CN-121999920-A - High-quality Raman spectrum rapid acquisition method and system

CN121999920ACN 121999920 ACN121999920 ACN 121999920ACN-121999920-A

Abstract

The invention relates to the field of data processing, and provides a rapid acquisition method and a rapid acquisition system for a high-quality Raman spectrum, which are used for denoising a rapidly acquired noise-containing Raman spectrum through a zero sample denoising model, so that a Raman spectrum with a high signal to noise ratio is acquired by adopting a shorter exposure time, any clean Raman spectrum is not required to be acquired, only a small amount of noise-containing spectrum and intrinsic noise data are required to be acquired, the training cost is reduced, the operation burden of a user in use is lightened, and the rapid fine adjustment and effective denoising effect stable on various samples are realized through migration training, the generalization capability of the rapid acquisition method is improved, and the rapid acquisition quality and efficiency of the high-quality Raman spectrum are improved.

Inventors

ZHANG JIFAN
Wan Mouqun
Rao Guishi
DAI YICHUAN
CHEN JIANFENG
CHEN XIAOXIAO

Assignees

南昌大学
江西省应急管理科学研究院

Dates

Publication Date: 20260508
Application Date: 20260408

Claims (10)

1. The rapid acquisition method of the high-quality Raman spectrum is characterized by comprising the following steps of: collecting intrinsic noise data of an instrument and preprocessing, wherein the preprocessing is based on standardized processing; Constructing a zero sample denoising model of the Raman spectrum, and optimizing according to the preprocessed intrinsic noise data of the instrument, wherein the zero sample denoising model is based on a coding-decoding architecture; carrying out rapid spectrum data acquisition and main component decomposition treatment on a target sample to construct a training set and a verification set, and carrying out migration training on a zero sample denoising model; And carrying out rapid spectrum data acquisition according to the zero sample denoising model after migration training so as to acquire final rapid acquisition spectrum data, wherein the final rapid acquisition spectrum data is acquired through differential reconstruction.
2. The rapid acquisition method of high quality raman spectra according to claim 1, wherein the step of acquiring and preprocessing intrinsic noise data of the instrument comprises: Setting the detector as short exposure time to perform rapid spectrum data acquisition, expanding a data acquisition area of the detector to synchronously acquire two-dimensional noise data, dividing the two-dimensional noise data into a plurality of one-dimensional noise data, and performing baseline removal and normalization processing to acquire intrinsic noise data of the instrument; the instrument intrinsic noise data is subjected to standardization processing, and the specific algorithm of the standardization processing is as follows: , Wherein, the Representing normalized instrument intrinsic noise data, Representing the intrinsic noise data of the instrument, Representing the mean value of the noise, Representing the variance of the noise.
3. The rapid acquisition method of high quality raman spectrum according to claim 1, wherein the step of constructing a zero sample denoising model of raman spectrum and optimizing according to the preprocessed intrinsic noise data of the instrument specifically comprises: The method comprises the steps of constructing a zero sample denoising model of a Raman spectrum, wherein the zero sample denoising model is formed based on a 6-layer sampling module, the sampling module comprises a downsampling module and an upsampling module, the downsampling module comprises a one-dimensional convolution layer and a mean value pooling layer, the upsampling module comprises two one-dimensional convolution layers and an upsampling layer, the convolution kernel size of the one-dimensional convolution layer is 9, the step size is 1, no offset exists, the kernel size of the mean value pooling layer is 2, and the upsampling module performs upsampling operation based on an adjacent interpolation method; optimizing according to the preprocessed instrument intrinsic noise data, wherein Adam is used as an optimizer in the optimizing, the optimized learning rate adjustment strategy is a linear learning rate, and the specific loss function of the optimizing is as follows: , Wherein, the Representing the loss of the optimization, The minimization function is represented as a function of the minimization, Representing the parameters of the network and, Representing a zero sample denoising model, Representing the instrument intrinsic noise data after the superimposed noise, Denote the L 2 norm.
