基于SSDAE深度神经网络的钛板电涡流检测图像分类研究
|
摘要
钛板电涡流成像检测易受工业现场中的噪声影响,包含噪声的检测图像往往难以提取较好的特征,从而影响分类识别精度。针对以上问题,提出了一种基于栈式稀疏降噪自编码(SSDAE)深度神经网络的钛板缺陷电涡流检测图像分类方法。将稀疏性限制引入降噪自编码器并进行逐层无监督自学习,然后将自编码器栈式组合后添加逻辑识别(LR)层,构建出SSDAE深度神经网络,网络在有监督微调后可实现钛板缺陷电涡流图像特征自动提取与分类识别。稀疏性限制的引入提高了特征学习能力,降噪自编码器的栈式组合提高了深度网络的鲁棒性。实验结果表明,相比其他常规方法,所提出方法不仅在理想环境下有更高的分类准确率,且该方法能有效抵抗噪声,在复杂工况下能更有效地对钛板缺陷进行分类识别。
Eddy current imaging detection of titanium plate is susceptible to noise from industrial field. It is difficult to extract effective features from the detected images containing noise,which affects the classification accuracy. To address this issue,a classification method for eddy current detection images of titanium plate defects based on stacked sparse denoising autoencoder( SSDAE) deep neural network is proposed. Sparsity constraint is introduced into the denoising autoencoders( DAE) and the autoencoders perform unsupervised self-learning layer-by-layer. Then,the autoencoders are stacked and a logistic regression( LR) layer is added to construct the deep neural network. The deep neural network can automatically extract features and classify eddy current detection images of titanium plate defects after supervised fine-tuning. The feature learning ability is improved by sparsity constraint,and the robustness of deep network is improved by stack combination of denoising autoencoders. Experimental results show that,compared with other traditional methods,the proposed method not only has higher classification accuracy in ideal conditions,but also can resist noise and classify defects of titanium plate in complex conditions more effectively.
Eddy current imaging detection of titanium plate is susceptible to noise from industrial field. It is difficult to extract effective features from the detected images containing noise,which affects the classification accuracy. To address this issue,a classification method for eddy current detection images of titanium plate defects based on stacked sparse denoising autoencoder( SSDAE) deep neural network is proposed. Sparsity constraint is introduced into the denoising autoencoders( DAE) and the autoencoders perform unsupervised self-learning layer-by-layer. Then,the autoencoders are stacked and a logistic regression( LR) layer is added to construct the deep neural network. The deep neural network can automatically extract features and classify eddy current detection images of titanium plate defects after supervised fine-tuning. The feature learning ability is improved by sparsity constraint,and the robustness of deep network is improved by stack combination of denoising autoencoders. Experimental results show that,compared with other traditional methods,the proposed method not only has higher classification accuracy in ideal conditions,but also can resist noise and classify defects of titanium plate in complex conditions more effectively.
引文
[1]李明利.我国钛及钛合金板带材应用现状分析[J].钛工业进展,2011,28(6):14-17.LI M L. Application status on domestic titanium and titanium alloys plate and strip[J]. Titanium Industry Progress,2011,2011,28(6):14-17.
[2]王瑞琴,黄先明,吴晓东,等.冷轧钛板材缺陷分析与讨论[J].热加工工艺,2014(13):147-149.WANG R Q,HUANG X M,WU X D,et al. Discussions and ananysis of defects for cold rolled titanium sheet[J].Hot Working Technology,2014(13):147-149.
[3]潘孟春,何赟泽,罗飞路.基于谱分析的脉冲涡流缺陷3D分类识别技术[J].仪器仪表学报,2010,31(9):2095-2100.PAN M CH,HE Y Z,LUO F L. Pulsed eddy current defect 3D classification technique based on frequency spectrum analysis[J]. Chinese Journal of Scientific Instrument,2010,31(9):2095-2100.
[4]周德强,张斌强,田贵云,等.脉冲涡流检测中裂纹的深度定量及分类识别[J].仪器仪表学报,2009,30(6):1190-1194.ZHOU D Q, ZHANG B Q, TIAN G Y, et al.Quantification of depth and classification of cracks using pulsed eddy current test technology[J]. Chinese Journal of Scientific Instrument,2009,30(6):1190-1194.
