Spectral analysis transfer learning method and system based on central symmetry

Abstract

The invention belongs to but not limited to the technical field of spectral analysis, and discloses a spectral analysis transfer learning method and system based on central symmetry, and the method comprises the steps: collecting spectral data of a source domain and a target domain, and making a component information table; preprocessing the spectral data of the source domain and the target domain, and performing feature extraction and enhancement on the spectral data; constructing a source domain deep learning model, pre-training the deep learning model by using the preprocessed source domain spectral data set, and optimizing the performance of the deep learning model on source domain data by adjusting model parameters; in the transfer learning process, the feature center of the source domain data is calculated, and the feature center of the target data is adjusted when the source domain data is transferred to the target data; adding a new layer in the migration network, wherein the new layer comprises an additional convolutional layer or a full connection layer; and retraining and optimizing the migrated deep learning model based on the target domain spectral data set. According to the method, the problems of high dimension and sparsity of the spectral data are solved, and the efficiency and effect of transfer learning are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to, but is not limited to, the field of spectral analysis technology, and particularly relates to a spectral analysis transfer learning method and system based on central symmetry. Background Art

[0002] Prior Art 1: A Raman spectroscopy intelligent analysis method based on transfer learning (CN201910616759.5); Prior Art 2: A method for analyzing Mars LIBS data by transfer learning based on a pre-trained deep CNN model (CN202410265821.1); Both Prior Art 1 and Prior Art 2 perform spectral analysis based on the framework of transfer learning. These two technologies both preprocess the original data, but the subsequent transfer strategies adopted are relatively simple, directly applying the common transfer learning methods in the field of computer vision. Their methods are limited to freezing certain fixed neural network layers and only training a few subsequent layers. This strategy ignores the essential differences between spectral analysis and computer vision in data processing and feature extraction, and fails to fully explore the potential of transfer learning in the field of spectral analysis.

[0003] Specifically, although Prior Art 1 and Prior Art 2 demonstrate the preliminary application of transfer learning in spectral analysis, they have obvious limitations. First, they fail to customize transfer learning strategies according to the unique properties of spectral data (such as high dimensionality, sparsity, etc.), resulting in mediocre performance of the model when adapting to spectral data. Second, the simple layer freezing and partial training strategies ignore the key information that may exist in spectral data, restricting the learning ability and generalization performance of the model.

[0004] In view of the above analysis, the technical problems urgently to be solved in the prior art are: the problems of high dimensionality and sparsity of spectral data, and the insufficient adaptability of the model during the transfer learning process. Existing methods often adopt a centralized spectral analysis transfer strategy, the core of which is to center-align the source data during the data processing stage, that is, by calculating the feature center after dimensionality reduction of the source data, and when migrating to the target data, adjusting the feature center of the target data to match the source data center. Although this method alleviates the problem of spectral feature complexity to a certain extent, there is still room for further improvement. Summary of the Invention

[0005] In view of the problems existing in the prior art, the present invention provides a spectral analysis transfer learning method and system based on central symmetry.

[0006] The present invention is implemented as follows. A spectral analysis transfer learning method based on central symmetry includes the following steps: Step 1, collect spectral data of the source domain and the target domain and make a component information table.

[0007] Step 2: Preprocess the spectral data of the source domain and the target domain, and perform feature extraction and enhancement on the spectral data.

[0008] Step 3: Construct a deep learning model for the source domain, pre-train the deep learning model using the preprocessed source domain spectral data set, and optimize its performance on the source domain data by adjusting the model parameters.

[0009] Step 4: During the transfer learning process, calculate the feature center of the source domain data and adjust the feature center of the target data when transferring to the target data; add new layers to the transfer network, including additional convolutional layers or fully connected layers; based on the target domain spectral data set, re-train and optimize the transferred deep learning model.

[0010] Furthermore, the specific steps of Step 1 include: Collect spectral data of the source domain and the target domain: Collect a large-scale source domain spectral data set containing known substance composition information and a target domain spectral data set.

[0011] Create a composition information table: Statistically analyze the content of various chemical components in the samples of the source domain and the target domain, create a total composition information table, and determine the composition label vector for each sample.

