Classification method and system fusing different two-dimensional spectral data

Abstract

The invention relates to a classification method and system fusing different two-dimensional spectral data. The method comprises the following steps: preprocessing one-dimensional spectral data; carrying out two-dimensional processing on the preprocessed one-dimensional spectral data by adopting a plurality of two-dimensional methods to generate a plurality of two-dimensional spectral images; a deep learning model is adopted to extract a feature map from each two-dimensional spectral image, and a plurality of feature maps are generated; performing feature fusion on the plurality of feature maps; evaluating the performance of the model, and optimizing the model according to an evaluation result; and performing classification prediction on the to-be-measured spectral data by using the trained model, and outputting a prediction result. Spectral data characteristics are obtained by fusing different two-dimensional methods, and complementary information of various methods is fully utilized, so that the classification accuracy is improved; by fusing multiple two-dimensional methods, the limitation caused by a single method can be reduced, so that the model has higher adaptability and robustness to different types of spectral data, and the processing efficiency and accuracy are improved.

Description

A classification method and system for fusing different two-dimensional spectral data Technical Field

[0001] This invention relates to the field of spectral data technology, and more specifically, to a classification method and system that integrates different two-dimensional spectral data. Background Technology

[0002] Spectral data plays a vital role in various fields, such as agriculture, medicine, and environmental science. However, traditional spectral data is one-dimensional, making it difficult to directly apply efficient algorithms from the field of machine vision for classification or regression tasks. To overcome this limitation, researchers have proposed several methods to convert one-dimensional spectral data into two-dimensional images, such as Principal Component Analysis (PCA), t-distributed random neighborhood embedding (t-SNE), and Linear Discriminant Analysis (LDA). These methods can convert one-dimensional spectral data into two-dimensional images, facilitating subsequent analysis using image processing techniques such as Convolutional Neural Networks (CNNs). However, different two-dimensionalization methods may capture different features of the spectral data, and a single method may not fully reflect the inherent information of the data. Summary of the Invention

[0003] The purpose of this invention is to provide a classification method and system that integrates different two-dimensional spectral data to solve the problem that different two-dimensional methods may capture different features of spectral data, while a single method may not be able to fully reflect the intrinsic information of the data.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a classification method for fusing different two-dimensional spectral data, comprising: preprocessing one-dimensional spectral data; performing two-dimensional processing on the preprocessed one-dimensional spectral data using multiple two-dimensional methods to generate multiple two-dimensional spectral images; extracting feature maps from each two-dimensional spectral image using a deep learning model to generate multiple feature maps; fusing the multiple feature maps; evaluating the performance of the model and optimizing the model based on the evaluation results; and using the trained model to classify and predict the spectral data to be tested, and outputting the prediction results.

[0005] In one embodiment, the preprocessing of the one-dimensional spectral data includes: denoising the collected one-dimensional spectral data; and normalizing it using the Z-score normalization method.

[0006] In one embodiment, the preprocessed one-dimensional spectral data is processed using multiple two-dimensionalization methods to generate multiple two-dimensional spectral images. These methods include: using PCA to reduce the dimensionality of the one-dimensional spectral data and convert it into a two-dimensional image while retaining the main components of the data; using t-SNE to embed the one-dimensional spectral data into a two-dimensional space while retaining the local and global structural information of the data; and applying LDA to project the one-dimensional spectral data onto the optimal discrimination vector space to achieve the two-dimensionalization of the data.

[0007] In one embodiment, the feature fusion of multiple feature maps includes: concatenating feature maps from PCA, t-SNE, and LDA methods along the channel dimension to obtain a fused feature map.

[0008] In one embodiment, the step of using a deep learning model to extract feature maps from each two-dimensional spectral image and generate multiple feature maps includes: extracting feature maps from each two-dimensional spectral image using a pre-trained ResNet convolutional neural network; and extracting feature representations through multi-layer convolution and pooling operations.

[0009] In one embodiment, the evaluation of the model's performance and the optimization of the model based on the evaluation results include: evaluating the performance of the support vector machine classifier using accuracy, recall, and F1 score metrics; and adjusting the parameters of the two-dimensionalization method, the structure of the feature extraction network, or the feature fusion strategy based on the evaluation results.

[0010] In one embodiment, before evaluating the performance of the model and optimizing the model based on the evaluation results, the method further includes: inputting the fused features into a classifier for training to learn the mapping relationship between spectral data and category labels or continuous values.

