CN-121978081-A - Self-adaptive Raman measurement method and system based on spectral structure decoupling

CN121978081ACN 121978081 ACN121978081 ACN 121978081ACN-121978081-A

Abstract

The invention discloses a self-adaptive Raman measurement method and system based on spectral structure decoupling. The method comprises the steps of defining a parameter space containing laser power and integration time, carrying out a small amount of initial measurement to obtain an initial spectrum data set, carrying out decoupling treatment on each spectrum, decomposing the spectrum into a scale component representing the overall intensity level and a structure component representing the relative morphological characteristics, constructing a probability spectrum proxy model capable of predicting spectrum distribution under any parameter based on the measurement parameters and the scale and structure components after decoupling, carrying out spectrum quality prediction on candidate measurement parameters by using the model and calculating uncertainty, selecting target parameters for next measurement by fusing decision functions of prediction quality and uncertainty, updating the data set after the measurement and reconstructing the model, iterating until termination conditions are met, outputting optimal measurement parameters and spectrums, and obviously improving the automation degree, efficiency and result stability of Raman measurement.

Inventors

LI YUTING
ZHANG YINUO
CHEN XIN
WANG ZILONG
LIANG PEI
HE YUN
ZHANG YULU

Assignees

中国计量大学

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (8)

1. The self-adaptive Raman measurement method based on spectral structure decoupling is characterized by comprising the following steps of: S1, defining a parameter space containing at least one adjustable Raman measurement parameter; S2, selecting a plurality of groups of measurement parameters in the parameter space to perform Raman spectrum measurement, and acquiring an initial data set containing the measurement parameters and corresponding Raman spectrums; S3, decoupling each Raman spectrum in the initial data set, and decomposing the Raman spectrum into a scale component representing the overall intensity level of the spectrum and a structural component representing the relative morphological characteristics of the spectrum; S4, constructing a probability spectrum proxy model based on the measurement parameters in the initial data set and the structural components and the scale components obtained through decoupling treatment, wherein the probability spectrum proxy model is used for predicting probability distribution of the structural components and the scale components of the Raman spectrum under the given measurement parameters; S5, selecting a plurality of groups of candidate measurement parameters which are not contained in the initial data set from the parameter space, predicting the spectrum quality of the candidate measurement parameters by utilizing the probability spectrum agency model, and calculating the uncertainty of a corresponding prediction result; s6, selecting a target parameter to be measured next time from the candidate measurement parameters according to the spectrum quality prediction result and the uncertainty thereof obtained in the step S5; S7, carrying out Raman spectrum measurement by using the target parameters, obtaining new spectrum data, generating a second data set, reconstructing the probability spectrum agent model based on the second data set, and continuing iteration; And S8, stopping iteration when the probability spectrum agent model meets a preset termination condition, and outputting an optimal measurement parameter and/or a corresponding optimal Raman spectrum.
2. An adaptive raman measurement method based on spectral structure decoupling according to claim 1, wherein said measurement parameter space comprises at least laser power and integration time.
3. The adaptive raman measurement method based on spectral structure decoupling according to claim 1, wherein in step S3, the decoupling process comprises: Each Raman spectrum in the initial dataset Decomposing into a scale component and a structural component; Scale component The definition is as follows: Wherein, the Represent the first The raman spectrum is at the first Intensity values at the wavenumber sampling points; structural component The definition is as follows: Wherein, the Indicating the relative morphological characteristics of the raman spectrum, To prevent an extremely small positive number from being unstable in value.
4. A method of adaptive raman measurement based on spectral structure decoupling according to claim 3, wherein in step S4, said constructing a probabilistic spectral proxy model comprises: For the structural component Performing dimension reduction treatment: Wherein the method comprises the steps of Is the average structure spectrum; In order to construct the base matrix of the structure, Is a low-dimensional feature dimension; is a corresponding structural coefficient vector; establishing a mapping relation between the measurement parameters and the structural coefficients: Establishing a mapping relation between the measurement parameters and the scale components: Wherein the method comprises the steps of And (3) with Is a probability prediction model; , Respectively is And (3) with Corresponding modeling errors.
5. The adaptive raman measurement method based on spectral structure decoupling according to claim 1, wherein in step S5, specifically comprising: Sampling from the probabilistic spectrum proxy model for any candidate measurement parameter to generate M groups of possible combinations of structural components and scale components; Reconstructing M predicted Raman spectra according to each group of combinations; calculating a quality score of each predicted raman spectrum by using a predefined spectrum quality evaluation function; and calculating the expected value and uncertainty of the spectrum quality corresponding to the candidate measurement parameters based on the M quality scores.
6. The method for adaptive raman measurement based on spectral structure decoupling according to claim 5, wherein in step S6, specifically comprising: Constructing a decision function based on the expected value and uncertainty of the spectrum quality: Wherein, the Is a weight for adjusting the degree of attention to the uncertain region; Expressed as spectral average mass; Expressed as uncertainty of model predictions; Selecting the decision function The candidate measurement parameter with the largest value is taken as the target parameter: 。
7. The method according to claim 1, wherein in step S8, the predetermined termination condition is that the number of iterations reaches a predetermined maximum value, or the uncertainty of prediction of all candidate measurement parameters in the parameter space is lower than a predetermined threshold.
8. An adaptive raman measurement system based on spectral structure decoupling for performing the adaptive raman measurement method according to any one of claims 1 to 7, comprising: a parameter configuration module for defining a parameter space containing at least one tunable raman measurement parameter; the spectrum acquisition module is used for selecting a plurality of groups of measurement parameters in the parameter space to carry out Raman spectrum measurement and acquiring an initial data set containing the measurement parameters and corresponding Raman spectrums; The decoupling processing module is used for carrying out decoupling processing on each Raman spectrum in the initial data set and decomposing the Raman spectrum into a scale component representing the overall intensity level of the spectrum and a structural component representing the relative morphological characteristics of the spectrum; The modeling prediction module is used for selecting a plurality of groups of candidate measurement parameters which are not contained in the initial data set from the parameter space, predicting the spectrum quality of the candidate measurement parameters by utilizing the probability spectrum agency model, and calculating the uncertainty of a corresponding prediction result; The decision module is used for selecting a target parameter to be measured next time from the candidate measurement parameters according to the spectrum quality prediction result and the uncertainty thereof; and the control module is used for performing Raman spectrum measurement by using the target parameters, obtaining new spectrum data, generating a second data set, reconstructing the probability spectrum proxy model based on the second data set, continuing iteration, stopping iteration when the probability spectrum proxy model meets a preset termination condition, and outputting optimal measurement parameters and/or corresponding optimal Raman spectrum.