[0012] Furthermore, the specific preprocessing of the data in Step 2 includes: Spectral data preprocessing: Preprocess the spectral data of the source domain and the target domain, including steps such as dark background subtraction, noise filtering, background baseline removal, wavelength calibration, channel stitching, and instrument radiation response correction, to ensure the quality and consistency of the spectral data.

[0013] Feature extraction and enhancement: Use methods such as interpolation processing, baseline correction, and normalization to perform feature extraction and enhancement on the spectral data, providing high-quality input data for the subsequent deep learning model.

[0014] Furthermore, the construction of the deep learning model for the source domain in Step 3 includes: According to the characteristics of the source domain spectral data, construct a suitable deep learning model structure, such as a deep convolutional neural network (CNN) or a deep neural network (DNN); the model should include batch normalization layers, convolutional layers, pooling layers, fully connected layers, etc., and determine the parameters and activation functions of each layer.

[0015] Furthermore, in Step 3, during the pre-training process, optimization algorithms such as Adam can be used to accelerate the training process and improve the model performance.

[0016] Furthermore, in Step 4, during the re-training process, the parameters of the source domain pre-trained model can be frozen, and only the newly added layers are re-trained; through repeated testing and optimization, the ideal parameters under the current target domain data set are determined.

[0017] Another object of the present invention is to provide a spectral analysis transfer learning system based on central symmetry for a spectral analysis transfer learning method based on central symmetry, including: A data preparation module that collects spectral data of the source domain and the target domain and makes a composition information table.

[0018] A data preprocessing module that preprocesses the spectral data of the source domain and the target domain, and extracts and enhances features of the spectral data.

[0019] Construction of a source domain deep learning model, constructing a source domain deep learning model, pre-training the deep learning model using the preprocessed source domain spectral data set, and optimizing its performance on the source domain data by adjusting model parameters.

[0020] A transfer learning module that, during the transfer learning process, calculates the feature center of the source domain data and adjusts the feature center of the target data when transferring to the target data; adds new layers to the transfer network, including additional convolutional layers or fully connected layers; and re-trains and optimizes the transferred deep learning model based on the target domain spectral data set.

[0021] Another object of the present invention is to provide a computer device, which includes a memory and a processor. When a computer program stored in the memory is executed by the processor, the processor executes the steps of the spectral analysis transfer learning method based on central symmetry.

[0022] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which when executed by a processor, causes the processor to execute the steps of the spectral analysis transfer learning method based on central symmetry.

[0023] Another object of the present invention is to provide an information data processing terminal, which includes the spectral analysis transfer learning system based on central symmetry.

[0024] Combined with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows: First, the present invention proposes a spectral analysis transfer learning method based on central symmetry, which not only inherits the basic idea of centralized transfer but also significantly optimizes it on this basis. Specifically, during the transfer learning process, the present invention not only achieves the alignment of feature centers but also introduces a more flexible transfer learning strategy, significantly improving the model adaptability and generalization ability. Specifically: Flexible migration strategy: By adding new layers and flexibly adjusting the migration layers, the present invention overcomes the problem of insufficient model adaptability in traditional transfer learning methods. The newly added layers can increase the amount of knowledge during the migration process, improve the adaptability of the model to the target domain data, and thus significantly reduce the forgetting phenomenon of the model during migration.

[0025] Feature center alignment optimization: On the basis of feature center alignment, the present invention further optimizes the transfer learning strategy, enabling the model to better capture the internal connection between the source domain and target domain spectral data and improving the effect of transfer learning.

[0026] Second, the central symmetric spectral migration algorithm of the present invention is expected to bring significant benefits and commercial value. By improving the accuracy and efficiency of spectral data migration, the algorithm can help researchers identify and analyze substance components faster, thus accelerating the scientific research process and improving production efficiency. Commercially, after lightweighting the algorithm, it is expected to be applied to instruments and promote the upgrade of related instrument equipment, enhance the market competitiveness of products, and create new profit opportunities for enterprises.