[0011] In one embodiment, the step of inputting the fused features into a classifier for training to learn the mapping relationship between spectral data and category labels or continuous values ​​includes: inputting the fused features into a support vector machine classifier for training; and optimizing the parameters of the support vector machine classifier using grid search and cross-validation methods.

[0012] This invention also provides a classification system that integrates different two-dimensional spectral data, comprising: a preprocessing module for preprocessing one-dimensional spectral data; a two-dimensional processing module for performing two-dimensional processing on the preprocessed one-dimensional spectral data using multiple two-dimensional processing methods to generate multiple two-dimensional spectral images; a feature extraction module for extracting feature maps from each two-dimensional spectral image using a deep learning model to generate multiple feature maps; a feature fusion module for fusing the multiple feature maps; a model evaluation and optimization module for evaluating the performance of the model and optimizing the model based on the evaluation results; and a classification prediction module for using the trained model to classify and predict the spectral data to be tested and output the prediction results.

[0013] In one embodiment, the classification system that fuses different two-dimensional spectral data further includes a training module, which is used to input the fused features into the classifier for training and to learn the mapping relationship between spectral data and category labels or continuous values.

[0014] As can be seen from the above technical solution, the present invention has at least the following advantages and positive effects compared with the prior art: 1. Improved classification accuracy: By integrating the spectral data features obtained by different two-dimensional methods, the complementary information of multiple methods is fully utilized, thereby improving the classification accuracy.

[0015] 2. Enhance model robustness: Integrating multiple 2D methods can reduce the limitations of a single method, making the model more adaptable and robust to different types of spectral data.

[0016] 3. Expand the scope of application: It is not only applicable to classification tasks of spectral data, but can also be extended to regression tasks and data processing in other related fields.

[0017] 4. Automation and efficiency: By using deep learning models for feature extraction and classification, the processing of spectral data is automated, improving processing efficiency and accuracy. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0019] Figure 1 is a flowchart illustrating a classification method that fuses different two-dimensional spectral data in one embodiment of the present invention; Figure 2 is a flowchart illustrating the preprocessing of one-dimensional spectral data in step S1 of the present invention; Figure 3 is a flowchart illustrating the two-dimensional processing of the preprocessed one-dimensional spectral data using multiple two-dimensional methods in step S2 of the present invention to generate multiple two-dimensional spectral images; Figure 4 is a flowchart illustrating the extraction of feature maps from each two-dimensional spectral image using a deep learning model in step S3 of the present invention to generate multiple feature maps; Figure 5 is a flowchart illustrating the input of the fused features into a classifier for training in step S5 of the present invention to learn the mapping relationship between spectral data and category labels or continuous values; Figure 6 is a flowchart illustrating the evaluation of model performance in step S6 of the present invention and the optimization of the model based on the evaluation results; Figure 7 is a structural diagram of the classification system that fuses different two-dimensional spectral data of the present invention; Figure 8 is a structural diagram of the feature fusion module shown in Figure 7.

[0020] The reference numerals in the attached figures are explained as follows: 210, preprocessing module; 220, 2D processing module; 230, feature extraction module; 240, feature fusion module; 250, training module; 260, model evaluation and optimization module; 270, classification prediction module. Detailed Implementation

[0021] To more clearly explain the purpose, technical solutions, and advantages of 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 of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein. On the contrary, these embodiments are provided so that the present invention will be more comprehensive and complete, and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0022] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a full understanding of embodiments of the invention. However, those skilled in the art will recognize that the technical solutions of the invention can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of the invention.

[0023] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0024] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0025] The present invention will be described in detail below with reference to specific embodiments: Embodiment 1

[0026] This invention provides a classification method that integrates different two-dimensional spectral data. This method is mainly applied to scenarios such as agricultural crop disease identification, medical tissue pathology analysis, or environmental pollutant monitoring. Referring to Figure 1, Figure 1 is a schematic flowchart of the classification method for integrating different two-dimensional spectral data according to this invention.

[0027] As shown in Figure 1, a classification method that integrates different two-dimensional spectral data includes the following steps.

[0028] S1. Preprocess the one-dimensional spectral data.

[0029] Spectral data is highly susceptible to instrument noise, changes in ambient light, and scattering effects from the sample surface during acquisition. Therefore, preprocessing one-dimensional spectral data is crucial to ensuring the accuracy of subsequent models.

[0030] Referring to Figure 2, which is a schematic diagram of the process of preprocessing one-dimensional spectral data in step S1 of the present invention.