[5] TIAN G Y,SOPHIAN A,TAYLOR D,et al. Wavelet-based PCA defect classification and quantification for pulsed eddy current NDT[J]. IEE ProceedingsScience,Measurement and Technology, 2005, 152(4):141-148.
[6] PENG Y,QIU X B,WEI J L,et al. Defect classification using PEC respones based on power spectral density analysis combined with EMD and EEMD[J]. Ndt&E International,2016(78):37-51.
[7]梁子千,玄文博,王婷,等.基于二维阻抗特征的管道环焊缝缺陷涡流检测[J].仪器仪表学报,2017,38(9):2138-2145.LIAN Z Q,XUAN W B,WANG T,et al. Eddy current NDt for the cracks of girth welds of pipes based on 2D impedance characteristics[J]. Chinese Journal of Scientific Instrument,2017,38(9):2138-2145.
[8] PASADAS D J,RAMOS H G,FENG B,et al. Defect classification with SVM and wideband excitation in multilayer aluminum plates[J]. IEEE Transactions on Instrumentation and Measurement, 2019, DOI:10. 1109/TIM.2019. 2893009.
[9] HINTON G E,SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks[J]. Science,2006,313(5786):504-507.
[10] XING CH, MA L, YANG X Q. Stacked denoise autoencoder based feature extraction and classification for hyperspectral images[J]. Journal of Sensors,2016,doi:10. 1155/2016/3632943.
[11]陈瑞娟,戚昊峰,李炳南,等.基于栈式自编码器的磁探测电阻抗成像算法研究[J].仪器仪表学报,2019,40(1):257-264.CHEN R J,QI H F,LI B N,et al. Study on magnetic detection electrical impedance tomography algorithm based on stacked auto-encoder[J]. Chinese Journal of Scientific Instrument,2019,40(1):257-264.
[12]陈仁祥,杨星,杨黎霞,等.栈式稀疏加噪自编码深度神经网络的滚动轴承损伤程度诊断[J].振动与冲击,2017,36(21):125-131.CHEN R X,YANG X,YANG L X,et al. Fault severity diagnosis method for rolling bearings based on a stacked sparse denoising auto-encoder[J]. Journal of Vibration&Shock,2017,36(21):125-131.
[13]赵光权,刘小勇,姜泽东,等.基于深度学习的轴承健康因子无监督构建方法[J].仪器仪表学报,2018,39(6):82-88.ZHAO G Q,LIU X Y,JIANG Z D,et al. Unsupervised health indicator of bearing based on deep learning[J].Chinese Journal of Scientific Instrument,2018,39(6):82-88.
[14] VINCENT P,LAROCHELLE H,BENGIO Y,et al.Extracting and composing robust features with denoising autoencoders[C]. International Conference on Machine Learning,ACM,2008:1096-1103.
[15] VINCENT P, LAROCHELLE H, LAJOIE I, et al.Stacked denoising autoencoders:Learning useful representations in a deep network with a local denoising criterion[J]. Journal of Machine Learning Research,2010,11(12):3371-3408
[16] BENGIO Y,LAMBLIN P,DAN P,et al. Greedy layerwise training of deep networks[J]. Advances in Neural Information Processing Systems,2007(19):153-160.
[17] YIN W L,CHEN G,CHEN L J,et al. The design of a digital magnetic induction tomography(MIT)system for metallic object imaging based on half cycle demodulation[J]. IEEE Sensors Journal, 2011,11(10):2233-2240.
[18] XU H Y,AVILA J R S,WU F F,et al. Imaging X70weld cross-section using electromagnetic testing[J].NDT&E International,2018(98):155-160.
[19] CUI Z Q,WANG H X,YIN W L,et al. Impulsive noise reduction in digital phase-sensitive demodulation by nonlinear filtering[J]. Measurement Science and Technology,2015,26(7):075401.
[2]王瑞琴,黄先明,吴晓东,等.冷轧钛板材缺陷分析与讨论[J].热加工工艺,2014(13):147-149.WANG R Q,HUANG X M,WU X D,et al. Discussions and ananysis of defects for cold rolled titanium sheet[J].Hot Working Technology,2014(13):147-149.
[3]潘孟春,何赟泽,罗飞路.基于谱分析的脉冲涡流缺陷3D分类识别技术[J].仪器仪表学报,2010,31(9):2095-2100.PAN M CH,HE Y Z,LUO F L. Pulsed eddy current defect 3D classification technique based on frequency spectrum analysis[J]. Chinese Journal of Scientific Instrument,2010,31(9):2095-2100.
