Hyperspectral image feature extraction method for unsupervised discriminant analysis
By introducing unsupervised discriminant analysis and hybrid projection technology in hyperspectral image feature extraction, the problem that existing methods cannot fully utilize the correlation of superpixel blocks is solved, and more efficient feature extraction and classification accuracy is achieved.
Patent Information
- Application Number
- CN202411926986.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-25
- Publication Date
- 2025-04-15
AI Technical Summary
The existing unsupervised feature extraction methods cannot fully consider the correlation between superpixel blocks, resulting in insufficient utilization of spatial information in hyperspectral images and the inability to further improve feature extraction performance.
A hyperspectral image feature extraction method for unsupervised discriminant analysis is proposed. Through superpixel block segmentation and local reconstruction, a feature extraction model based on unsupervised mixed projection and discriminant analysis is constructed, local and global features of hyperspectral images are extracted, and full-picture classification is performed through a classifier.
By fully considering the correlation between superpixel blocks, the feature extraction performance of hyperspectral images is improved, and better classification accuracy and noise removal effect are obtained.
Smart Images
Figure CN119832272A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of hyperspectral image classification, and in particular to a hyperspectral image feature extraction method for unsupervised discriminant analysis. Background Art
[0002] Traditional unsupervised feature extraction models do not fully consider the spatial information of hyperspectral images, and the separability of the extracted low-dimensional features of hyperspectral images is not ideal. Compared with traditional unsupervised feature extraction models, unsupervised feature extraction models based on superpixels can make good use of the hyperspectral spatial information after superpixel block segmentation, thereby obtaining better classification accuracy. Its pioneering work is that Jiang et al. applied entropy rate superpixel block (ERS) segmentation to hyperspectral images in the literature 1 (J. Jiang, J. Ma, C. Chen, Z. Wang, Z.Cai, and L. Wang, “SuperPCA: A superpixelwise PCA approach for unsupervised feature extraction of hyperspectral imagery,” IEEE Trans. Geosci. RemoteSens., vol. 56, no. 8, pp. 4581–4593, Aug. 2018.), and then performed local feature extraction based on principal component analysis (PCA) within each uniform superpixel block. Subsequently, Zhang et al. proposed superpixel-based local reconstruction for hyperspectral image denoising, as well as PCA-based global-local and spectral-spatial feature extraction in reference 2 (X. Zhang, X. Jiang, J. Jiang, Y. Zhang, X. Liu, and Z. Cai, “Spectral-spatial and superpixelwise PCA for unsupervised feature extraction of hyperspectral imagery,” IEEE Trans. Geosci. Remote Sens., vol.60, 2022, Art. no. 5502210).
[0003] The above methods can utilize the spatial information of hyperspectral images to a certain extent, thereby obtaining low-dimensional features of hyperspectral images with good separability. However, since the relevant information between superpixel blocks is not considered, the spatial information of hyperspectral images is not fully utilized, making it impossible for the unsupervised feature extraction model to further obtain better performance.
[0004] On this basis, Lu et al. proposed an unsupervised discriminant analysis model based on pseudo labels provided by superpixel block segmentation in the literature 3 (P. Lu, X. Jiang, Y. Zhang, X. Liu, Z. Cai, J. Jiang, and A. Plaza, “Spectral–spatial and superpixelwise unsupervised linear discriminant analysis for feature extraction and classification of hyperspectral images,” IEEE Transactions on Geoscience and Remote Sensing, vol. 61, pp. 1–15, 2023.). This model considers the relationship between superpixel blocks by obtaining pseudo labels through superpixel block segmentation, which further improves the performance of the unsupervised feature extraction model based on superpixels.
[0005] However, this method only considers the differences between superpixel blocks, but does not consider the correlation between superpixel blocks, so the model performance still cannot be improved. Therefore, the current unsupervised feature extraction model based on superpixels still considers the relationship between superpixel blocks in a relatively simple way, which inhibits the feature extraction performance of this type of model to a certain extent. Summary of the invention
[0006] The purpose of the present invention is to provide a hyperspectral image feature extraction method for unsupervised discriminant analysis in order to solve the problem that the existing unsupervised feature extraction method cannot well consider the correlation between superpixel blocks.
