A raman spectrum analysis method based on two-dimensional raman diagram combined with deep learning
By converting one-dimensional Raman spectral data into two-dimensional Raman images and combining them with a deep learning model, the problem of insufficient accuracy in existing Raman spectral data analysis methods is solved, enabling efficient sample identification and discrimination and improving analytical precision.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TSINGHUA UNIVERSITY
- Filing Date
- 2022-10-10
- Publication Date
- 2026-04-21
AI Technical Summary
Existing Raman spectroscopy data analysis methods are insufficient in terms of accuracy and precision, especially in small sample sets and one-dimensional data processing, making it difficult to effectively utilize the potential of artificial intelligence models.
One-dimensional Raman spectral data is converted into two-dimensional Raman images using four encoding methods (recursive graph of the spectrum, Grami angle deformation field, spectrogram, and Markov transition field). The data conversion is combined with deep learning model training, and the appropriate conversion method is selected according to the complexity of the spectral data to improve data adaptability and accuracy.
It achieves high-accuracy sample identification and differentiation, improves the precision and efficiency of Raman spectroscopy analysis, and is suitable for the analysis of biological samples.
Smart Images
Figure CN115565004B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of spectral analysis technology, and in particular to a Raman spectral analysis method based on two-dimensional Raman images combined with deep learning. Background Technology
[0002] Raman spectroscopy is a label-free optical technique with the advantage of specificity, allowing analysis of the biochemical properties of substances based on factors such as the position and intensity of Raman peaks. Therefore, it has significant applications in sample analysis and identification.
[0003] There are many methods for analyzing and processing Raman spectral data. Peak intensity comparison is the most direct method, but its accuracy is not very high and it is mainly suitable for small sample sets. Multivariate statistical methods are one of the most commonly used Raman spectral data analysis methods, and they have many sub-methods, such as principal component analysis and principal component analysis-linear discriminant analysis. However, these methods have certain limitations in terms of accuracy. Therefore, it is necessary to find new methods to improve accuracy.
[0004] With the development of information technology, artificial intelligence (AI) technology has great potential for application in data analysis, especially in image classification and identification. However, Raman spectroscopy data is a type of one-dimensional data. Although AI methods can directly load one-dimensional data into models for training, the accuracy is not particularly high. Therefore, it is necessary to find a method to transform one-dimensional Raman data into two-dimensional images and select appropriate AI methods for training. Summary of the Invention
[0005] The present invention aims to at least partially solve one of the technical problems in the related art.
[0006] To address this, this invention proposes a Raman spectroscopy analysis method based on two-dimensional Raman images combined with deep learning. The concept of "two-dimensional Raman image" is introduced, referring to the process of transforming one-dimensional Raman spectra into two-dimensional Raman images through a certain encoding method, making it easier to load the spectra into artificial intelligence models. Four encoding methods are proposed for the transformation of one-dimensional Raman data, and the deep learning models used for training and corresponding evaluation parameters are given. Furthermore, matching transformation methods are provided based on the complexity of the processed spectra.
[0007] Another objective of this invention is to propose a Raman spectroscopy analysis device based on two-dimensional Raman images combined with deep learning.
[0008] To achieve the above objectives, this invention proposes a Raman spectroscopy analysis method based on two-dimensional Raman images combined with deep learning, comprising:
[0009] Acquire raw Raman spectral data of biological samples and various Raman spectral data conversion methods;
[0010] Based on the attribute characteristics of the original Raman spectral data, the complexity of the original Raman spectral data is judged to obtain a complexity judgment result;
[0011] Based on the complexity judgment result and the preset conversion matching mechanism, the optimal Raman spectral data conversion method is matched from the various Raman spectral data conversion methods, and the preprocessed original Raman spectral data is converted to obtain a two-dimensional Raman image sample set.
[0012] A deep learning network model is trained based on the two-dimensional Raman image sample set to obtain a trained deep learning network model. The actual two-dimensional Raman image set is then input into the trained deep learning network model for image classification to obtain the image classification result.
[0013] The Raman spectroscopy analysis method based on two-dimensional Raman images combined with deep learning according to embodiments of the present invention may also have the following additional technical features:
[0014] Furthermore, in one embodiment of the present invention, the step of training a deep learning network model based on sample two-dimensional Raman images to obtain a trained deep learning network model includes:
[0015] The two-dimensional Raman image sample set is divided into a training set and a test set;
[0016] A deep learning network model is constructed based on the two-dimensional Raman image sample set. The training set is input into the deep learning network model for training. The performance of the deep learning network model is tested using the test set. The trained deep learning network model is obtained based on the performance test results.