[0031] Preprocessing one-dimensional spectral data includes the following steps.

[0032] S101. Denoising the collected one-dimensional spectral data. Wavelet transform denoising is used, and wavelet basis functions are selected to decompose the original one-dimensional spectral signal into five levels. A soft thresholding function is used to process the high-frequency coefficients, effectively filtering out high-frequency random noise while preserving the abrupt changes in the spectral signal, resulting in a smooth spectral curve after reconstruction.

[0033] S102. Normalization is performed using the Z-score standardization method. For the denoised spectral data matrix X, the mean μ and standard deviation σ of each column are calculated using the formula... The transformation is then performed. This step eliminates the effects of differences in wavelength dimensions and baseline drift, ensuring the data follows a standard normal distribution and accelerating the convergence speed of the subsequent neural network.

[0034] By denoising the collected one-dimensional spectral data and normalizing it using the Z-score normalization method, it can be ensured that the one-dimensional spectral data are of the same order of magnitude.

[0035] S2. Multiple two-dimensional transformation methods are used to process the preprocessed one-dimensional spectral data into two-dimensional images to generate multiple two-dimensional spectral images.

[0036] Single two-dimensional methods can often only reflect one aspect of the data. This invention employs three complementary methods in parallel: PCA, t-SNE, and LDA.

[0037] Referring to Figure 3, Figure 3 is a schematic diagram of the process in step S2 of the present invention, which uses various two-dimensionalization methods to process the preprocessed one-dimensional spectral data into two-dimensional images to generate multiple two-dimensional spectral images.

[0038] Multiple two-dimensional transformation methods are used to process the preprocessed one-dimensional spectral data into two-dimensional images, specifically including the following steps.

[0039] S201. The PCA method is used to reduce the dimensionality of one-dimensional spectral data and convert it into a two-dimensional image, preserving the main components of the data. The eigenvalues ​​and eigenvectors of the covariance matrix are calculated, and the top two principal components (PC1, PC2) with a cumulative contribution rate exceeding 95% are selected. The scores of each sample on these two principal components are used as the horizontal and vertical coordinates, mapped onto a two-dimensional plane to generate a grayscale image. This image mainly preserves the global variance information and the distribution of the main components of the spectral data.

[0040] S202. The t-SNE method is used to embed one-dimensional spectral data into a two-dimensional space, preserving the local and global structural information of the data. The perplexity is set to 30, the learning rate to 200, and the number of iterations to 1000. This method minimizes the KL divergence between the probability distribution in the high-dimensional space and the t-distribution in the low-dimensional space. The generated two-dimensional image can excellently preserve the local neighborhood structure and clustering features of the spectral data, allowing similar samples to cluster tightly in the image space.

[0041] S203. The LDA method is applied to project the one-dimensional spectral data onto the optimal discrimination vector space, realizing the two-dimensionalization of the data. Using known class labels, the inter-class scatter matrix and intra-class scatter matrix are calculated, and the generalized eigenvalue problem is solved to find the optimal projection vector that maximizes the inter-class distance and minimizes the intra-class distance. The data is then projected onto the two-dimensional space formed by the first two discrimination vectors. This image highlights the discriminative information between different disease categories.

[0042] Ultimately, each original spectral sample is converted into three two-dimensional spectral images corresponding to PCA, t-SNE, and LDA, respectively.

[0043] S3. Use a deep learning model to extract feature maps from each two-dimensional spectral image and generate multiple feature maps.

[0044] Feature maps are extracted from each two-dimensional spectral image using a deep learning model. It should be noted that the feature maps include, but are not limited to, important information such as the spatial structure and texture of the spectral data.

[0045] Referring to Figure 4, Figure 4 is a schematic diagram of the process of extracting feature maps from each two-dimensional spectral image and generating multiple feature maps in step S3 of the present invention using a deep learning model.

[0046] The deep learning model is used to extract feature maps from each two-dimensional spectral image to generate multiple feature maps, specifically including the following steps.

[0047] S301. Use a pre-trained ResNet convolutional neural network to extract feature maps from each two-dimensional spectral image.

[0048] S302. Feature representations are extracted through multi-layer convolution and pooling operations.

[0049] S4. Perform feature fusion on multiple feature maps.

[0050] Feature maps from different 2D transformation methods can be effectively fused. Stitching, weighted summation, or other advanced fusion strategies can be employed to fully utilize the complementary information between different feature maps. Specifically, in this embodiment, feature maps from PCA, t-SNE, and LDA methods are stitched together along the channel dimension to obtain the fused feature map. The stitching operation preserves all information from the different feature maps, achieving comprehensive information fusion.