[0027] The technical solution of the present invention solves the technical problems that people have been eager to solve but have never succeeded in: the field of spectral analysis and substance identification has always faced the problems of inaccurate and unstable spectral data migration. Traditional spectral migration algorithms are often interfered by various factors, resulting in unsatisfactory results. The central symmetric spectral migration algorithm of the present invention effectively solves this problem through unique algorithm design, providing people with a more reliable means of spectral data analysis.

[0028] The technical solution of the present invention overcomes the technical prejudice: In the past, there may have been some technical prejudices in the field of spectral data processing, believing that certain types of spectral data are difficult to migrate or match effectively. However, the central symmetric spectral migration algorithm of the present invention has been verified through practice, proving that even complex spectral data can achieve accurate migration and matching through reasonable algorithm design. This achievement has successfully overcome the technical prejudice, bringing new ideas and possibilities to the field of spectral data processing and analysis. Brief Description of the Drawings

[0029] Figure 1 is the flowchart of the spectral analysis transfer learning method based on central symmetry provided by the embodiment of the present invention.

[0030] Figure 2 is the structural diagram of the migration network provided by the embodiment of the present invention.

[0031] Figure 3 is the structural diagram of the spectral analysis transfer learning system based on central symmetry provided by the embodiment of the present invention.

[0032] Figure 4It is a visualization display graph after processing a certain spectral data of corn and mango respectively on the left and right provided by an embodiment of the present invention. Detailed implementation manners

[0033] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0034] As Figure 1 shown, a spectral analysis transfer learning method based on central symmetry provided by an embodiment of the present invention includes the following steps: Step 1, data preparation.

[0035] Collect source domain and target domain spectral data: Collect a large-scale source domain spectral data set and a target domain spectral data set containing known substance composition information.

[0036] Make a composition information table: Statistically analyze the content of various chemical components in the samples of the source domain and the target domain, and make a total composition information table to determine the composition label vector of each sample.

[0037] Step 2, data preprocessing.

[0038] Spectral data preprocessing: Preprocess the source domain and target domain spectral data, including steps such as dark background deduction, noise filtering, background baseline removal, wavelength calibration, channel stitching, and instrument radiation response correction, to ensure the quality and consistency of the spectral data.

[0039] Feature extraction and enhancement: Use methods such as interpolation processing, baseline correction, and normalization to extract and enhance features from the spectral data, providing high-quality input data for subsequent deep learning models.

[0040] Step 3, construct a source domain deep learning model.

[0041] Design the model structure: According to the characteristics of the source domain spectral data, design a suitable deep learning model structure, such as a deep convolutional neural network (CNN) or a deep neural network (DNN). The model should include a batch normalization layer, a convolutional layer, a pooling layer, a fully connected layer, etc., and determine the parameters and activation functions of each layer.

[0042] Model pre-training: Use the preprocessed source domain spectral data set to pre-train the deep learning model, and optimize its performance on the source domain data by adjusting the model parameters. During the pre-training process, optimization algorithms such as Adam can be used to accelerate the training process and improve the model performance.

[0043] Step 4, transfer learning strategy.

[0044] Feature center alignment: During the transfer learning process, first, the alignment of the spectral data feature centers between the source domain and the target domain is achieved to alleviate the problem of spectral feature complexity. This can be realized by calculating the feature center of the source domain data and adjusting the feature center of the target data when transferring to the target data.

[0045] Adding new layers and flexible transfer: Different from the traditional method of simply freezing the transfer layers, the present invention adds new layers, such as additional convolutional layers or fully connected layers, to the transfer network. These new layers can increase the amount of knowledge during the transfer process, improve the adaptability and generalization ability of the model. At the same time, by flexibly adjusting the number and position of the transfer layers, the forgetting phenomenon that the network may encounter during the transfer can be reduced.

[0046] Model retraining and optimization: Based on the spectral data set of the target domain, the transferred deep learning model is retrained and optimized. During the retraining process, the parameters of the pre-trained model in the source domain can be frozen, and only the newly added layers are retrained. Through repeated testing and optimization, the ideal parameters under the current target domain data set are determined.