[0017] Furthermore, in one embodiment of the present invention, before converting the preprocessed raw Raman spectral data to obtain a two-dimensional Raman image sample set, the method further includes:
[0018] The raw Raman spectral data is preprocessed to obtain preprocessed data; wherein, the data preprocessing includes smoothing and baseline processing;
[0019] The preprocessed data is output as data in a preset format, and the preset format data is converted into a beam sequence.
[0020] Furthermore, in one embodiment of the present invention, the multiple Raman spectral data conversion methods include: a recursive plot of the spectrum, a Grammy angle deformation field of the spectrum, a spectrogram of the spectrum, and a Markov transition field of the spectrum; the step of matching the optimal Raman spectral data conversion method from the multiple Raman spectral data conversion methods based on the complexity judgment result and a preset conversion matching mechanism includes:
[0021] If the complexity of the original Raman spectral data is determined to be greater than a preset threshold, then the recursive graph of the spectrum and the Grammy angle deformation field of the spectrum are matched from the various Raman spectral data conversion methods.
[0022] Conversely, from the various Raman spectral data conversion methods, we can match the spectrogram of the spectrum and the Markov transition field of the spectrum.
[0023] Furthermore, in one embodiment of the present invention, the conversion method of the recursive graph of the spectrum includes: scaling the beam sequence to obtain a first preset interval sequence, reconstructing the first preset interval sequence using the Taken delay method, and generating a recursive graph of the Raman spectrum based on the reconstructed sequence.
[0024] The conversion method of the Grammy angle deformation field of the spectrum includes: scaling the beam sequence to obtain a second preset interval sequence, and performing matrix transformation on the second preset interval sequence to generate the Grammy angle deformation field of the Raman spectrum.
[0025] Furthermore, in one embodiment of the present invention, the method for converting the spectrogram of the spectrum includes: scaling the beam sequence to obtain a third preset interval sequence, segmenting the third preset interval sequence, performing a short-time Fourier transform on the segmented sequence, and performing a discrete Fourier transform on the sequence after the short-time Fourier transform to generate the spectrogram of the spectrum.
[0026] The conversion method of the Markov transition field of the spectrum includes: scaling the beam sequence to obtain a fourth preset interval sequence; dividing the fourth preset interval sequence into multiple quantile bins according to the value range of the fourth preset interval sequence; constructing a Markov transition matrix based on the multiple quantile bins; and normalizing the Markov transition matrix to generate the Markov transition field of the Raman spectrum.
[0027] To achieve the above objectives, another aspect of the present invention proposes a Raman spectroscopy analysis device based on two-dimensional Raman images combined with deep learning, comprising:
[0028] The data acquisition module is used to acquire raw Raman spectral data of biological samples and various Raman spectral data conversion methods;
[0029] The feature judgment module is used to judge the complexity of the original Raman spectral data based on the attribute features of the original Raman spectral data and obtain a complexity judgment result;
[0030] The conversion matching module, based on the complexity judgment result and the preset conversion matching mechanism, matches the optimal Raman spectral data conversion method from the various Raman spectral data conversion methods, and converts the preprocessed original Raman spectral data to obtain a two-dimensional Raman image sample set;
[0031] The image classification module is used to train a deep learning network model based on the two-dimensional Raman image sample set, obtain a trained deep learning network model, input the actual two-dimensional Raman image set into the trained deep learning network model to perform image classification, and obtain the image classification result.
[0032] This invention presents a Raman spectroscopy analysis method and apparatus based on two-dimensional Raman images combined with deep learning. When converting one-dimensional Raman spectral data into two-dimensional images, the one-dimensional Raman spectrum is transformed into a two-dimensional Raman image through a certain encoding method. This image is more suitable for training deep learning models. Based on the aforementioned two-dimensional Raman images, four different encoding methods are provided to convert one-dimensional Raman data into two-dimensional Raman images. During the integration of these four two-dimensional Raman images with deep learning, the appropriate conversion process can be matched according to the complexity of the spectral data. The combination of two-dimensional Raman images and deep learning achieves high-accuracy recognition and improves the ability to identify samples.