Description

Self-adaptive Raman measurement method and system based on spectral structure decoupling Technical Field The invention relates to the technical field of spectrum analysis, in particular to a self-adaptive Raman measurement method and system based on spectrum structure decoupling. Background The Raman spectrum technology is used as an important molecular vibration spectrum analysis means, has the advantages of non-contact, no need of sample pretreatment, capability of providing molecular structure information of substances and the like, and is widely applied to the fields of material analysis, biomedical analysis, industrial production and the like. In practical applications, the measurement quality of raman spectra is affected by a variety of experimental parameters, of which laser power and integration time are the most critical parameters. The selection of laser power and integration time directly relates to the Raman signal intensity, signal-to-noise ratio and spectrum stability, and incorrect parameter setting can cause the problems of over-weak signal, poor signal-to-noise ratio, even sample ablation and the like. In the existing raman measurement practice, the measurement parameters are usually selected depending on manual experience or preset fixed parameters. When measuring the same substance, an operator often adjusts the laser power and the integration time step by step through multiple experiments to obtain a relatively ideal raman spectrum. However, the parameter setting method relying on experience has the problems of low efficiency, poor repeatability, strong subjectivity and the like, and is difficult to adapt to the rapid measurement requirement under complex samples or variable measurement environments. In the scenes of industrial online detection or on-site rapid analysis and the like, the frequent dependence on manual parameter adjustment not only increases the cost of measurement time, but also reduces the automation and intelligent level of the system. To cope with this problem, research into raman spectrum adaptive measurement is started. The existing Raman spectrum self-adaptive measurement focuses on acquiring a Raman spectrum under the condition of a set measurement parameter, and carrying out subsequent analysis around the extraction, enhancement or discrimination of spectrum characteristics. For example, WO2023272749A1 proposes a method for extracting and detecting characteristic information of a small molecule raman spectrum signal, which improves the detection accuracy of a target molecule by partitioning, extracting characteristic and statistically analyzing the raman spectrum. The method mainly focuses on the expression mode and analysis algorithm of the spectral characteristics, can effectively extract target information under specific measurement conditions, but the measurement process is generally based on preset experimental parameters such as laser power, integration time and the like, and lacks a dynamic optimization and self-adaptive adjustment mechanism for the measurement parameters. In addition, some prior art attempts have been made to manually or semi-automatically adjust the measurement parameters through multiple experiments to obtain superior raman spectrum quality. However, the method needs to correct parameters in a gradual heuristic way, so that not only is the measurement efficiency low, but also consistent and stable measurement results are difficult to obtain under different samples or different environmental conditions. With the development of intelligent measurement and data-driven analysis technologies, some researches start to attempt to introduce a machine learning method to optimize raman measurement parameters. Attempts have been made to assist in determining a better combination of measured parameters by constructing a regression model between the measured parameters and the signal strength or signal to noise ratio. However, most of the existing regression model building methods simplify the raman spectrum into a single scalar index, such as peak intensity, signal-to-noise ratio or integral intensity, and ignore the abundant structural information contained in the raman spectrum as a high-dimensional signal. Although the method can improve the efficiency of parameter selection to a certain extent, the influence of the measured parameter change on the overall form, peak structure and background characteristics of the spectrum is difficult to comprehensively reflect. On the other hand, existing raman measurement optimization methods based in part on machine learning generally employ a way to model the whole spectrum directly as a high-dimensional output. Because the Raman spectrum has higher dimensionality and strong correlation among wave number points, the modeling mode is often greatly dependent on the number of samples, and a relatively stable prediction result can be obtained only by acquiring a large amount of spectrum data under different para