[0047] The spectral analysis transfer learning method based on central symmetry proposed by the present invention aims to solve the possible distribution difference problem in component recognition between the spectral data of the source domain and the target domain. First, this method collects the spectral data of the source domain and the target domain, and makes corresponding component information tables to systematically and label the component information of each sample. The core of this step is to ensure the consistency of the data source and the label, so that the subsequent model can accurately learn and adapt to the spectral characteristics of different domains.

[0048] In the data preprocessing stage, in order to improve the quality and consistency of the spectral data, a series of operations such as dark background subtraction, noise filtering, and background baseline removal are carried out. These preprocessing methods ensure the purity of the spectral data during deep learning training. In addition, through feature extraction and enhancement means such as interpolation processing, baseline correction, and normalization, the feature representation ability of the spectral data is improved, so that the model can receive high-quality input data, laying a solid foundation for subsequent modeling.

[0049] Then, in order to build a basic deep learning model on the source domain data, this method designs a deep convolutional neural network (CNN) or deep neural network (DNN) structure suitable for spectral data. The model includes different levels of operations such as batch normalization, convolution, pooling, and fully connected, ensuring that the model has sufficient flexibility and accuracy in capturing the complex features of spectral data. In the model pre-training, by adopting optimization algorithms such as the Adam optimization algorithm to accelerate the adjustment of model parameters, the performance of the model on the source domain data is optimized.

[0050] The transfer learning strategy is the core of this method. Through the alignment of feature centers and the design of flexible transfer of new layers, it ensures that the source domain model can be effectively transferred to the target domain data. Specifically, by calculating the source domain feature center and adjusting the target domain feature center, the center alignment of the two-domain features is achieved, effectively alleviating the problems brought by cross-domain feature differences. In addition, during the transfer process, additional convolutional layers or fully connected layers are added. Different from the traditional method of freezing layers, the addition of these new layers can improve the adaptability of the transfer model.

[0051] During the retraining and optimization process of the model after transfer learning, this method adopts the strategy of gradually freezing the parameters of the source domain model and only retrains the newly added layers. This strategy avoids the model from overly forgetting the source domain knowledge and can well adapt to the data distribution of the target domain. Through continuous testing and optimization, the parameters are gradually adjusted to ensure that the performance of the transfer model on the target domain spectral dataset reaches the best state.

[0052] Through a series of innovative combinations of orderly data preprocessing, deep model construction, and transfer learning strategies, the present invention can achieve precise compositional analysis transfer learning between different spectral data domains.

[0053] Embodiment 1 A spectral analysis transfer learning method based on central symmetry includes the following steps: S101, Data preparation Collect spectral data.

[0054] Corn dataset: 80 samples, channel range 1100 - 2498, and the number of channels after processing is 700.

[0055] Mango dataset: 11,691 samples, channel range 300 - 1200, and the number of channels after processing is 306.

[0056] Make a compositional information table: Count the chemical component contents for each sample and make a compositional information table to determine the regression label vector (4 components for corn and 1 component for mango).

[0057] S102, Data preprocessing Spectral data preprocessing: Perform dark background subtraction, noise filtering, and background baseline removal on the data.

[0058] Feature extraction and enhancement: Perform interpolation processing, baseline correction, and normalization to unify the data in the 0 - 2500 channel range and channels with an interval of 1 to provide high-quality input data.

[0059] S103, Construct the source domain deep learning model As Figure 2 shown, the model structure includes three-layer network: 1. A fully connected layer of 2500×50.

[0060] 2. A ReLU activation layer.

[0061] 3. A fully connected layer of 50×output dimension (for corn, the output dimension is 4; for mango, the output dimension is 1).

[0062] Model pre-training: Use the pre-processed source domain spectral data (possibly mango data because of the large amount of data) for pre-training.

[0063] Adopt the Adam optimization algorithm.

[0064] As Figure 4 shown, on the left and right are the visualizations of the processed spectral data of corn and mango respectively.

[0065] S104, Transfer learning strategy Feature center alignment: Calculate the feature center of the source domain data and adjust the feature center of the target data to achieve alignment.