[0033] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0034] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:
[0035] Figure 1 This is a flowchart of a Raman spectroscopy analysis method based on two-dimensional Raman images combined with deep learning according to an embodiment of the present invention;
[0036] Figure 2 This is a schematic diagram of a Raman spectroscopy analysis architecture based on two-dimensional Raman images combined with deep learning according to an embodiment of the present invention;
[0037] Figure 3 This is a recursive graph of the spectrum according to an embodiment of the present invention;
[0038] Figure 4 This is a Grammy angle distortion difference field diagram of the spectrum according to an embodiment of the present invention;
[0039] Figure 5 A spectrogram of the spectrum according to an embodiment of the present invention;
[0040] Figure 6 This is a Markov transition field diagram of the spectrum according to an embodiment of the present invention;
[0041] Figure 7 This is a schematic diagram of a Raman spectroscopy analysis device based on two-dimensional Raman images combined with deep learning according to an embodiment of the present invention. Detailed Implementation
[0042] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0043] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0044] The Raman spectroscopy analysis method and apparatus based on two-dimensional Raman images combined with deep learning, according to embodiments of the present invention, are described below with reference to the accompanying drawings.
[0045] Figure 1 This is a flowchart of a Raman spectroscopy analysis method based on two-dimensional Raman images combined with deep learning, according to an embodiment of the present invention.
[0046] like Figure 1 As shown, the method includes, but is not limited to, the following steps:
[0047] S1 acquires raw Raman spectral data of biological samples and various Raman spectral data conversion methods.
[0048] Specifically, in this embodiment of the invention, biological samples are obtained, such as cells and tissues. The raw Raman spectral data of the biological samples are then acquired.
[0049] Transformation of one-dimensional Raman spectra into two-dimensional Raman diagrams. This invention provides four encoding methods for transforming one-dimensional Raman data into two-dimensional Raman diagrams: recursive plot of the spectrum, Grami angle deformation field of the spectrum, spectrogram of the spectrum, and Markov transition field of the spectrum.
[0050] S2, based on the attribute characteristics of the original Raman spectral data, the complexity of the original Raman spectral data is judged to obtain the complexity judgment result.
[0051] S3, based on the complexity judgment result and the preset conversion matching mechanism, matches the optimal Raman spectral data conversion method from multiple Raman spectral data conversion methods, and converts the preprocessed original Raman spectral data to obtain a two-dimensional Raman image sample set.
[0052] Specifically, the architecture of Raman spectroscopy analysis based on two-dimensional Raman images combined with deep learning in this embodiment of the invention is as follows: Figure 2 As shown.
[0053] Understandably, the acquired raw Raman spectral data is first preprocessed. The preprocessing steps mainly include smoothing and baseline setting. Preferably, the baseline-processed data is output in TXT format. For each preprocessed Raman spectrum, it can be converted and recorded as a beam sequence W = (w1, w2, ..., w...). n ), where n is the wave number sequence number.
[0054] Furthermore, for spectral data collected from samples, the analysis can be based on the range of the spectrum, the intensity variation, and the number of peaks. For spectral data with a wide range, numerous intensity variations, and many peaks, the recursive plot of the spectrum or the Grammy angle deformation field of the spectrum can be selected. Conversely, the spectrogram of the spectrum or the Markov transformation field of the spectrum can be selected for conversion.
[0055] As an example, if the complexity of the original Raman spectral data is determined to be greater than a preset threshold, two conversion methods are selected from multiple Raman spectral data conversion methods: the recursive graph of the spectrum and the Grami angle deformation field of the spectrum. Conversely, two conversion methods are selected from multiple Raman spectral data conversion methods: the spectrogram of the spectrum and the Markov transition field of the spectrum.
[0056] Specifically, the steps for the above four transformation methods are as follows:
[0057] like Figure 3 As shown, the spectrum recurrence relation (SRP) is as follows:
[0058] a) For the preprocessed Raman spectrum sequence W = (w1, w2, ..., w n ), and scale it to the [0,1] interval, that is,
[0059]
[0060] b) Reconstruct the scaled sequence as follows: The reconstruction method utilizes the Taken delay approach, namely:
[0061]
[0062] c) Generate a recursive graph of the Raman spectrum. The recursive graph R of the spectrum can be generated from the reconstructed spectrum. Represented as:
[0063]
[0064] like Figure 4As shown, the spectral Grammy angle deformation field (SGACF) is:
[0065] a) Scale the preprocessed Raman spectral data. Scale the preprocessed Raman spectral series W = (w1, w2, ..., w...). n ), scale the interval [-1,1], and use This means, that is:
[0066]
[0067] b) Generating the Grammy angle deformation field for the Raman spectrum. (Based on scaling...) After performing a matrix transformation, the Grammy angle deformation field of the Raman spectrum can be expressed as:
[0068]
[0069]
[0070] The present invention provides two different Grammi angular deformation fields: GACSF (Gramian Angular Conversion Summation Field) is the Grammi angular deformation sum field, and GACDF (Gramian Angular Conversion Difference Field) is the Grammi angular deformation difference field. Both can generate spectral Grammi angular deformation fields. Figure 4 It demonstrates Grami angle deformation and field.
