Hyperspectral blind unmixing method

Abstract

The invention discloses a hyperspectral blind unmixing method, which comprises the following steps of: constructing a hyperspectral blind unmixing model, inputting hyperspectral image data into the hyperspectral blind unmixing model, encoding spatial features and spectral features of a mixed spectrum of each pixel in the hyperspectral image data through a shallow encoder, and obtaining a hyperspectral blind unmixing model; synchronously transmitting the combined features to an end member extraction branch and an abundance estimation branch through a feature sharing mechanism; the end member extraction branch receives hidden end member features extracted from the joint features through a feature extraction module, and then pure end members are extracted from the joint features by using a Colmogolov-Arnold network; according to the abundance estimation branch, an abundance graph corresponding to a pure end member is extracted from the joint features by using a windowed multi-head self-attention mechanism, and an unmixing result of a mixed spectrum is obtained. The method does not need to depend on prior end member information, the dependence of a traditional method on a known end member library is reduced, and the de-mixing efficiency, stability and accuracy of the system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of hyperspectral image processing, in particular to a hyperspectral blind unmixing method, and in particular to a hyperspectral blind unmixing method driven by a Kolmogorov-Arnold network. Background Art

[0002] Hyperspectral imaging technology has been widely used in remote sensing, agriculture, environmental monitoring and other fields because it can accurately capture the composition information of different substances in the scene through spectral information of multiple bands. Each pixel contained in a hyperspectral image is often a mixture of spectra of multiple substances. How to decompose the mixed spectrum into the pure spectrum (endmembers) of each substance and its corresponding abundance map has become an important task of hyperspectral unmixing.

[0003] Existing hyperspectral unmixing methods usually rely on known pure endmember spectral information or need to build an endmember library. However, in practical applications, it is extremely challenging to obtain complete and accurate endmember information. Therefore, traditional blind hyperspectral unmixing methods have the following defects: 1) they rely on complex model structures, have large computational complexity, and have low unmixing efficiency; 2) the stability and accuracy of the unmixing results also need to be improved. Summary of the invention

[0004] In view of this, an object of the present invention is to provide a hyperspectral blind unmixing method to solve the technical problems mentioned in the prior art.

[0005] A hyperspectral blind unmixing method, characterized in that the method comprises the following steps:

[0006] Constructing a hyperspectral blind unmixing model, wherein the hyperspectral blind unmixing model has a shallow encoder, an endmember extraction branch, and an abundance estimation branch;

[0007] The hyperspectral image data is input into the hyperspectral blind unmixing model, and the spatial features and spectral features of the mixed spectrum of each pixel in the hyperspectral image data are encoded by a shallow encoder to generate a joint feature, and the joint feature is synchronously transmitted to the end member extraction branch and the abundance estimation branch through a feature sharing mechanism;

[0008] The endmember extraction branch receives the joint features output by the shallow encoder, extracts hidden endmember features from the joint features through a feature extraction module, and then extracts pure endmembers from the joint features using a Kolmogorov-Arnold network;

[0009] The abundance estimation branch receives the joint features output by the shallow encoder, and uses a windowed multi-head self-attention mechanism to extract an abundance map corresponding to the pure endmember from the joint features to obtain an unmixing result of the mixed spectrum.

[0010] Optionally, the method further comprises linearly combining the extracted pure end members and the abundance map to reconstruct a hyperspectral image, and evaluating the accuracy of an unmixing result of the hyperspectral blind unmixing model by comparing the hyperspectral image with the consistency of the hyperspectral image data input by the hyperspectral blind unmixing model;

[0011] If the hyperspectral image is consistent with the hyperspectral image data input by the hyperspectral blind unmixing model, it means that the unmixing result of the hyperspectral blind unmixing model has high accuracy;

[0012] If the hyperspectral image is inconsistent with the hyperspectral image data input by the hyperspectral blind unmixing model, it means that the accuracy of the unmixing result of the hyperspectral blind unmixing model is low.

