and maintain a better recognition effect even when the data is highly hybrid,and the noise content does not have a significant impact on the number of DRSN iterations.Compared with CNN recognition method,DRSN has obvious advantages.

关 键 词：气体管道 泄漏检测 深度残差收缩网络 声波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]