[0071] like Figure 5 As shown, the spectrum spectrum (SSTFT):
[0072] a) For the preprocessed Raman spectrum sequence W = (w1, w2, ..., w n ), and scale it to the [0,1] interval, that is,
[0073]
[0074] b) Spectral Division. For the scaled spectrum x(t), t = 0, 1, ..., T-1, where t is the Raman frequency shift and T is the length of the acquired Raman spectrum, the spectrum x(t) can be divided into segments, which can be represented as x n (m), n=0,1,···,r-1, where n is the number of segment sequences, r is the total number of segments, and m is the Raman frequency shift of segment length n.
[0075] c) Perform a short-time Fourier transform. For the above x(t), perform a short-time Fourier transform, i.e.:
[0076]
[0077] w(n) is a window function.
[0078] d) Perform Discrete Fourier Transform. Further discretizing x(n) yields the Discrete-Time Fourier Transform (DTFT) of the Raman scattering intensity x(n), i.e.:
[0079]
[0080] The Discrete Fourier Transform can also be expressed as:
[0081]
[0082] Where 0≤k≤N-1, ω=2πk / N.
[0083] e) Generate the Raman spectrogram. |X(n,k)| is the short-time intensity spectrum estimate of x(n). Define P(n,k) as the energy density function (or power spectrum function) of the intensity spectrum at a Raman frequency shift of m, i.e.:
[0084] P(n,k)=|X(n,k)| 2 =(X(n,k))×(conj(X(n,k)))
[0085] This yields a Raman spectrogram, a two-dimensional image with wavenumber n on the x-axis, frequency k on the y-axis, and chromaticity represented by P(n,k). Color depth is measured in dB and expressed as 10*lg(P(n,k)). Figure 5 As shown.
[0086] like Figure 6 As shown, the Markov transition field (SMTF) of the spectrum:
[0087] a) For the preprocessed Raman spectrum sequence W = (w1, w2, ..., w n First, scale to the [0,1] interval, that is,
[0088]
[0089] b) Divide the wavenumber sequence into quantile bins. Based on its value range, it is divided into Q quantile bins, and each data point wi belongs to a unique q bin. j (j∈[1,Q]);
[0090] c) Construct the Markov transition matrix. Construct a Markov transition matrix V of size Q×Q, where v i,j Represented as quantile bin q j The elements in are in the quantile bin q i The probability of an element following in the sequence, i.e.:
[0091] v ij =P(x t ∈q i |x t-1 ∈q j )
[0092] d) The Markov transition field that generates the spectrum. (The ∑...) j v ij Normalizing by 1, we can obtain the Markov transition field M of the spectrum, that is:
[0093]
[0094] S4. Train a deep learning network model based on a two-dimensional Raman image sample set to obtain a trained deep learning network model. Input the actual two-dimensional Raman image set into the trained deep learning network model to perform image classification and obtain the image classification result.
[0095] It is understood that, in this embodiment of the invention, the two-dimensional Raman image sample set is divided into a training set and a test set; a deep learning network model is constructed based on the two-dimensional Raman image sample set, the training set is input into the deep learning network model for training, and the performance of the deep learning network model is tested using the test set, and the trained deep learning network model is obtained based on the performance test results.
[0096] Specifically, deep learning model creation involves using four different images—transformed from one-dimensional spectral data into two-dimensional Raman images—as input for deep learning. Different model layers and their numbers are set for each image, including convolutional layers, batch normalization layers, fully connected layers, pooling layers, and activation function layers.
[0097] Deep learning model training. The model training and validation process uses a 5x cross-validation method to evaluate the stability of each classification model and prevent overfitting.