[0007] The above-mentioned purpose of the present application is achieved through the following technical solutions: S1: Acquire hyperspectral images and perform superpixel block segmentation; S2: Construct an unsupervised feature extraction model based on unsupervised hybrid projection and discriminant analysis; extract the local and global features of the hyperspectral image through the unsupervised feature extraction model, classify the hyperspectral image, and obtain the classification result of the whole image.
[0008] Optionally, step S1 includes: The ERS segmentation algorithm is used to segment the hyperspectral image into a preset number of super-pixel blocks based on the resolution and inherent characteristics of the hyperspectral image; the inherent characteristics include: color, brightness and texture.
[0009] Optionally, step S2 includes: S21: Based on the hyperspectral image segmented by superpixel blocks, in each superpixel block, locally reconstruct the th pixel through other pixels within the k nearest neighbors of the th pixel to obtain a hyperspectral image after noise removal; S22: sending the hyperspectral image after noise removal to a local feature extraction module based on unsupervised mixed projection in the unsupervised feature extraction model to obtain local features of the hyperspectral image; S23: sending the hyperspectral image after noise removal to a global feature extraction module based on unsupervised discriminant analysis in the unsupervised feature extraction model to obtain global features of the hyperspectral image; S24: Connect the global features and local features after feature extraction to form the final fusion features; input the fusion features into the classifier for full-image classification to obtain the classification result of the full image.
[0010] Optionally, step S21 includes: S21a: Hyperspectral image after superpixel segmentation The The first Pixel , select pixels based on the K nearest neighbor principle The set of neighboring pixels , for pixels Reconstruction is performed and the calculation is as follows:
[0011] in, By pixel and its neighboring pixels Similarity weights calculated using the Euclidean distance metric and the Gaussian kernel function; S21b: Reconstruct each pixel in the hyperspectral image after superpixel block segmentation to obtain a hyperspectral image after noise removal .
[0012] Optionally, step S22 includes: S22a: Let all superpixel blocks after noise removal be the set , a total of S superpixel blocks; Average all pixels in each superpixel block to obtain a superpixel feature set ; Target superpixel patch The superpixel feature is , select the superpixel block The superpixel features of other superpixel blocks that are spatially contacted are reconstructed as follows:
[0013] in, Represents the superpixel features of the jth superpixel block; represents the reconstruction coefficient; The normalization constraint of the reconstruction coefficients makes the reconstruction coefficients It is not always greater than 0. The reconstruction coefficient is used to perform a domain screening based on the reconstructed superpixel block level on each superpixel block to obtain the screening result, which specifically includes: the superpixel feature corresponding to the reconstruction coefficient greater than 0 is regarded as the superpixel feature corresponding to the superpixel block. Similar superpixel blocks, the remaining superpixel features are considered to be similar to the superpixel blocks. Dissimilar superpixel patches; S22b: Combine the screening results to perform self-projection matrix The design is as follows: Based on the superpixel information, the first scatter matrix is defined as follows:
[0014] in and The similarity weights of two superpixel features calculated by Euclidean distance metric and Gaussian kernel function; is the number of similar superpixel features in the screening results; Is the filter result The number of dissimilar superpixel features; Based on the pixel information, the second scatter matrix is defined as follows:
[0015] in is the similarity weight between pixel and superpixel features calculated by Euclidean distance metric and Gaussian kernel function, It is The number of all pixels in the block superpixel; By using the first scatter matrix and the second scatter matrix, we can solve the generalized eigenvalue optimization problem. ,as follows:
[0016] S22c: Combine the screening results to perform the neighbor projection matrix Designed only for Similar superpixel blocks are optimized using the classic PCA algorithm as follows:
[0017] in is with The pixel set of similar superpixel blocks, perform the same operation on all similar blocks, and get The set of neighbor projection matrices ; represents the trace of the matrix; express The transpose of express The covariance matrix of represents the unit matrix; S22d: Projection matrix and all neighbor projection matrices Acting on the target superpixel block The results are combined as follows:
[0018] in is a custom balance parameter. is the target superpixel block Low-dimensional local features; S22e: Perform the operations of steps S22a to S22d on each superpixel block to obtain the low-dimensional local features of each superpixel block; obtain the local features of the hyperspectral image by aggregating the low-dimensional local features of all superpixel blocks. .