[0066] Adding new layers and flexible transfer: Add new layers (such as additional convolutional layers or fully connected layers) to the transfer network to increase the amount of knowledge; flexibly adjust the number and position of the transfer layers to reduce the phenomenon of forgetting.

[0067] Model re-training and optimization: Re-train the transferred model based on the target domain spectral data set (such as corn data); freeze the parameters of the source domain pre-trained model and only re-train the newly added layers.

[0068] As Figure 2 shown, the network structure for adding the training target data: 1. A fully connected layer of 50×2048.

[0069] 2. A ReLU activation layer.

[0070] 3. A fully connected layer of 2048×50.

[0071] Data analysis and evaluation.

[0072] Model performance evaluation: Use appropriate evaluation metrics (such as mean square error MSE, predicted average relative error MREP, etc.) to evaluate the performance of the model after transfer learning. By comparing the performance of the transfer learning model and the non-transfer learning model on the target domain data, verify the effectiveness and superiority of the present invention.

[0073] The present invention selects to add a new layer to the migration network. This design can make full use of the particularity of spectral data, effectively increase the amount of knowledge in the migration process, and at the same time reduce the forgetting phenomenon that the network may encounter during migration, thereby improving the generalization ability and accuracy of the model on the target data.

[0074] As Figure 3 shown, a spectral analysis transfer learning system based on central symmetry provided by an embodiment of the present invention includes: A data preparation module that collects source domain and target domain spectral data and makes a composition information table.

[0075] A data preprocessing module that preprocesses the source domain and target domain spectral data, and performs feature extraction and enhancement on the spectral data.

[0076] Construction of a source domain deep learning model, constructing a source domain deep learning model, pre-training the deep learning model using the preprocessed source domain spectral data set, and optimizing its performance on the source domain data by adjusting model parameters.

[0077] A transfer learning module that, during the transfer learning process, calculates the feature center of the source domain data and adjusts the feature center of the target data when migrating to the target data; adds a new layer to the transfer network, including an additional convolutional layer or fully connected layer; and re-trains and optimizes the transferred deep learning model based on the target domain spectral data set.

[0078] The models trained on the mango data set are obtained through three different methods, namely ordinary model training, ordinary transfer, and transfer method based on centralization. The results are shown in the following table.

[0079] 1. R2 (coefficient of determination) analysis: Non-transfer model: The R2 value is 0.863, indicating that the model has a good fitting degree for the data, but there is still room for improvement.

[0080] Ordinary transfer: The R2 value is increased to 0.9117, indicating that through the ordinary transfer method, the fitting degree of the model has been significantly improved.

[0081] Transfer method based on centralization: The R2 value is further increased to 0.9376, which is the highest value among the three models. This indicates that the transfer method based on centralization of the present invention can more accurately describe and predict data, and has a better fitting effect.

[0082] 2. RMSEP (root mean square error prediction) analysis: Non-transfer model: The RMSEP value is 0.0996, indicating that the prediction error of the model is relatively large.

[0083] Ordinary migration: The RMSEP value is reduced to 0.0799, indicating that the prediction error of the model has been reduced through the ordinary migration method.

[0084] Migration method based on centralization: The RMSEP value is further reduced to 0.0672, which is the lowest value among the three models. This indicates that the migration method based on centralization of the present invention has higher prediction accuracy and smaller error.

[0085] Comprehensive analysis: From the two indicators of R2 and RMSEP, the migration method based on centralization of the present invention shows significant advantages in terms of data fitting and prediction accuracy compared with the non-migration model and the ordinary migration method. Specifically, the R2 value of this method is higher, indicating that its description and prediction of data are more accurate; at the same time, its RMSEP value is lower, indicating that its prediction error is smaller and it has higher reliability.

[0086] Therefore, it can be considered that the migration method based on centralization of the present invention has achieved a significant improvement in technical effects, providing a more accurate and reliable method for spectral data processing and analysis. This result provides strong evidence of technical effects for the embodiments of the present invention.