[0098] Deep learning model testing. Selecting m data points from a set of two-dimensional Raman image samples as the training set, then the remaining N... t =NN m The selected test set data is used to test the model trained in the previous step, and the accuracy, sensitivity, and specificity of sample classification are finally obtained. At the same time, the subject worker characteristic curve of the model is obtained, which is used to evaluate the recognition effect based on two-dimensional Raman diagrams combined with deep learning.
[0099] The Raman spectroscopy analysis method based on two-dimensional Raman images combined with deep learning according to embodiments of the present invention can match a suitable conversion process according to the complexity of the spectral data. The combination of two-dimensional Raman images and deep learning achieves high-accuracy identification and improves the ability to distinguish samples.
[0100] To achieve the above embodiments, such as Figure 7 As shown, this embodiment also provides a Raman spectroscopy analysis device 10 based on two-dimensional Raman images combined with deep learning. The device 10 includes: a data acquisition module 100, a feature judgment module 200, a conversion matching module 300, and an image classification module 400.
[0101] The data acquisition module 100 is used to acquire raw Raman spectral data of biological samples and various Raman spectral data conversion methods;
[0102] The feature judgment module 200 is used to judge the complexity of the original Raman spectral data based on the attribute features of the original Raman spectral data and obtain the complexity judgment result.
[0103] The conversion matching module 300, based on the complexity judgment result and the preset conversion matching mechanism, matches the optimal Raman spectral data conversion method from multiple Raman spectral data conversion methods, and converts the preprocessed original Raman spectral data to obtain a two-dimensional Raman image sample set.
[0104] The image classification module 400 is used to train a deep learning network model based on a two-dimensional Raman image sample set, obtain a trained deep learning network model, input the actual two-dimensional Raman image set into the trained deep learning network model for image classification, and obtain the image classification result.
[0105] Furthermore, the image classification module 400 described above is also used for:
[0106] The two-dimensional Raman image sample set is divided into a training set and a test set;
[0107] A deep learning network model is constructed based on a two-dimensional Raman image sample set. The training set is input into the deep learning network model for training, and the performance of the deep learning network model is tested using a test set. The trained deep learning network model is obtained based on the performance test results.
[0108] Furthermore, the aforementioned device 10 also includes:
[0109] The data preprocessing module is used to preprocess the raw Raman spectral data to obtain preprocessed data; the data preprocessing includes smoothing and baseline processing.
[0110] The format conversion module is used to output preprocessed data in a preset format and convert the preset format data into a beam sequence.
[0111] Furthermore, the aforementioned various Raman spectral data conversion methods include: recursive plots of the spectrum, Grammy angle deformation fields of the spectrum, spectrograms of the spectrum, and Markov transition fields of the spectrum; the aforementioned conversion matching module 300 is also used for:
[0112] If the complexity of the original Raman spectral data is determined to be greater than the preset threshold, two conversion methods are selected from multiple Raman spectral data conversion methods: the recursive graph of the spectrum and the Grammy angle deformation field of the spectrum.
[0113] Conversely, from various Raman spectral data conversion methods, we can match two conversion methods: spectrogram of the spectrum and Markov transition field of the spectrum.
[0114] The Raman spectroscopy analysis device based on two-dimensional Raman images combined with deep learning according to embodiments of the present invention can match a suitable conversion process according to the complexity of the spectral data. The combination of two-dimensional Raman images and deep learning achieves high-accuracy identification and improves the ability to identify samples.
[0115] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.
[0116] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0117] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.