[0013] Optionally, the method further comprises, using a joint loss function to respectively constrain the endmember extraction branch and the abundance estimation branch of the hyperspectral blind unmixing model, so as to iteratively optimize the hyperspectral blind unmixing model during training, wherein;

[0014] The joint loss function includes a spectral angle distance loss function, a mean square error loss function, a total variation loss function and a regularization loss function;

[0015] The spectral angle distance loss and the mean square error loss are used to ensure that the hyperspectral image reconstructed according to the extracted pure end members and abundance map can accurately approximate the hyperspectral image data input by the hyperspectral blind unmixing model;

[0016] The total variation loss and the regularization loss are used to enhance the perception of subtle texture information and structural information in the hyperspectral image data, so as to ensure that the abundance map can capture edge features and local features in the hyperspectral image data.

[0017] Optionally, each pixel of the hyperspectral image data corresponds to a mixed spectrum of different substances, and the mixed spectrum includes spatial features and spectral features;

[0018] The spectral feature has several bands, and each band has a different wavelength range.

[0019] Optionally, the convolutional neural network structure of the shallow encoder is 1×1 convolution.

[0020] Optionally, the method of extracting pure end members from the joint features by the Kolmogorov-Arnold network is:

[0021] The joint features are separated by a nonlinear multivariable function, and the spectral features separated from the joint features are clustered using a clustering technique to obtain the pure end members.

[0022] Optionally, the neural network of the endmember extraction branch has an activation function layer to constrain the pure endmembers extracted by the Kolmogorov-Arnold network so that the pure endmembers output by the endmember extraction branch are non-negative values.

[0023] Optionally, the abundance estimation branch is a Swin Transformer module.

[0024] Optionally, the output features of the Swin Transformer module are smoothed, and the smoothing is implemented by a smoothing layer so that the sum of the abundance values ​​of all pixels in the target abundance matrix output by the abundance estimation branch is constrained to be 1;

[0025] The smoothing layer includes a 3×3 convolutional layer and a softmax function.

[0026] The beneficial effects that the present invention can produce include:

[0027] A hyperspectral blind unmixing method provided by the present invention does not need to rely on prior endmember information, reduces the dependence of traditional methods on known endmember libraries, and improves the flexibility and applicability of the system. And adopts a dual-branch deep learning architecture, so that endmember extraction and abundance estimation can be carried out in parallel, which not only improves the unmixing efficiency, but also ensures the accuracy and stability of the unmixing results. In addition, the introduction of the feature sharing mechanism enables the spatial features and spectral features extracted by the shallow encoder to be used simultaneously, enhances the information consistency between endmember extraction and abundance estimation, and thus optimizes the unmixing performance. Among them, in terms of abundance estimation, the adopted Swin Transformer module effectively improves the perception ability of the spatial structure of the hyperspectral image, and better captures the long-distance dependencies in the image through the windowed self-attention mechanism and the window offset mechanism. In order to further optimize the unmixing results, the present invention designs a joint loss function, combining the spectral angle distance loss, the mean square error loss, the total variation loss and the regularization loss, to ensure the dual advantages of the unmixing results in terms of accuracy and detail retention. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A flow chart of a hyperspectral blind unmixing method of the present invention;

[0029] Figure 2 Schematic diagram of the unmixing result of the Jasper Ridge data set in the present invention;

[0030] Figure 3Schematic diagram of the unmixing result of the Urban4 data set in the present invention;

[0031] Figure 4 Schematic diagram of the unmixing result of the Urban5 dataset in the present invention;

