A NMR Relaxation Time Inversion Method Based on Supervised Deep Neural Networks

Abstract

The invention discloses an NMR relaxation time inversion method based on a supervised deep neural network, constructing a sample pair data set, the sample pair data set is a collection of sample pairs composed of NMR relaxation signals and corresponding NMR relaxation time spectra ;Construct a supervised deep neural network model, construct a loss function; generate a training data set and a verification data set according to the sample pair data set, train a supervised neural network model, and obtain the best simulated NMR relaxation signal and NMR relaxation time spectrum The mapping relationship is recorded as a prediction model; the NMR relaxation signal to be inverted is input into the prediction model for prediction, and the corresponding predicted NMR relaxation time spectrum is output. The prediction process of the present invention is fully automatic without manual participation and prior information; the predicted NMR relaxation time spectrum is more accurate and can provide more reliable quantitative information.

Description

technical field [0001] The invention belongs to the technical field of nuclear magnetic resonance, and in particular relates to an NMR relaxation time inversion method based on a supervised deep neural network. Background technique [0002] In the field of nuclear magnetic resonance (NMR) research, the NMR relaxation time of the sample being studied is closely related to the structure and dynamic process of the material molecule and the environment in which it is located, and is a characteristic parameter that characterizes the relationship between the properties of the material and the environment in which it is located. There are two types of NMR relaxation times most commonly used in research: the longitudinal (spin-lattice) relaxation time T 1 and the transverse (spin-spin) relaxation time T 2 . For the NMR sample of a simple system (such as pure water), the relaxation process is a monoexponential time-varying function, and the relaxation time of the sample (T 1 and T...

AI Technical Summary

Problems solved by technology

However, these traditional inversion methods generally require prior information, and the regularization factor needs to be adjusted dynamically. Unmatched regularization parameters are likely to cause broadening of the relaxation time spectrum or significant changes in the weight of the neural network. To a certain extent, this type of inversion method is limited in its versatility and quantification accuracy, especially in the study of systems with complex structures or sample distributions, which will further aggravate the uncertainty of test results

In addition, this type of method usually finds the optimal solution of the objective function in an iterative manner, and the algorithm is very time-consuming

Images