4. The rapid acquisition method of high-quality raman spectrum according to claim 1, wherein the steps of performing rapid spectrum data acquisition and main component decomposition processing on the target sample specifically comprise: The method comprises the steps of carrying out rapid spectrum data acquisition on a target sample, obtaining noisy spectrum data, carrying out main component decomposition processing, and calculating the average signal-to-noise ratio contribution of a main component, wherein the specific algorithm of the average signal-to-noise ratio contribution is as follows: , , Wherein, the Representing the average signal-to-noise contribution, The number of the principal component is represented, The number of spectra is represented by the number of spectra, The number of the spectrum is represented, Representing the signal-to-noise ratio contribution, Represents the total number of principal components, Representing the projection of the spectral data onto the principal component, The main component is represented by the formula, Representing the average spectrum; Forward contribution screening is carried out according to the average signal-to-noise ratio contribution of the main components so as to obtain the main components of the forward contribution and carry out noise spectrum reconstruction, and the specific algorithm of the noise spectrum reconstruction is as follows: , Wherein, the Representing the components of the reconstructed spectral signal, Representing the principal component of the forward contribution, Representing the forward contribution.
5. The rapid acquisition method of high quality raman spectra according to claim 1, wherein the step of constructing a training set and a validation set comprises: And superposing the one-dimensional spectrum data acquired by the rapid spectrum data as the spectrum data of the intrinsic noise and the reconstruction noise to construct a training set and a verification set, wherein the specific algorithm for constructing the training set and the verification set is as follows: , Wherein, the The training set and the verification set are represented, Representing the reconstructed spectral data of the light spectrum, Representing the one-dimensional spectral data, The scaling factor is represented as such, Representing the translation coefficient.
6. The rapid acquisition method of high quality raman spectra according to claim 1, wherein the step of performing migration training on the zero sample denoising model comprises: and (3) performing rapid spectrum data acquisition, setting short exposure time and acquisition times of the detector, reading and accumulating all pixels in a corresponding row of signals to serve as initial measurement spectrum data, and superposing the intrinsic noise data of the instrument on the initial measurement spectrum data based on a data synthesis algorithm to obtain small sample data for transfer learning and perform transfer training.
7. The rapid acquisition method of high quality raman spectra according to claim 1, wherein the step of performing rapid spectral data acquisition according to the migration trained zero sample denoising model to obtain final rapid acquired spectral data specifically comprises: and carrying out rapid spectrum data acquisition according to the zero sample denoising model after migration training, and obtaining final rapid acquisition spectrum data according to a differential reconstruction algorithm, wherein the specific algorithm of the differential reconstruction is as follows: , Wherein, the Representing final fast acquired spectral data of the differential reconstruction, Representing a zero sample denoising model, And representing the intrinsic noise data of the instrument after the noise is superimposed.
8. A rapid acquisition system for high quality raman spectroscopy comprising: The acquisition module is used for acquiring intrinsic noise data of the instrument and preprocessing, and the preprocessing is based on standardized processing; the de-noising model construction module is used for constructing a zero sample de-noising model of the Raman spectrum, optimizing according to the preprocessed intrinsic noise data of the instrument, and the zero sample de-noising model is based on a coder-decoder architecture; The migration training module is used for carrying out rapid spectrum data acquisition on the target sample and main component decomposition processing so as to construct a training set and a verification set, and carrying out migration training on the zero sample denoising model; And the differential reconstruction module is used for carrying out rapid spectrum data acquisition according to the zero sample denoising model after migration training so as to acquire final rapid acquisition spectrum data, wherein the final rapid acquisition spectrum data is acquired through differential reconstruction.
9. A storage medium storing one or more programs which when executed by a processor implement a method of rapid acquisition of high quality raman spectra according to any one of claims 1 to 7.
10. A computer device comprising a memory and a processor, wherein: The memory is used for storing a computer program; The processor is configured to implement the rapid acquisition method of high quality raman spectra according to any one of claims 1 to 7 when executing a computer program stored on the memory.