[0051] It should be noted that, in order to reduce dimensionality and further integrate information, a convolutional layer was used to compress the number of channels from 6144 to 512, followed by global average pooling to obtain the final fused feature vector. This vector contains both global variance information and integrates local structure and class discrimination information.

[0052] S5. Input the fused features into the classifier for training to learn the mapping relationship between spectral data and category labels or continuous values.

[0053] It should be noted that classifiers include, but are not limited to, support vector machine classifiers, random forest classifiers, and other classifiers.

[0054] Referring to Figure 5, Figure 5 is a flowchart of step S5 in this invention, in which the fused features are input into a classifier for training to learn the mapping relationship between spectral data and category labels or continuous values.

[0055] The fused features are input into a classifier for training to learn the mapping relationship between spectral data and category labels or continuous values, specifically including the following steps.

[0056] S501. Input the fused features into a support vector machine classifier for training. Use the fused feature vector obtained in step S4 as the input to the support vector machine classifier, and the corresponding class label as the output.

[0057] S502. Optimize the parameters of the support vector machine classifier using grid search and cross-validation methods.

[0058] It should be noted that the parameters of a support vector machine classifier include the penalty coefficient and the kernel function parameters.

[0059] S6. Evaluate the performance of the model and optimize the model based on the evaluation results.

[0060] Referring to Figure 6, which is a flowchart of evaluating the performance of the model in step S6 of the present invention and optimizing the model based on the evaluation results.

[0061] The model's performance is evaluated using metrics such as cross-validation, accuracy, recall, and F1 score, and the model is optimized based on the evaluation results.

[0062] Evaluate the model's performance and optimize it based on the evaluation results, specifically including the following steps.

[0063] S601. Evaluate the performance of the support vector machine classifier using accuracy, recall, and F1 score metrics. Evaluate model performance on an independent test set. Calculate not only the overall accuracy but also the precision, recall, and F1 score for each category, and plot a confusion matrix to analyze easily confused categories.

[0064] S602. Adjust the parameters of the 2D transformation method, the structure of the feature extraction network, or the feature fusion strategy based on the evaluation results. If the recall rate is lower than the threshold, backtrack and adjust the parameters of the 2D transformation process in step S2, or fine-tune the deep learning model in step S3, and retrain until all indicators meet the requirements of practical applications.

[0065] S7. Use the trained model to classify and predict the spectral data to be tested, and output the prediction results.

[0066] A trained support vector machine classifier is used to classify and predict new spectral data, and the prediction results are output. The prediction results can be used in multiple fields such as agricultural disease diagnosis, medical disease screening, and environmental monitoring.

[0067] In summary, this application improves classification accuracy by fusing spectral data features obtained from different two-dimensionalization methods, fully utilizing the complementary information of multiple methods. Fusing multiple two-dimensionalization methods reduces the limitations of a single method, making the model more adaptable and robust to different types of spectral data. It is not only suitable for spectral data classification tasks but can also be extended to regression tasks and data processing in other related fields. Utilizing deep learning models for feature extraction and classification automates spectral data processing, improving processing efficiency and accuracy. Example 2

[0068] Referring to Figure 7, the present invention provides a classification system that integrates different two-dimensional spectral data. Figure 7 is a schematic diagram of the structure of the classification system that integrates different two-dimensional spectral data according to the present invention.

[0069] As shown in Figure 7, a classification system that integrates different two-dimensional spectral data includes: a preprocessing module 210, a two-dimensional processing module 220, a feature extraction module 230, a feature fusion module 240, a training module 250, a model evaluation and optimization module 260, and a classification prediction module 270.

[0070] The preprocessing module 210 is used to preprocess one-dimensional spectral data.

[0071] The two-dimensional processing module 220 is used to perform two-dimensional processing on the preprocessed one-dimensional spectral data using various two-dimensional methods to generate multiple two-dimensional spectral images.

[0072] The feature extraction module 230 is used to extract feature maps from each two-dimensional spectral image using a deep learning model, generating multiple feature maps.

[0073] The feature fusion module 240 is used to fuse multiple feature maps. For example, Figure 8 illustrates the structural diagram of the feature fusion module 240 in this invention. The main function of the feature fusion module 240 is to effectively integrate multiple independent feature maps generated by different two-dimensional methods to form a composite feature map containing information from all sources, so as to make full use of the complementary information extracted by each method.