[0032] Figure 5 Schematic diagram of the unmixing result of the synthetic data set in the present invention. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] like Figure 1 As shown, the present invention provides a hyperspectral blind unmixing method, which includes the following steps: constructing a hyperspectral blind unmixing model, the hyperspectral blind unmixing model has a shallow encoder, an endmember extraction branch and an abundance estimation branch; inputting hyperspectral image data into the hyperspectral blind unmixing model, encoding the spatial features and spectral features of the mixed spectrum of each pixel in the hyperspectral image data through the shallow encoder to generate a joint feature, and synchronously transmitting the joint feature to the endmember extraction branch and the abundance estimation branch through a feature sharing mechanism; wherein the joint feature includes spatial features and spectral features, so that the encoded spatial features and spectral features are shared with the endmember extraction branch and the abundance estimation branch, which can ensure that the endmember extraction branch and the abundance estimation branch jointly use low-level features to ensure the consistency of information, thereby optimizing the learning effect of the dual-branch deep learning architecture, and improving The accuracy and stability of the unmixing results of the hyperspectral blind unmixing model; the endmember extraction branch receives the joint features output by the shallow encoder, and extracts the hidden endmember features from the joint features through the feature extraction module, and then uses the Kolmogorov-Arnold network to extract pure endmembers from the hyperspectral image data; it should be noted that the present invention replaces the traditional multi-layer perceptron with the Kolmogorov-Arnold network, which can directly extract pure endmember spectra from the hyperspectral image data, avoids the dependence of the traditional method on the known endmember library, simplifies the unmixing process, and can adapt to applications in more practical scenarios; the abundance estimation branch receives the joint features output by the shallow encoder, and uses the windowed multi-head self-attention mechanism to extract the abundance map corresponding to the pure endmember from the joint features to obtain the unmixing result of the mixed spectrum, wherein the abundance map represents the relative proportion of each substance in each pixel.

[0035] In the above, hyperspectral image data is collected by a spectral imaging device, and each pixel of the hyperspectral image data corresponds to a mixed spectrum of different substances, and the mixed spectrum includes spatial features and spectral features; wherein the spectral features have several bands, and each band has a different wavelength range.

[0036] In the above, the convolutional neural network structure of the shallow encoder is a 1×1 convolution, which is used to reduce the dimensionality of the input high-dimensional hyperspectral image data while retaining its key spatial and spectral features. Specifically, the joint feature F extracted by the shallow encoder is:

[0037] F = Encoder(Conv(Y));

[0038] Among them, Y represents the input hyperspectral image data, Conv is a 1×1 convolution operation for dimensionality reduction, and Encoder is the encoder.

[0039] Furthermore, the method of extracting pure end members from the joint features by the Kolmogorov-Arnold network is as follows: the joint features are separated by a nonlinear multivariable function, and the spectral features separated from the joint features are clustered by a clustering technique to obtain pure end members. The nonlinear multivariable function f(x) is:

[0040]

[0041] Among them, φ i and φ i,j are continuous single variable functions, x j represents the jth component of the input vector, and the i-th continuous univariate function can be based on x j It should be noted that through this nonlinear multivariable function decomposition method, the Kolmogorov-Arnold network can decompose the mixed spectrum into pure end members of different substances and use E hid Extract the separated pure end members;

[0042]

[0043] Among them, E ext represents the pure end member extracted from the Kolmogorov-Arnold network, KAN represents the operation of the Kolmogorov-Arnold network, E i Indicates E hid The i-th end member curve, e mrepresents the spectral feature, and P is the number of endmembers; in the endmember extraction branch, the number of branch layers of the Kolmogorov-Arnold network is [N, 2N, P]. In this embodiment, the neural network of the endmember extraction branch has an activation function layer to constrain the pure endmembers extracted by the Kolmogorov-Arnold network so that the pure endmembers output by the endmember extraction branch are non-negative values.