[0087] Result comparison table

[0088] It should be noted that the embodiments of the present invention can be implemented through hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those of ordinary skill in the art can understand that the above devices and methods can be implemented using computer-executable instructions and / or included in processor control code, such as provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits of programmable hardware devices such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable logic devices such as field programmable gate arrays, or can be implemented by software executed by various types of processors, or can be implemented by a combination of the above hardware circuits and software such as firmware.

[0089] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention by those skilled in the art within the technical scope disclosed by the present invention shall be covered by the protection scope of the present invention.

Claims

1. A spectral analysis transfer learning method based on central symmetry, characterized in that: The following steps are involved: Step 1: Collect source domain and target domain spectral data and make a component information table; Step 2: preprocess the source domain and target domain spectral data, and extract and enhance the features of the spectral data; Step 3: construct a source domain deep learning model, use the preprocessed source domain spectral dataset to pre-train the deep learning model, and adjust the model parameters to optimize its performance on the source domain data; Step 4: During the transfer learning process, the feature center of the source domain data is calculated, and the feature center of the target data is adjusted when migrating to the target data; Add new layers to the migrated network, including additional convolutional layers or fully connected layers; retrain and optimize the migrated deep learning model based on the target domain spectral dataset.

2. The spectral analysis transfer learning method based on central symmetry according to claim 1, characterized in that: The step 1 specifically includes: Collect source domain and target domain spectral data: Collect large-scale source domain spectral datasets and target domain spectral datasets containing known material composition information; Create a composition information table: Count the contents of various chemical components in the source and target domain samples, create a composition information table, and determine the composition label vector of each sample.

3. The spectral analysis transfer learning method based on central symmetry according to claim 1, characterized in that: The data preprocessing in step 2 specifically includes: Spectral data preprocessing: Preprocess the source and target domain spectral data, including dark background subtraction, noise filtering, background baseline removal, wavelength calibration, channel splicing and instrument radiation response correction to ensure the quality and consistency of spectral data; Feature extraction and enhancement: Interpolation, baseline correction and normalization methods are used to extract and enhance the features of spectral data.

4. The spectral analysis transfer learning method based on central symmetry according to claim 1, characterized in that: Constructing the source domain deep learning model in step 3 includes: constructing a suitable deep learning model structure according to the characteristics of the source domain spectral data, including a deep convolutional neural network (CNN) or a deep neural network (DNN); the model includes a 500×50 fully connected layer, a Relu activation layer, and a 50× output dimension fully connected layer, and determining the parameters and activation functions of each layer; In the step three, during the pre-training process, the Adam optimization algorithm is used to accelerate the training process and improve the model performance.

5. The spectral analysis transfer learning method based on central symmetry according to claim 1, characterized in that: The network to which the training target data is added in step 4 is composed of a 50×2048 fully connected layer, a Relu activation layer and a 2048×50 fully connected layer.

6. The spectral analysis transfer learning method based on central symmetry according to claim 1, characterized in that: In the step 4, during the retraining process, the parameters of the source domain pre-training model are frozen, and only the newly added layers are retrained; through repeated testing and optimization, the ideal parameters under the current target domain data set are determined.

7. A spectral analysis transfer learning system based on central symmetry according to the spectral analysis transfer learning method based on central symmetry as claimed in any one of claims 1 to 6, characterized in that: include: Data preparation module, collects source domain and target domain spectral data and makes a composition information table; The data preprocessing module preprocesses the source domain and target domain spectral data and performs feature extraction and enhancement on the spectral data; Source domain deep learning model construction: construct a source domain deep learning model, use the preprocessed source domain spectral dataset to pre-train the deep learning model, and adjust the model parameters to optimize its performance on the source domain data; The transfer learning module calculates the feature center of the source domain data during the transfer learning process and adjusts the feature center of the target data when migrating to the target data; Add new layers to the migrated network, including additional convolutional layers or fully connected layers; retrain and optimize the migrated deep learning model based on the target domain spectral dataset.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the spectral analysis transfer learning method based on central symmetry as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the processor executes the steps of the spectral analysis transfer learning method based on central symmetry as described in any one of claims 1 to 6.

10. An information data processing terminal, characterized in that: The information data processing terminal includes the spectral analysis transfer learning system based on central symmetry as described in claim 7.

Images