[0019] Optionally, step S23 includes: S23a: For each superpixel block, select other superpixel blocks that are k nearest neighbors from all superpixel blocks to obtain a set of similar original superpixel features in the global scope. And the superpixel feature set after denoising , the remaining superpixel blocks are regarded as dissimilar original superpixel feature sets And the superpixel feature set after denoising ; S23b: Based on the original hyperspectral image and the denoised hyperspectral image, design the superpixel block-level and pixel-level scatter matrix; Constructing intra-class scatter matrix at superpixel block level and the between-class scatter matrix as follows:
[0020]
[0021] in, , , and It is the similarity weight of two superpixel features calculated by Euclidean distance metric and Gaussian kernel function; The pixel-level intra-class scatter matrix is constructed as follows:
[0022] in, represents the i-th pixel in the s-th superpixel before denoising; Represents the features of the sth superpixel before denoising; represents the i-th pixel in the s-th superpixel after denoising; Represents the features of the sth superpixel after denoising; Represents the number of pixels in the s-th superpixel; S23c: Based on the scattered matrix, the global projection matrix is optimized by solving the generalized eigenvalues ,as follows:
[0023] in represents the trace of a matrix; express The transpose of Indicates custom balance parameters; calculate , Represents the denoised hyperspectral image and obtains the global features of the hyperspectral image.
[0024] An electronic device comprises a processor, a memory, a user interface and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device performs a hyperspectral image feature extraction method of unsupervised discriminant analysis.
[0025] A computer-readable storage medium stores instructions. When the instructions are executed, a hyperspectral image feature extraction method of unsupervised discriminant analysis is performed.
[0026] The beneficial effects of the technical solution provided by this application are: 1. A local feature extraction method based on unsupervised mixed projection of superpixel blocks is designed, which can fully consider the information between mixed superpixel blocks. It supplements the lack of existing unsupervised feature extraction methods based on superpixel blocks in the hyperspectral field.
[0027] 2. A global feature extraction method based on unsupervised discriminant analysis of superpixel blocks is designed to supplement the method of analyzing the relationship between superpixel blocks.
[0028] 3. The local reconstruction combined with superpixel blocks can effectively remove the noise in the hyperspectral image and improve the feasibility of subsequent feature extraction work. The subtle spatial relationship between superpixel blocks is captured by calculating different projection matrices for each superpixel block and its spatially adjacent superpixel blocks. By selecting other superpixel blocks for each superpixel block in the global range of the hyperspectral image through the screening rules of mutual neighbors, the negative impact caused by the incorrect analysis of the relationship between superpixel blocks can be effectively reduced. The existing unsupervised feature extraction model based on superpixels in the field of hyperspectral images mainly analyzes each superpixel block independently or only considers the differences between superpixel blocks. Therefore, the unsupervised feature extraction model based on superpixels in the present application supplements the unsupervised feature extraction in the hyperspectral field, which can fully consider the complex correlations between superpixel blocks and has broad market prospects. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] The present application will be further described below with reference to the accompanying drawings and embodiments, in which: Figure 1 is a step diagram in an embodiment of the present application; Figure 2 It is a schematic diagram of the results in the examples of the present application; Figure 3 It is a schematic diagram of the structure of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION
[0030] In order to have a clearer understanding of the technical features, purposes and effects of the present application, the specific implementation methods of the present application are now described in detail with reference to the accompanying drawings.
[0031] The embodiment of the present application provides a hyperspectral image feature extraction method for unsupervised discriminant analysis.