[0044] Furthermore, the abundance estimation branch is a Swin Transformer module. The Swin Transformer module extracts local features through a windowed multi-head self-attention mechanism (W-MSA) to achieve smooth conversion between local windows and global windows, where the multi-head self-attention mechanism (W-MSA) is expressed as:

[0045]

[0046] Where Q is the query matrix, K is the key matrix, V is the value matrix, B is the bias matrix, d is the dimension of the query matrix or key matrix, and T is the transpose. The Swin Transformer module (window-based Transformer module) shares information between local and global windows through the self-attention mechanism to capture long-range dependencies in joint features, thereby generating an accurate abundance map. In order to further improve computational efficiency, a window offset mechanism (SW-MAS) is introduced in the Swin Transformer module to achieve cross-window information interaction by offsetting the window by several pixels between two consecutive windows in each layer of self-attention calculation, thereby improving the global perception ability of the abundance estimation branch.

[0047] In the above, by utilizing the nonlinear function decomposition capability of the Kolmogorov-Arnold network, the hyperspectral blind unmixing model can capture more complex spectral features in the abundance estimation process, thereby improving the accuracy of the abundance map generation. Figure 1 As shown, the calculation process of the pixel window offset between two consecutive windows in the Swin Transformer module is as follows:

[0048]

[0049] Among them, z l is the encoding information of the lth layer, is the attention encoding of the lth layer, LN is layer normalization, KAN is the Kolmogorov-Arnold network linear layer, W-MSA is the multi-head self-attention mechanism, and SW-MSA is the multi-head self-attention mechanism with sliding window.

[0050] In the above, in order to constrain the sum of the abundance map to 1, the present invention smoothes the output features of the Swin Transformer module, and the smoothing layer implements the constraint that the sum of the abundance values ​​of all pixels in the target abundance matrix output by the abundance estimation branch is 1; the smoothing layer includes a 3×3 convolutional layer and a softmax function.

[0051] Furthermore, in order to evaluate the accuracy of the unmixing results of the hyperspectral blind unmixing model, the extracted pure endmembers and abundance maps are linearly combined to reconstruct the hyperspectral image, and the consistency of the hyperspectral image and the hyperspectral image data input into the hyperspectral blind unmixing model is compared; if the two are consistent, it means that the accuracy of the unmixing results of the hyperspectral blind unmixing model is high; if the two are inconsistent, it means that the accuracy of the unmixing results of the hyperspectral blind unmixing model is low.

[0052] Furthermore, a joint loss function is used to constrain the endmember extraction branch and abundance estimation branch of the hyperspectral blind unmixing model respectively, so as to iteratively optimize the hyperspectral blind unmixing model during the training process. The joint loss function includes the spectral angular distance (SAD) loss function, the mean square error (MSE) loss function, the total variation (TV) loss function and the l 1 / 2 Regularization loss function; specifically, spectral angle distance loss and mean square error loss are used to ensure that the hyperspectral image reconstructed based on the extracted pure endmembers and abundance map can accurately approximate the hyperspectral image data input by the hyperspectral blind unmixing model; total variation loss and regularization loss are used to enhance the perception of subtle texture and structural information in the hyperspectral image data to ensure that the abundance map can capture the edge features and local features in the hyperspectral image data.

[0053] It should be noted that in the unmixing task based on reconstructed hyperspectral images, the commonly used loss functions are the spectral angular distance (SAD) loss function SADLoss and the mean square error (MSE) loss function MSELoss, which are used together to evaluate the reconstructed hyperspectral image Y re The performance of its combined loss function Loss m The expression is:

[0054] Loss m (Y,Y re ) = α·SADLoss(Y, Y re )+β·MSELoss(Y,Y re );

[0055] The above combined loss function Loss mWithout considering the texture information in complex scenes, in order to make the abundance matrix A more sensitive to the target edge and structure information, and to further reduce noise interference and smooth the abundance matrix, the present invention introduces the total variation (TV) loss to encourage the abundance matrix to be optimized with less noise and smooth edge transition. The expression of the TV loss function TVLoss is as follows:

[0056]

[0057] in, is the horizontal gradient of A, is the vertical gradient of A.