Claims
1. A method of Raman spectrum analysis based on two-dimensional Raman mapping combined with deep learning, characterized in that, Includes the following steps: The system acquires raw Raman spectral data of biological samples and various Raman spectral data conversion methods, including recursive graph conversion, Grami angle deformation field conversion, spectrogram conversion, and Markov transition field conversion. The raw Raman spectral data were preprocessed to obtain a one-dimensional wavenumber sequence; The complexity of a one-dimensional wavenumber sequence is determined by analyzing the acquisition range, intensity variation, and number of peaks, and the complexity assessment result is obtained. Based on the complexity assessment results and the preset conversion matching mechanism, the optimal Raman spectral data conversion method is matched from the various Raman spectral data conversion methods: If the complexity assessment result is that the complexity is greater than the preset threshold, then the two conversion methods of matching the recursive graph of the spectrum and the Grammy angle deformation field of the spectrum are used. Conversely, the two conversion methods are matched: spectrogram of the spectrum and Markov transition field of the spectrum. The one-dimensional wavenumber sequence is transformed using the matched transformation method to obtain the corresponding two-dimensional Raman plot sample set: If the conversion method of the recursive graph of the spectrum is matched, the conversion steps include: scaling the wavenumber sequence to obtain a first preset interval sequence, reconstructing the first preset interval sequence using the Taken delay method, and generating a recursive graph of the Raman spectrum based on the reconstructed sequence. If the conversion method of the Grammy angle deformation field of the spectrum is matched, the conversion steps include: scaling the wavenumber sequence to obtain a second preset interval sequence, and performing matrix transformation on the second preset interval sequence to generate the Grammy angle deformation field of the Raman spectrum. If the conversion method of the spectrogram of the spectrum is matched, the conversion steps include: scaling the wavenumber sequence to obtain a third preset interval sequence, segmenting the third preset interval sequence, performing a short-time Fourier transform on the segmented sequence, and performing a discrete Fourier transform on the sequence after the short-time Fourier transform to generate the spectrogram of the spectrum. If the conversion method of the Markov transition field of the spectrum is matched, the conversion steps include: scaling the wavenumber sequence to obtain a fourth preset interval sequence, dividing the fourth preset interval sequence into multiple quantile bins according to the value range of the fourth preset interval sequence, constructing a Markov transition matrix based on the multiple quantile bins, and normalizing the Markov transition matrix to generate the Markov transition field of the Raman spectrum. A deep learning network model is trained based on a two-dimensional Raman image sample set. The trained deep learning network model is then input into the actual two-dimensional Raman image set to perform image classification and obtain the image classification results.
2. The method of claim 1, wherein, The process of training a deep learning network model based on a two-dimensional Raman image sample set to obtain a trained deep learning network model includes: The two-dimensional Raman image sample set is divided into a training set and a test set; A deep learning network model is constructed based on the two-dimensional Raman image sample set. The training set is input into the deep learning network model for training. The performance of the deep learning network model is tested using the test set. The trained deep learning network model is obtained based on the performance test results.
3. The method of claim 2, wherein, The preprocessing of the raw Raman spectral data to obtain a one-dimensional wavenumber sequence includes: The raw Raman spectral data is smoothed and baseline-stretched to obtain preprocessed data; the preprocessed data is output as data in a preset format and then converted into a wavenumber sequence.
4. A Raman spectroscopy analysis device based on two-dimensional Raman mapping combined with deep learning using the method of claim 1, characterized in that, include: The data acquisition module is used to acquire raw Raman spectral data of biological samples and various Raman spectral data conversion methods; The feature judgment module is used to judge the complexity of the original Raman spectral data based on the attribute features of the original Raman spectral data and obtain a complexity judgment result; The conversion matching module, based on the complexity judgment result and the preset conversion matching mechanism, matches the optimal Raman spectral data conversion method from the various Raman spectral data conversion methods, and converts the preprocessed original Raman spectral data to obtain a two-dimensional Raman image sample set; The image classification module is used to train a deep learning network model based on the two-dimensional Raman image sample set, obtain a trained deep learning network model, input the actual two-dimensional Raman image set into the trained deep learning network model to perform image classification, and obtain the image classification result.
5. The apparatus of claim 4, wherein, The image classification module is also used for: The two-dimensional Raman image sample set is divided into a training set and a test set; A deep learning network model is constructed based on the two-dimensional Raman image sample set. The training set is input into the deep learning network model for training. The performance of the deep learning network model is tested using the test set. The trained deep learning network model is obtained based on the performance test results.
6. The apparatus of claim 5, wherein, The device further includes: The data preprocessing module is used to smooth and perform baseline processing on the raw Raman spectral data to obtain preprocessed data. The format conversion module is used to output the preprocessed data as preset format data and convert the preset format data into wavenumber sequences.
7. The apparatus of claim 6, wherein, The various Raman spectral data conversion methods include: recursive plots of the spectrum, Grammy angle deformation fields of the spectrum, spectrograms of the spectrum, and Markov transition fields of the spectrum; the conversion matching module is also used for: If the complexity of the original Raman spectral data is determined to be greater than a preset threshold, then the recursive graph of the spectrum and the Grammy angle deformation field of the spectrum are matched from the various Raman spectral data conversion methods. Conversely, from the various Raman spectral data conversion methods, we can match the spectrogram of the spectrum and the Markov transition field of the spectrum.