Description

High-quality Raman spectrum rapid acquisition method and system Technical Field The invention relates to the field of data processing, in particular to a rapid acquisition method and system of high-quality Raman spectrum. Background Raman spectroscopy is a powerful non-invasive analysis technique that plays an increasingly important role in material science and biomedicine, and provides chemical structure information by analyzing molecular vibration modes, and can be used for disease diagnosis, drug discovery and material characterization, however, in practical applications, raman spectroscopy data is often interfered by noise, which makes it often difficult for current techniques to quickly acquire high quality raman spectra. In the prior art, the denoising method based on deep learning has great application potential in the aspect of quickly acquiring high-quality spectrums, potential signals are reserved mainly by identifying and removing Noise characteristics in quickly acquired Raman spectrums, so that the quality of the spectrums is improved, and a self-supervision denoising algorithm (such as Noise2Noise, noise2Void and the like) avoids using Raman clean data in a training set, so that the method has practical value, however, the existing method and model are usually only optimized for specific samples, often perform poorly when facing new samples with unknown sources, and also face key defects of difficult acquisition of training data, and are difficult to meet the requirements of quickly acquiring Raman spectrums in complex or dynamic scenes. Therefore, how to design a rapid acquisition method of high-quality Raman spectrum to avoid the influence of new samples with unknown sources and complex dynamic scenes, and improve the acquisition quality and efficiency of Raman spectrum becomes a problem to be solved urgently. Disclosure of Invention Based on the method and the system for rapidly acquiring the high-quality Raman spectrum, disclosed by the invention, the rapid acquisition of the noise-containing Raman spectrum is performed through the zero sample denoising model, the acquisition of the Raman spectrum with high signal to noise ratio by adopting shorter exposure time is realized, the acquisition of any clean Raman spectrum is not required, only a small amount of noise-containing spectrum and intrinsic noise data are required to be acquired, the training cost is reduced, the operation burden of a user in use is lightened, and the stable rapid fine adjustment and effective denoising effect on various samples are realized through migration training, so that the generalization capability of the rapid acquisition of the high-quality Raman spectrum is improved. The invention provides a rapid acquisition method of a high-quality Raman spectrum, which comprises the following steps: collecting intrinsic noise data of an instrument and preprocessing, wherein the preprocessing is based on standardized processing; Constructing a zero sample denoising model of the Raman spectrum, and optimizing according to the preprocessed intrinsic noise data of the instrument, wherein the zero sample denoising model is based on a coding-decoding architecture; carrying out rapid spectrum data acquisition and main component decomposition treatment on a target sample to construct a training set and a verification set, and carrying out migration training on a zero sample denoising model; And carrying out rapid spectrum data acquisition according to the zero sample denoising model after migration training so as to acquire final rapid acquisition spectrum data, wherein the final rapid acquisition spectrum data is acquired through differential reconstruction. In summary, according to the rapid acquisition method of high-quality Raman spectrum, the noise of the rapidly acquired noise-containing Raman spectrum is removed through the zero sample denoising model, so that the Raman spectrum with high signal to noise ratio is acquired by adopting shorter exposure time, no clean Raman spectrum is required to be acquired, only a small amount of noise-containing spectrum and intrinsic noise data are required to be acquired, the training cost is reduced, the operation burden of a user in use is lightened, and the stable rapid fine tuning and effective denoising effect on various samples are realized through migration training, so that the generalization capability of the rapid acquisition of the high-quality Raman spectrum is improved. Specifically, intrinsic noise data of an instrument are collected and preprocessed, a zero sample denoising model of a Raman spectrum is constructed based on standardized processing, optimization is carried out according to the intrinsic noise data of the instrument after preprocessing, the zero sample denoising model is based on a coder-decoder framework, the purpose that the Raman spectrum with high signal to noise ratio is obtained by adopting shorter exposure time is achieved, no clean Raman spectrum i