Data processing is a key link in seismic exploration, and the UNet network, as one of the typical neural network architectures, has also been used in the field of seismic exploration in recent years as a means of suppressing random noise. The UNet network is based on its symmetric encoding and decoding structure, which can extract a wide range of contextual information. However, due to the excessive use of down-sampling operations in its encoding part, it is easy to lose the spatial details of the input image. Secondly, the UNet architecture is a single-stage model with a simple network structure, making it difficult to achieve a balance between spatial accuracy and multi-scale information. Based on this, a multi-stage progressive UNet network (MPUNet) is proposed. The first two-stages of the network use an encoder-decoder to learn rich multi-scale information, and the last stage preserves accurate spatial details through the original resolution sub network. Introducing supervised attention modules between each two-stages to recalibrate features entering the next stage, and introducing cross stage feature fusion mechanisms to make the entire network framework more tightly connected and avoid the loss of effective information. The experimental results of artificial synthetic records and actual seismic data show that compared to time-frequency peak filtering (TFPF) and residual dense networks (RDNet), traditional UNet and UNet with residual dense blocks (RDBUNet), MPUNet has a more significant denoising effect, which can effectively improve the signal-to-noise ratio and resolution of seismic data, providing a favorable basis for subsequent analysis and interpretation of seismic data.
牛永效.基于频率及 FK 域联合应用的地震波去噪技术研究[J].铁道标准设计, 2017, 61(2): 47-49.
[2]
NIUYongxiao. Research on seismic wave denoising technology based on frequency and F-K domain joint application.Railway[J].Standard Design, 2017, 61(2): 47-49.(in Chinese)
[3]
LIZ, GAOJ, LIUN, et al.Random noise suppression of seismic data by time-frequency peak filtering with variational mode decomposition [J].Exploration Geophysics, 2019, 50(6): 634-644.
[4]
LIUN, YANGY, LIZ, et al.Seismic signal de-noising using time-frequency peak filtering based on empirical wavelet transform[J].Acta Geophysica, 2020, 68(2): 425-434.
[5]
LUY, HUANGY, XUEW, et al.Seismic data processing method based on wavelet transform for de-noising[J].Cluster Computing, 2019, 22(3): 6609-6620.
[6]
SIWEIZ, DAYONGZ, XIAOKANGY, et al.Noise reduction method based on curvelet theory of seismic data[J]. Petroleum Science and Technology, 2023, 41(24): 2344-2361.
[7]
JIG, WANGC.A Denoising method for seismic data based on SVD and deep learning[J].Applied Sciences, 2022, 12(24): 12840.
[8]
ZHANGF, CAIN, WUJ, et al.Image denoising method based on a deep convolution neural network[J].IET Image Processing, 2018, 12(4): 485-493.
CUILiang, XUYubing, CHENYaoyu. Application of an improved FCM clustering algorithm in noise image segmentation[J]. Journal of North University of China(Natural Science Edition), 2016, 37(1): 76-82. (in Chinese)
[11]
YANGYang, CHENGuangwu, WANGJingwen. Research on train combination positioning method based on grey neural network[J]. Journal of Measurement Science and Instrumentation, 2019, 10(2): 143-149.
[12]
WUZ, ZHAOD, LIANGQ, et al.Dynamic sparsity neural networks for automatic speech recognition[C]//ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).IEEE, 2021: 6014-6018.
[13]
BABKENA, HIDDEB, MAA W. Towards Advanced diagnosis and management of inherited arrhythmia syndromes: harnessing the capabilities of artificial intelligence and machine learning[J]. Heart Rhythm, 2023, 20(10): 1399-1407.
WANGYuqing, LUWenkai, LIUJinlin, et al.Earthquake random noise suppression based on data augmentation and CNN[J].Journal of Geophysics.2019, 62(1): 421-433.(in Chinese)
LIHaishan, CHENDewu, WUjie, et al.Pre-stack random noise suppression with deep residual network[J].Oil Geophysical Prospecting, 2020, 55(3): 493-503.(in Chinese)
[20]
ZHANGY, TIANY, KONGY, et al. Residual dense network for image super-resolution[J].IEEE Access, 2018, 43: 2472-2481.
[21]
GAOZ, ZHANGS, CAIJ, et al. Research on deep convolutional neural network time-frequency domain seismic signal denoising combined with residual dense blocks[J]. Frontiers in Earth Science, 2021, 9: 1-14.
[22]
JIANXIANC, LIW, JIANGSHANZ, et al. Denoising method for seismic co-band noise based on a u-net network combined with a residual dense block[J]. Applied Sciences, 2023, 13(3): 1324-1324.
[23]
ZHAOZ, XIONGB, GAIS, et al.Improved deep multi-patch hierarchical network with nested module for dynamic scene deblurring[J].IEEE Access, 2020, 8: 62116-62126.