[0058] In the above, since the degree of freedom is too high when solving the abundance matrix A and the end member matrix E in the unmixing task, in order to enhance the robustness and feature extraction ability of the model, the present invention introduces l 1 / 2 Regularize the loss function to stabilize the abundance matrix, l 1 / 2 Regularization loss function Loss 1 / 2 The expression is:

[0059]

[0060] The final joint loss function Loss H U (Y,Y re , A) is:

[0061] Loss HU (Y,Y re , A) = α·SADLoss(Y, Y re )+β·MSELoss(Y,Y re )+γ·TVLoss(A)+λ·Loss 1 / 2 (A);

[0062] Among them, α=1, β=0.95, γ=10 -7 ,λ=10 _8 .

[0063] It is worth noting that the hyperspectral blind unmixing method provided by the present invention does not need to rely on prior endmember information, reduces the dependence of traditional methods on known endmember libraries, and improves the flexibility and applicability of the system. And the dual-branch deep learning architecture is adopted, so that endmember extraction and abundance estimation can be carried out in parallel, which not only improves the unmixing efficiency, but also ensures the accuracy and stability of the unmixing results. In addition, the introduction of the feature sharing mechanism enables the spatial features and spectral features extracted by the shallow encoder to be used simultaneously, enhances the information consistency between endmember extraction and abundance estimation, and thus optimizes the unmixing performance. Among them, in terms of abundance estimation, the adopted Swin Transformer module effectively improves the perception ability of the spatial structure of the hyperspectral image, and better captures the long-distance dependencies in the image through the windowed self-attention mechanism and the window offset mechanism. In order to further optimize the unmixing results, the present invention designs a joint loss function, combining the spectral angular distance (SAD) loss function, the mean square error (MSE) loss function, the total variation (TV) loss function and l 1 / 2 The regularized loss function ensures that the deconvolution results have the dual advantages of accuracy and detail preservation.

[0064] Based on the above embodiments, in order to verify the unmixing results of the hyperspectral blind unmixing model, the present invention conducts multiple experiments to evaluate the effectiveness and accuracy of the method proposed in the present invention in hyperspectral unmixing. Specifically, the experimental method includes the following steps:

[0065] 1) Dataset selection: This embodiment selects multiple public hyperspectral datasets, including the Jasper Ridge dataset, the Urban dataset, and a synthetic dataset. These datasets are widely used for algorithm verification in the field of hyperspectral unmixing.

[0066] In the above, the Jasper Ridge dataset was collected in central California, USA, using the AVIRIS sensor for data collection. The collection scene includes four materials: water, soil, trees, and roads. The data cube consists of 100×100 pixels, and the original data contains 224 bands. The spectral range is 0.38 to 2.5. After removing the influence of water vapor and atmospheric interference, the data retains 198 bands.

[0067] In the above, the Urban dataset was collected in the urban area of ​​Copperas Cove, Texas, USA, using the HYDICEHSI sensor for data collection. Among them, the EMs (endmembers) in the dataset are divided into 4 and 5 categories. Category 4 (Urban4) includes asphalt, grass, trees, and roofs; while category 5 (Urban5) adds soil to category 4 (Urban4). The data cube contains 307×307 pixels. After removing dense water vapor and atmospheric effects, the dataset retains 162 bands with a spectral range of 400 to 2500 and a spectral resolution of 10.

[0068] In the above, the synthetic dataset is a synthetic dataset that contains various mineral electromagnetic curves from the US Geological Survey spectral library. This dataset can generate datasets of different sizes using its built-in multilinear mixed model and has a mixture of various pure EMs. In this method, the present invention selected 10 mineral spectra and created a dataset with a spatial dimension of 128×128 pixels and 216 bands for experiments.