[0032] Please refer to Figure 1 , Figure 1 : is a step diagram of a hyperspectral image feature extraction method for unsupervised discriminant analysis in an embodiment of the present application, comprising: S1: Acquire hyperspectral images and perform superpixel block segmentation; S2: Construct an unsupervised feature extraction model based on unsupervised hybrid projection and discriminant analysis; extract the local and global features of the hyperspectral image through the unsupervised feature extraction model, classify the hyperspectral image, and obtain the classification result of the whole image.
[0033] Step S1 includes: The ERS segmentation algorithm is used to segment the hyperspectral image into a preset number of super-pixel blocks based on the resolution and inherent characteristics of the hyperspectral image; the inherent characteristics include: color, brightness and texture.
[0034] As an embodiment, the hyperspectral image is constructed into a weighted graph , where the vertex is the pixel in the hyperspectral image to be segmented, and the edge set It consists of the similarity between pixels.
[0035] ERS segmentation algorithm attempts to select edge A subset of Depend on The objective function of ERS is defined as:
[0036] in, It is an entropy rate function, the purpose is to obtain uniform and compact clustering; is a balancing term, the purpose is to obtain clusters of similar size; is a balance parameter. ERS uses a greedy heuristic algorithm to solve and obtain the optimal superpixel block segmentation.
[0037] Step S2 includes: S21: Based on the hyperspectral image segmented by superpixel blocks, in each superpixel block, The other pixels within the k nearest neighbors of the pixel are Pixels are locally reconstructed to obtain a hyperspectral image after noise removal; S22: sending the hyperspectral image after noise removal to a local feature extraction module based on unsupervised mixed projection in the unsupervised feature extraction model to obtain local features of the hyperspectral image; S23: sending the hyperspectral image after noise removal to a global feature extraction module based on unsupervised discriminant analysis in the unsupervised feature extraction model to obtain global features of the hyperspectral image; S24: Connect the global features and local features after feature extraction to form the final fusion features; input the fusion features into the classifier for full-image classification to obtain the classification result of the full image.
[0038] As an embodiment, the classifier used is a SVM classifier with a radial basis function kernel (RBF) and grid search parameters, and the final classification result of the entire image is obtained after passing through the classifier.
[0039] In order to illustrate the effect of the present invention, the following examples are provided for experimental comparison.
[0040] The experiments were verified using three hyperspectral image datasets: PaviaU, Houston 2013, and XuZhou. For fairness, for all the comparison methods, when extracting dimensionality reduction features, the number of training samples is 10 sample pixels per class, and the remaining sample pixels are used as test data.
[0041] Classification accuracy and Kappa coefficient are commonly used classification performance evaluation indicators in the hyperspectral field. The closer the classification accuracy is to 1, the higher the proportion of samples correctly classified by the classifier, and the better the classification effect.
[0042] The calculation formula for classification accuracy is: ,in, is the number of test sample data, and Respectively The predicted class label information and the true class label information of the test sample data are is an indicator function, when equal 1 if the value is 0, otherwise it is 0.
[0043] The calculation formula of Kappa coefficient is:
[0044] in , .
[0045] This method is experimentally compared and verified with references 1, 2, 3, and 4 (S. Zhang, T. Lu, W. Fu, and S. Li, "Superpixel-level hybrid discriminant lysis for hyperspectral image feature extraction," IEEE Transactions n Geoscience and Remote Sensing, vol. 60, pp.1–13, 2022.). The comparison of classification accuracy obtained from the experiment is shown in Table 1. For three hyperspectral image datasets with different distribution types, the technical solution of this application is superior to the methods in the literature.
[0046] Table 1
[0047] Step S21 includes: S21a: Hyperspectral image after superpixel segmentation The The first Pixel , select pixels based on the K nearest neighbor principle The set of neighboring pixels , for pixels Reconstruction is performed and the calculation is as follows:
[0048] in, By pixel and its neighboring pixels Similarity weights calculated using the Euclidean distance metric and the Gaussian kernel function; S21b: Reconstruct each pixel in the hyperspectral image after superpixel block segmentation to obtain a hyperspectral image after noise removal .
