Page 141 - 《应该声学》2022年第2期

第 41 卷 第 2 期 高子洋等： 卷积神经网络的缺陷类型识别分析 309

trasonics NDE[J]. NDT & E International, 2002, 35(8): [16] Khumaidi A, Yuniarno E M, Purnomo M H. Welding de-
567–572. fect classification based on convolution neural network
[8] Veiga J L B C, Carvalho A A D, de Silva I C, et al. (CNN) and Gaussian kernel[C]. IEEE, 2017: 261–265.
The use of artificial neural network in the classification of [17] Munir N, Kim H J, Song S J, et al. Investigation of deep
pulse-echo and TOFD ultra-sonic signals[J]. Journal of the neural network with drop out for ultrasonic flaw classifi-
Brazilian Society of Mechanical Sciences and Engineering, cation in weldments[J]. Journal of Mechanical Science and
2005, 27(4): 394–398. Technology, 2018, 32(7): 3073–3080.
[9] Sambath S, Nagaraj P, Selvakumar N. Automatic de- [18] Munir N, Kim H J, Park J, et al. Convolutional neu-
fect classification in ultrasonic NDT using artificial in-
ral network for ultrasonic weldment flaw classification in
telligence[J]. Journal of Nondestructive Evaluation, 2011,
noisy conditions[J]. Ultrasonics, 2019, 94: 74–81.
30(1): 20–28.
[19] 张重远, 岳浩天, 王博闻, 等. 基于相似矩阵盲源分离与卷积
[10] Filho E F S, Silva M M, Farias P C M A, et al. Flexi-
神经网络的局部放电超声信号深度学习模式识别方法 [J]. 电
ble decision support system for ultrasound evaluation of
网技术, 2019, 43(6): 1900–1907.
fiber–metal laminates implemented in a DSP[J]. NDT &
Zhang Chongyuan, Yue Haotian, Wang Bowen, et al.
E International, 2016, 79: 38–45.
Pattern recognition of partial discharge ultrasonic signal
[11] Cruz F C, Simas Filho E F, Albuquerque M C S, et al. Ef-
based on similar matrix bss and deep learning CNN[J].
ficient feature selection for neural network based detection
Power System Technology, 2019, 43(6): 1900–1907.
of flaws in steel welded joints using ultrasound testing[J].
[20] Munir N, Park J, Kim H J, et al. Performance enhance-
Ultrasonics, 2017, 73: 1–8.
ment of convolutional neural network for ultrasonic flaw
[12] Lecun Y, Bottou L, Bengio Y, et al. Gradient-based learn-
classification by adopting autoencoder[J]. NDT & E In-
ing applied to document recognition[J]. Proceedings of the
ternational, 2020, 111: 102218.
IEEE, 1998, 86: 2278–2324.
[21] Nair V, Hinton G E. Rectified linear units improve
[13] Krizhevsky A, Sutskever I, Hinton G E. ImageNet classi-
restricted boltzmann machines[C]. Proceeding, Twenty-
fication with deep convolutional neural networks[J]. Com-
munications of The ACM 2017, 60(6): 84–90. Seventh International Conference on Machine Learning
[14] 施成龙, 师芳芳, 张碧星. 利用深度神经网络和小波包变换进 (ICML 2010), Haifa, Israel, 2010: 807–814.
行缺陷类型分析 [J]. 声学学报, 2016, 41(4): 499–506. [22] Maas A L, Hannun A Y, Ng A Y. Rectifier nonlineari-
Shi Chenglong, Shi Fangfang, Zhang Bixing. Analysis on ties improve neural network acoustic models[C]. Atlanta,
defect classification by deep neural networks and wavelet Georgia, USA: Proceedings of the 30 th International Con-
packet transform[J]. Acta Acustica, 2016, 41(4): 499–506. ference on Machine Learning, 2013.
[15] Meng M, Chua Y J, Wouterson E, et al. Ultrasonic signal [23] Santurkar S, Tsipras D, Ilyas A, et al. How does batch
classification and imaging system for composite materials normalization help optimization?[C]. Montréal, Canada:
via deep convolutional neural networks[J]. Neurocomput- 32nd Conference on Neural Information Processing Sys-
ing, 2017, 257: 128–135. tems(NeurIPS), 2018.