[0074] Training module 250 is used to input the fused features into the classifier for training, learning the mapping relationship between spectral data and class labels or continuous values.

[0075] The model evaluation and optimization module 260 is used to evaluate the performance of the model and optimize the model based on the evaluation results; the classification and prediction module 270 is used to classify and predict the spectral data to be tested using the trained model and output the prediction results.

[0076] This system achieves fully automated processing from raw spectra to classification results through hardware and software collaboration, significantly improving the intelligence level of spectral analysis.

[0077] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the claims. It should be understood that the invention is not limited to the precise structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A classification method that fuses different two-dimensional spectral data, characterized in that, include: Preprocessing of one-dimensional spectral data; Multiple two-dimensional transformation methods are used to process the preprocessed one-dimensional spectral data into two-dimensional images, generating multiple two-dimensional spectral images. A deep learning model is used to extract feature maps from each two-dimensional spectral image, generating multiple feature maps; Multiple feature maps are fused; the model performance is evaluated and optimized based on the evaluation results; the trained model is used to classify and predict the spectral data to be tested, and the prediction results are output.

2. The classification method for fusing different two-dimensional spectral data according to claim 1, characterized in that, The preprocessing of the one-dimensional spectral data includes: denoising the collected one-dimensional spectral data; and normalizing it using the Z-score normalization method.

3. The classification method for fusing different two-dimensional spectral data according to claim 1, characterized in that, The method employs multiple 2D transformation techniques to process the preprocessed one-dimensional spectral data into two-dimensional images, including: using PCA to reduce the dimensionality of the one-dimensional spectral data and convert it into a two-dimensional image while retaining the main components of the data; using t-SNE to embed the one-dimensional spectral data into a two-dimensional space while retaining the local and global structural information of the data; and applying LDA to project the one-dimensional spectral data onto the optimal discrimination vector space to achieve the 2D transformation of the data.

4. The classification method for fusing different two-dimensional spectral data according to claim 2, characterized in that, The feature fusion of multiple feature maps includes: concatenating feature maps from PCA, t-SNE, and LDA methods along the channel dimension to obtain a fused feature map.

5. The classification method for fusing different two-dimensional spectral data according to claim 1, characterized in that, The method of extracting feature maps from each two-dimensional spectral image using a deep learning model to generate multiple feature maps includes: extracting feature maps from each two-dimensional spectral image using a pre-trained ResNet convolutional neural network; and extracting feature representations through multi-layer convolution and pooling operations.

6. The classification method for fusing different two-dimensional spectral data according to claim 1, characterized in that, The evaluation model's performance is assessed, and the model is optimized based on the evaluation results, including: evaluating the performance of the support vector machine classifier using accuracy, recall, and F1 score metrics; and adjusting the parameters of the two-dimensionalization method, the structure of the feature extraction network, or the feature fusion strategy based on the evaluation results.

7. The classification method for fusing different two-dimensional spectral data according to claim 1, characterized in that, Before evaluating the performance of the model and optimizing the model based on the evaluation results, the method further includes: inputting the fused features into a classifier for training to learn the mapping relationship between spectral data and category labels or continuous values.

8. The classification method for fusing different two-dimensional spectral data according to claim 7, characterized in that, The step of inputting the fused features into a classifier for training to learn the mapping relationship between spectral data and category labels or continuous values ​​includes: inputting the fused features into a support vector machine classifier for training; and optimizing the parameters of the support vector machine classifier using grid search and cross-validation methods.

9. A classification system that integrates different two-dimensional spectral data, characterized in that, include: The preprocessing module is used to preprocess one-dimensional spectral data; The two-dimensional processing module is used to perform two-dimensional processing on the preprocessed one-dimensional spectral data using various two-dimensional methods to generate multiple two-dimensional spectral images; The feature extraction module is used to extract feature maps from each two-dimensional spectral image using a deep learning model, generating multiple feature maps; The feature fusion module is used to fuse multiple feature maps. The model evaluation and optimization module is used to evaluate the performance of the model and optimize the model based on the evaluation results. The classification and prediction module is used to classify and predict the spectral data under test using a trained model and output the prediction results.

10. The classification system for fusing different two-dimensional spectral data according to claim 9, characterized in that, The classification system that integrates different two-dimensional spectral data also includes a training module, which is used to input the fused features into the classifier for training and learn the mapping relationship between spectral data and category labels or continuous values.

Images