[0049] Step S22 includes: S22a: Let all superpixel blocks after noise removal be the set , a total of S superpixel blocks; Average all pixels in each superpixel block to obtain a superpixel feature set ; Target superpixel patch The superpixel feature is , select the superpixel block The superpixel features of other superpixel blocks that are spatially contacted are reconstructed as follows:
[0050] in, Represents the superpixel features of the jth superpixel block; represents the reconstruction coefficient; The normalization constraint of the reconstruction coefficients makes the reconstruction coefficients It is not always greater than 0. The reconstruction coefficient is used to perform a domain screening based on the reconstructed superpixel block level on each superpixel block to obtain the screening result, which specifically includes: the superpixel feature corresponding to the reconstruction coefficient greater than 0 is regarded as the superpixel feature corresponding to the superpixel block. Similar superpixel blocks, the remaining superpixel features are considered to be similar to the superpixel blocks. Dissimilar superpixel patches; S22b: Combine the screening results to perform self-projection matrix The design is as follows: Based on the superpixel information, the first scatter matrix is defined as follows:
[0051] in and The similarity weights of two superpixel features calculated by Euclidean distance metric and Gaussian kernel function; is the number of similar superpixel features in the screening results; Is the filter result The number of dissimilar superpixel features; Based on the pixel information, the second scatter matrix is defined as follows:
[0052] in is the similarity weight between pixel and superpixel features calculated by Euclidean distance metric and Gaussian kernel function, It is The number of all pixels in the block superpixel; By using the first scatter matrix and the second scatter matrix, we can solve the generalized eigenvalue optimization problem. ,as follows:
[0053] S22c: Combine the screening results to perform the neighbor projection matrix Designed only for Similar superpixel blocks are optimized using the classic PCA algorithm as follows:
[0054] in is with The pixel set of similar superpixel blocks, perform the same operation on all similar blocks, and get The set of neighbor projection matrices ; represents the trace of the matrix; express The transpose of express The covariance matrix of represents the unit matrix; S22d: Projection matrix and all neighbor projection matrices Acting on the target superpixel block The results are combined as follows:
[0055] in is a custom balance parameter. is the target superpixel block Low-dimensional local features; S22e: Perform the operations of steps S22a to S22d on each superpixel block to obtain the low-dimensional local features of each superpixel block; obtain the local features of the hyperspectral image by aggregating the low-dimensional local features of all superpixel blocks. .
[0056] As an embodiment, the local feature extraction module based on unsupervised hybrid projection includes: performing domain screening at the reconstructed superpixel block level on each superpixel block, designing its own projection matrix, designing its neighbor projection matrix, and local feature extraction based on unsupervised hybrid projection.
[0057] Step S23 includes: S23a: For each superpixel block, select other superpixel blocks that are k nearest neighbors from all superpixel blocks to obtain a set of similar original superpixel features in the global scope. And the superpixel feature set after denoising , the remaining superpixel blocks are regarded as dissimilar original superpixel feature sets And the superpixel feature set after denoising ; S23b: Based on the original hyperspectral image and the denoised hyperspectral image, design the superpixel block-level and pixel-level scatter matrix; Constructing intra-class scatter matrix at superpixel block level and the between-class scatter matrix as follows:
[0058]
[0059] in, , , and It is the similarity weight of two superpixel features calculated by Euclidean distance metric and Gaussian kernel function; The pixel-level intra-class scatter matrix is constructed as follows:
[0060] in, represents the i-th pixel in the s-th superpixel before denoising; Represents the features of the sth superpixel before denoising; represents the i-th pixel in the s-th superpixel after denoising; Represents the features of the sth superpixel after denoising; Represents the number of pixels in the s-th superpixel; S23c: Based on the scattered matrix, the global projection matrix is optimized by solving the generalized eigenvalues ,as follows:
[0061] in represents the trace of a matrix; express The transpose of Indicates custom balance parameters; calculate , Represents the denoised hyperspectral image and obtains the global features of the hyperspectral image.
[0062] As an embodiment, the global feature extraction module based on unsupervised discriminant analysis includes: screening superpixel features of k-nearest neighbors of each superpixel block, designing a scatter matrix based on the original hyperspectral image and the hyperspectral image after noise removal, and global feature extraction based on unsupervised discriminant analysis.
