A Deep Learning-based Method for LF-NMR Relaxation Time Spectrum Inversion

doi:10.11938/cjmr20263201

Abstract

Abstract:

Low-field nuclear magnetic resonance (LF-NMR) relaxation time spectrum analysis has been widely applied in petroleum engineering and geological exploration. However, conventional NMR relaxation time spectrum inversion algorithms require time-domain signals with high signal-to-noise ratios (SNRs), which are often difficult to obtain in practical scenarios such as on-site core analysis. To achieve accurate and stable inversion of low-SNR NMR signals, this paper proposes a deep learning-based inversion network named UDMCA (U-Net Denoising and Multi-scale Cross Attention Inversion Network), which incorporates a multi-scale cross-attention mechanism to extract and fuse relaxation features at multiple scales according to the decay characteristics of NMR signals, thereby improving inversion accuracy for low-SNR multi-component signals. In addition, a U-Net denoising module is introduced to further enhance noise robustness. Experimental results show that the proposed method achieves accurate and stable inversion for low-SNR signals and demonstrates good applicability in core NMR signal inversion, providing an effective solution for low-SNR LF-NMR data processing and inversion.

Key words: low-field nuclear magnetic resonance (LF-NMR), relaxation time spectrum, deep learning, 1D inversion

CLC Number:

O482.53

LIU Kewen, JIANG Yukang, CHEN Fang, CHEN Junfei, LU Yuan, CHEN Li, LIU Chaoyang. A Deep Learning-based Method for LF-NMR Relaxation Time Spectrum Inversion[J]. Chinese Journal of Magnetic Resonance, 2026, 43(3): 307-320.

Figures/Tables 14

Fig. 1

Fig. 2

Fig. 3

Fig. 4

Fig. 5

Fig. 6

Table 1

Fig. 7

Fig. 8

Fig. 9

Fig. 10

Table 2

Fig. 11

Fig. 12

References 20

[1]	GUO J C, ZHOU H Y, ZENG J, et al. Advances in low-field nuclear magnetic resonance (NMR) technologies applied for characterization of pore space inside rocks: a critical review[J]. Pet Sci, 2020, 17(5): 1281-1297. doi: 10.1007/s12182-020-00488-0
[2]	FAJT M, MAZUR-ROSMUS W, STEFAŃSKA A, et al. Chert outcrops differentiation by means of low-field NMR relaxometry[J]. Sci Rep, 2024, 14(1): 25280. doi: 10.1038/s41598-024-75945-6 pmid: 39455649
[3]	LU Y, LIU K, WANG Y. Applying NMR T₂ spectral parameters in pore structure evaluation—an example from an Eocene low-permeability sandstone reservoir[J]. Appl Sci, 2021, 11(17): 8027. doi: 10.3390/app11178027
[4]	MITCHELL J, HOWE A M, CLARKE A. Real-time oil-saturation monitoring in rock cores with low-field NMR[J]. J Magn Reson, 2015, 256: 34-42. doi: S1090-7807(15)00098-1 pmid: 25996514
[5]	PRANGE M, SONG Y Q. Understanding NMR spectral uncertainty[J]. J Magn Reson, 2010, 204(1): 118-123. doi: 10.1016/j.jmr.2010.02.010
[6]	高汉宾, 张振芳. 核磁共振原理与实验方法[M]. 武汉: 武汉大学出版社, 2008.
[7]	TIKHONOV A N. On the solution of ill-posed problems and the method of regularization[J]. Dokl Akad Nauk SSSR, 1963, 151(3): 501-504.
[8]	BUTLER J P, REEDS J A, DAWSON S V. Estimating solutions of first kind integral equations with nonnegative constraints and optimal smoothing[J]. SIAM J Numer Anal, 1981, 18(3): 381-397. doi: 10.1137/0718025
[9]	DAY I J. On the inversion of diffusion NMR data: Tikhonov regularization and optimal choice of the regularization parameter[J]. J Magn Reson, 2011, 211(2): 178-185. doi: 10.1016/j.jmr.2011.05.014 pmid: 21683632
[10]	BECK A, TEBOULLE M. A fast iterative shrinkage-thresholding algorithm for linear inverse problems[J]. SIAM J Imaging Sci, 2009, 2(1): 183-202. doi: 10.1137/080716542
[11]	BERMAN P, LEVI O, PARMET Y, et al. Laplace inversion of low-resolution NMR relaxometry data using sparse representation methods[J]. Concepts Magn Reson Part A, 2013, 42(3): 72-88. doi: 10.1002/cmr.a.v42.3
[12]	ZOU Y, XIE R, LIU M, et al. Nuclear magnetic resonance spectrum inversion based on the residual hybrid L₁/L₂ norm[J]. IEEE Geosci Remote Sens Lett, 2018, 15(8): 1194-1198. doi: 10.1109/LGRS.2018.2835457
[13]	LUO S H, XIAO L Z, JIN Y, et al. Low-field NMR inversion based on low-rank and sparsity restraint of relaxation spectra[J]. Pet Sci, 2022, 19(6): 2741-2756. doi: 10.1016/j.petsci.2022.10.020
[14]	PARASRAM T, DAOUD R, XIAO D. T₂ analysis using artificial neural networks[J]. J Magn Reson, 2021, 325: 106930. doi: 10.1016/j.jmr.2021.106930
[15]	LUO G, XIAO L Z, LUO S H, et al. A study on multi-exponential inversion of nuclear magnetic resonance relaxation data using deep learning[J]. J Magn Reson, 2023, 346: 107358. doi: 10.1016/j.jmr.2022.107358
[16]	CHEN B, WU L, CUI X, et al. High-quality reconstruction for Laplace NMR based on deep learning[J]. Anal Chem, 2023, 95(31): 11596-11602. doi: 10.1021/acs.analchem.3c00537
[17]	DEY K K, ASH J T, TREASE N M, et al. Trading sensitivity for information: Carr-Purcell-Meiboom-Gill acquisition in solid-state NMR[J]. J Chem Phys, 2010, 133(5): 054501. doi: 10.1063/1.3463653
[18]	WHITTALL K P, MACKAY A L. Quantitative interpretation of NMR relaxation data[J]. J Magn Reson, 1989, 84(1): 134-152.
[19]	RONNEBERGER O, FISCHER P, BROX T. U-Net: Convolutional networks for biomedical image segmentation[C]// Medical Image Computing and Computer-Assisted Intervention-MICCAI 2015. Cham: Springer, 2015: 234-241.
[20]	KOMATSU R, GONSALVES T. Comparing U-Net based models for denoising color images[J]. AI, 2020, 1(4): 465-486. doi: 10.3390/ai1040029

SNR	MSE		R²
SNR	UDMCA	ATT-CNN	UDMCA	ATT-CNN
10	0.00001384	0.00001610	0.892	0.877
20	0.00000622	0.00000647	0.950	0.948
30	0.00000402	0.00000392	0.968	0.968
40	0.00000260	0.00000292	0.980	0.977
50	0.00000199	0.00000246	0.984	0.980

模型	峰位置/ms	积分比
UDMCA	18.19/266.2	0.95:1
ATT-CNN	19.56/286.2	1:0.88
BRD	12.66/214.1	0.95:1