De-Noising Method for Seismic Data via Improved Convolution Neural Network

Shaohua Cui; Suwen Li; Xude Wang

doi:10.3788/LOP57.061501

Journals >Laser & Optoelectronics Progress >Volume 57 >Issue 6 >Page 061501 > Article

Laser & Optoelectronics Progress
Vol. 57, Issue 6, 061501 (2020)

De-Noising Method for Seismic Data via Improved Convolution Neural Network

Shaohua Cui^*, Suwen Li, and Xude Wang

Author Affiliations

College of Physics and Electronic Information, Huaibei Normal University, Huaibei, Anhui 235000, China

show less

DOI: 10.3788/LOP57.061501 Cite this Article Set citation alerts

Shaohua Cui, Suwen Li, Xude Wang. De-Noising Method for Seismic Data via Improved Convolution Neural Network[J]. Laser & Optoelectronics Progress, 2020, 57(6): 061501 Copy Citation Text

show less

Fig. 1. Three stages of single-layer convolution neural network

Download full size

Fig. 2. LeNet-5 system

Download full size

Fig. 3. Denoising results of different algorithms for pre-stack seismic data. (a) Original seismic data; (b) noisy seismic data; (c) denoising results of SVD algorithm; (d) denoising results of BP algorithm; (e) denoising results of algorithm proposed in Ref. [9]; (f) denoising results of proposed algorithm

Download full size

Fig. 4. Denoising results of different algorithms for post-stack seismic data. (a) Original seismic data; (b) noisy seismic data; (c) denoising results of SVD; (d) denoising results of BP; (e) denoising results of algorithm proposed in Ref. [9]; (f) denoising results of proposed algorithm

Download full size

Number of feature maps	3×3	5×5	7×7	9×9
1	0.0475	0.0520	0.0585	0.0680
2	0.0360	0.0245	0.0555	0.0445
3	0.0340	0.0275	0.0365	0.0355
4	0.0370	0.0250	0.0315	0.0300
5	0.0312	0.0265	0.0345	0.0330
6	0.0420	0.0520	0.0315	0.5000
7	0.5000	0.0335	0.5000	0.0285
8	0.5000	0.5000	0.0305	0.0285
9	0.5000	0.0235	0.0385	0.5000
10	0.5000	0.5000	0.0327	0.0290
11	0.5000	0.5000	0.5000	0.0312
12	0.5000	0.5000	0.5000	0.0441
13	0.5000	0.5000	0.5000	0.5000
14	0.5000	0.5000	0.5000	0.5000
15	0.5000	0.5000	0.5000	0.5000
16	0.5000	0.5000	0.5000	0.5000

Table 1. M_SE values corresponding to different sizes of convolution kernels and numbers of feature maps in C1 layer

Number of feature maps	1×1	3×3	5×5
1	0.0620	0.0460	0.0510
2	0.0565	0.0425	0.0435
3	0.0570	0.0545	0.0435
4	0.0405	0.0410	0.0430
5	0.0415	0.0340	0.0345
6	0.0535	0.0190	0.0330
7	0.0495	0.0370	0.0295
8	0.0395	0.0325	0.0340
9	0.0480	0.0395	0.0260

Table 2. M_SE values corresponding to different sizes of convolution kernels and feature map numbers of C3 layer when the convolution kernel size is 5×5 and the number of feature maps is 9 for

Seismic data	Algorithm	P_SNR /dB	S_NR /dB	M_SE
	SVD	62.9160	25.7189	0.0095
Pre-stack data	BP	61.0352	24.4899	0.0156
	Algorithm in Ref. [9]	64.6129	27.3903	0.0080
	Proposed algorithm	64.6236	28.1604	0.0067
	SVD	61.2323	27.4449	0.0079
Post-stack data	BP	60.3104	24.6603	0.0150
	Algorithm in Ref. [9]	64.9023	29.6088	0.0048
	Proposed algorithm	66.6712	31.9728	0.0022

Table 3. Results of different algorithms for pre-stack and post-stack seismic data

Noise level /%	P_SNR /dB	S_NR /dB	M_SE
5	80.9861	38.3296	0.0005
10	71.0896	34.3800	0.0016
15	66.6712	31.9728	0.0022
20	65.5404	26.8308	0.0091
25	64.3720	23.7155	0.0150
30	63.7596	22.1031	0.0217

Table 4. Denoising results of proposed algorithm at different noise levels

Shaohua Cui, Suwen Li, Xude Wang. De-Noising Method for Seismic Data via Improved Convolution Neural Network[J]. Laser & Optoelectronics Progress, 2020, 57(6): 061501

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information