[0063] The present application provides a hyperspectral image feature extraction system for unsupervised discriminant analysis, the system comprising: a hyperspectral image denoising module, a local feature extraction module, a global feature extraction module, a feature fusion module and a classifier; the hyperspectral image denoising module is connected to the local feature extraction module; the hyperspectral image denoising module is connected to the global feature extraction module; the local feature extraction module is connected to the feature fusion module; the global feature extraction module is connected to the feature fusion module; the feature fusion module is connected to the classifier;.
[0064] As an example, a feature extraction method based on unsupervised hybrid projection and discriminant analysis (such as Figure 2 The method is mainly divided into four parts. The first part is a denoising module based on local reconstruction for removing the noise of the hyperspectral image. The second part is a local feature extraction module based on unsupervised mixed projection, which can obtain the local features of the hyperspectral image. The third part is a global feature extraction module based on unsupervised discriminant analysis, which can obtain the global features of the hyperspectral image. The fourth part is a feature fusion module that connects the global features and local features after feature extraction to form the final fusion features.
[0065] The present application also discloses an electronic device. Figure 3 , Figure 3 The electronic device 500 may include: at least one processor 501 , at least one network interface 504 , a user interface 503 , a memory 505 , and at least one communication bus 502 .
[0066] The communication bus 502 is used to realize the connection and communication between these components.
[0067] The user interface 503 may include a display screen, and the optional user interface 503 may also include a standard wired interface or a wireless interface.
[0068] The network interface 504 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).
[0069] The present application also discloses a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded by a processor to execute the above-mentioned method for extracting features from hyperspectral images by unsupervised discriminant analysis.
[0070] The above are only exemplary embodiments of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure.
[0071] This application is intended to cover any variation, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the art not described in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.
Claims
1. A hyperspectral image feature extraction method for unsupervised discriminant analysis, characterized in that: The method comprises the following steps: S1: Acquire hyperspectral images and perform superpixel block segmentation; S2: Construct an unsupervised feature extraction model based on unsupervised hybrid projection and discriminant analysis; extract the local and global features of the hyperspectral image through the unsupervised feature extraction model, classify the hyperspectral image, and obtain the classification result of the whole image.
2. The method for extracting features from hyperspectral images by unsupervised discriminant analysis according to claim 1, characterized in that: Step S1 includes: The ERS segmentation algorithm is used to segment the hyperspectral image into a preset number of super-pixel blocks based on the resolution and inherent characteristics of the hyperspectral image; the inherent characteristics include: color, brightness and texture.
3. The method for extracting features from hyperspectral images by unsupervised discriminant analysis according to claim 1, characterized in that: Step S2 includes: S21: Based on the hyperspectral image segmented by superpixel blocks, in each superpixel block, The other pixels within the k nearest neighbors of the pixel are Pixels are locally reconstructed to obtain a hyperspectral image after noise removal; S22: sending the hyperspectral image after noise removal to a local feature extraction module based on unsupervised mixed projection in the unsupervised feature extraction model to obtain local features of the hyperspectral image; S23: sending the hyperspectral image after noise removal to a global feature extraction module based on unsupervised discriminant analysis in the unsupervised feature extraction model to obtain global features of the hyperspectral image; S24: Connect the global features and local features after feature extraction to form the final fusion features; input the fusion features into the classifier for full-image classification to obtain the classification result of the full image.
4. The method for extracting features from hyperspectral images by unsupervised discriminant analysis according to claim 1, characterized in that: Step S21 includes: S21a: Hyperspectral image after superpixel segmentation The The first Pixel , based on the K nearest neighbor principle, pixels are selected The set of neighboring pixels , for pixels Reconstruction is performed and the calculation is as follows: in, By pixel and its neighboring pixels Similarity weights calculated using the Euclidean distance metric and the Gaussian kernel function; S21b: Reconstruct each pixel in the hyperspectral image after superpixel block segmentation to obtain a hyperspectral image after noise removal .