[4]周德强,张斌强,田贵云,等.脉冲涡流检测中裂纹的深度定量及分类识别[J].仪器仪表学报,2009,30(6):1190-1194.ZHOU D Q, ZHANG B Q, TIAN G Y, et al.Quantification of depth and classification of cracks using pulsed eddy current test technology[J]. Chinese Journal of Scientific Instrument,2009,30(6):1190-1194.
[5] TIAN G Y,SOPHIAN A,TAYLOR D,et al. Wavelet-based PCA defect classification and quantification for pulsed eddy current NDT[J]. IEE ProceedingsScience,Measurement and Technology, 2005, 152(4):141-148.
[6] PENG Y,QIU X B,WEI J L,et al. Defect classification using PEC respones based on power spectral density analysis combined with EMD and EEMD[J]. Ndt&E International,2016(78):37-51.
[7]梁子千,玄文博,王婷,等.基于二维阻抗特征的管道环焊缝缺陷涡流检测[J].仪器仪表学报,2017,38(9):2138-2145.LIAN Z Q,XUAN W B,WANG T,et al. Eddy current NDt for the cracks of girth welds of pipes based on 2D impedance characteristics[J]. Chinese Journal of Scientific Instrument,2017,38(9):2138-2145.
[8] PASADAS D J,RAMOS H G,FENG B,et al. Defect classification with SVM and wideband excitation in multilayer aluminum plates[J]. IEEE Transactions on Instrumentation and Measurement, 2019, DOI:10. 1109/TIM.2019. 2893009.
[9] HINTON G E,SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks[J]. Science,2006,313(5786):504-507.
[10] XING CH, MA L, YANG X Q. Stacked denoise autoencoder based feature extraction and classification for hyperspectral images[J]. Journal of Sensors,2016,doi:10. 1155/2016/3632943.
[11]陈瑞娟,戚昊峰,李炳南,等.基于栈式自编码器的磁探测电阻抗成像算法研究[J].仪器仪表学报,2019,40(1):257-264.CHEN R J,QI H F,LI B N,et al. Study on magnetic detection electrical impedance tomography algorithm based on stacked auto-encoder[J]. Chinese Journal of Scientific Instrument,2019,40(1):257-264.
[12]陈仁祥,杨星,杨黎霞,等.栈式稀疏加噪自编码深度神经网络的滚动轴承损伤程度诊断[J].振动与冲击,2017,36(21):125-131.CHEN R X,YANG X,YANG L X,et al. Fault severity diagnosis method for rolling bearings based on a stacked sparse denoising auto-encoder[J]. Journal of Vibration&Shock,2017,36(21):125-131.
[13]赵光权,刘小勇,姜泽东,等.基于深度学习的轴承健康因子无监督构建方法[J].仪器仪表学报,2018,39(6):82-88.ZHAO G Q,LIU X Y,JIANG Z D,et al. Unsupervised health indicator of bearing based on deep learning[J].Chinese Journal of Scientific Instrument,2018,39(6):82-88.
[14] VINCENT P,LAROCHELLE H,BENGIO Y,et al.Extracting and composing robust features with denoising autoencoders[C]. International Conference on Machine Learning,ACM,2008:1096-1103.
[15] VINCENT P, LAROCHELLE H, LAJOIE I, et al.Stacked denoising autoencoders:Learning useful representations in a deep network with a local denoising criterion[J]. Journal of Machine Learning Research,2010,11(12):3371-3408
[16] BENGIO Y,LAMBLIN P,DAN P,et al. Greedy layerwise training of deep networks[J]. Advances in Neural Information Processing Systems,2007(19):153-160.
[17] YIN W L,CHEN G,CHEN L J,et al. The design of a digital magnetic induction tomography(MIT)system for metallic object imaging based on half cycle demodulation[J]. IEEE Sensors Journal, 2011,11(10):2233-2240.
[18] XU H Y,AVILA J R S,WU F F,et al. Imaging X70weld cross-section using electromagnetic testing[J].NDT&E International,2018(98):155-160.
[19] CUI Z Q,WANG H X,YIN W L,et al. Impulsive noise reduction in digital phase-sensitive demodulation by nonlinear filtering[J]. Measurement Science and Technology,2015,26(7):075401.