5. The method for extracting features from hyperspectral images by unsupervised discriminant analysis according to claim 1, characterized in that: Step S22 includes: S22a: Let all superpixel blocks after noise removal be the set , a total of S superpixel blocks; Average all pixels in each superpixel block to obtain a superpixel feature set ; Target superpixel patch The superpixel feature is , select the superpixel block The superpixel features of other superpixel blocks that are spatially contacted are reconstructed as follows: in, Represents the superpixel features of the jth superpixel block; represents the reconstruction coefficient; The normalization constraint of the reconstruction coefficients makes the reconstruction coefficients It is not always greater than 0. The reconstruction coefficient is used to perform a domain screening based on the reconstructed superpixel block level on each superpixel block to obtain the screening result, which specifically includes: the superpixel feature corresponding to the reconstruction coefficient greater than 0 is regarded as the superpixel feature corresponding to the superpixel block. Similar superpixel blocks, the remaining superpixel features are considered to be similar to the superpixel blocks. Dissimilar superpixel patches; S22b: Combine the screening results to perform self-projection matrix The design is as follows: Based on the superpixel information, the first scatter matrix is defined as follows: in and The similarity weights of two superpixel features calculated by Euclidean distance metric and Gaussian kernel function; is the number of similar superpixel features in the screening results; Is the filter result The number of dissimilar superpixel features; Based on the pixel information, the second scatter matrix is defined as follows: in is the similarity weight between pixel and superpixel features calculated by Euclidean distance metric and Gaussian kernel function, It is The number of all pixels in the block superpixel; By using the first scatter matrix and the second scatter matrix, we can solve the generalized eigenvalue optimization problem. ,as follows: S22c: Combine the screening results to perform the neighbor projection matrix Designed only for Similar superpixel blocks are optimized using the classic PCA algorithm as follows: in is with The pixel set of similar superpixel blocks, perform the same operation on all similar blocks, and get The set of neighbor projection matrices ; represents the trace of the matrix; express The transpose of express The covariance matrix of represents the unit matrix; S22d: Projection matrix and all neighbor projection matrices Acting on the target superpixel block The results are combined as follows: in is a custom balance parameter. is the target superpixel block Low-dimensional local features; S22e: Perform the operations of steps S22a to S22d on each superpixel block to obtain the low-dimensional local features of each superpixel block; obtain the local features of the hyperspectral image by aggregating the low-dimensional local features of all superpixel blocks. .
6. The method for extracting features from hyperspectral images by unsupervised discriminant analysis according to claim 1, characterized in that: Step S23 includes: S23a: For each superpixel block, select other superpixel blocks that are k nearest neighbors from all superpixel blocks to obtain a set of similar original superpixel features in the global scope. And the superpixel feature set after denoising , the remaining superpixel blocks are regarded as dissimilar original superpixel feature sets And the superpixel feature set after denoising ; S23b: Based on the original hyperspectral image and the denoised hyperspectral image, design the superpixel block-level and pixel-level scatter matrix; Constructing intra-class scatter matrix at superpixel block level and the between-class scatter matrix as follows: in, , , and It is the similarity weight of two superpixel features calculated by Euclidean distance metric and Gaussian kernel function; The pixel-level intra-class scatter matrix is constructed as follows: in, represents the i-th pixel in the s-th superpixel before denoising; Represents the features of the sth superpixel before denoising; represents the i-th pixel in the s-th superpixel after denoising; Represents the features of the sth superpixel after denoising; Represents the number of pixels in the s-th superpixel; S23c: Based on the scatter matrix, the global projection matrix is optimized by solving the generalized eigenvalues ,as follows: in represents the trace of a matrix; express The transpose of Indicates custom balance parameters; calculate , Represents the denoised hyperspectral image and obtains the global features of the hyperspectral image.
7. An electronic device, characterized in that: The electronic device comprises a processor (501), a memory (505), a user interface (503) and a network interface (504), wherein the memory (505) is used to store instructions, the user interface (503) and the network interface (504) are used to communicate with other devices, and the processor (501) is used to execute the instructions stored in the memory (505) so that the electronic device executes the method according to any one of claims 1 to 6.
8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed by a computer, the method according to any one of claims 1 to 6 is executed.