[0069] 2) Setting evaluation index: In quantitative analysis, the present invention uses spectral angle distance (SAD) and root mean square error (RMSE) to evaluate the extracted end member matrix and abundance matrix. Among them, spectral angle distance (SAD) is used to evaluate the estimated end member matrix, which measures the similarity between the estimated end members and the pure end members of each material. The expression of SAD is:

[0070]

[0071] in, represents the ith estimated endmember, E i represents the i-th pure endmember, and P is the number of endmembers in the hyperspectral image. The root mean square error (RMSE) is used to evaluate the abundance matrix, and the expression of RMSE is:

[0072]

[0073] in, represents the abundance matrix of the ith estimate, A i represents the i-th abundance map.

[0074] It should be noted that the ideal values ​​of SAD and RMSE are both 0.

[0075] 3) Experimental method selection: This embodiment selects 8 methods for comparison, namely, fully constrained least squares linear unmixing algorithm (FCLSU), robust non-negative matrix factorization (rNMF), extended linear mixed model (ELMM), linear mixed model (PLMM), three-dimensional convolutional neural network (3DCNN), window sliding based Transformer (swt), spatial-spectral attention bilateral hyperspectral unmixing network (SSABN) and abundance-guided hyperspectral unmixing network (A2SAN). Among them, FCLSU, rNMF, ELMM, PLMM belong to traditional machine learning methods, while 3D-CNN, SwinT, SSABN, A2SAN belong to deep learning methods. In order to ensure a fair comparison, the present invention adopts VCA method to obtain a priori end members, so as to initialize the method requiring a priori end members. In addition, the parameters of the hyperspectral blind unmixing model are all set to default values.

[0076] 4) Experimental results and analysis: Figure 2 As shown in the figure, the performance of the above experimental methods on the Jasper Ridge dataset is demonstrated. For abundance estimation, rNMF and ELMM incorrectly classified the "road" area as "water", while FCLSU, ELMM and PLMM also failed to accurately estimate the "road" area in their abundance estimation. In contrast, deep learning methods outperform traditional machine learning methods in overall performance. However, compared with the ground truth, 3D-CNN's abundance estimation of "road" is significantly overestimated, SSABN has obvious deficiencies in estimating the "soil" area, and Swin Transformer and A2SAN also face difficulties in distinguishing "road" from "soil" in abundance estimation, resulting in some "roads" being incorrectly included in the estimation results of "soil". Compared with traditional methods, the method proposed in the present invention shows higher accuracy in abundance estimation. In the comparison of endmember extraction, the extraction results of "road" have the largest difference, showing that most methods perform poorly in extracting "roads", which is also reflected in abundance estimation. In addition, the endmembers output by some experimental methods produced non-positive results, such as PLMM with negative values ​​in the EM curve of "trees" and rNMF with zero values ​​in the EM curve of "soil". In general, the method proposed in the present invention obtains more accurate results in both abundance estimation and endmember extraction. As shown in Table 1, the present invention lists the performance indicators of all experimental methods on the Jasper Ridge dataset.

[0077] Table 1:

[0078]

[0079] As can be seen from Table 1, the SAD values ​​of traditional machine learning methods are relatively high. This is because most ML methods require an endmember extraction algorithm first, and then further refine the endmember and abundance estimates based on the extraction results. In contrast, data-driven deep learning methods perform better in terms of metric results. Compared with the suboptimal method A2SAN, the method proposed in this paper significantly reduces SAD and RMSE, further demonstrating its effectiveness.

[0080] like Figure 3 As shown in the figure, the performance of the above experimental methods on the Urban4 dataset is demonstrated. Obviously, compared with the JasperRidge dataset, the abundance estimation in GT is more complicated; FCLSU, rNMF, ELMM and PLMM not only have lower accuracy in abundance estimation, but also have various misestimation problems, such as PLMM mistakenly includes the area that should be attributed to "asphalt" in the estimation of "grass"; rNMF mistakenly includes "asphalt" in the estimation of "roof"; and FCLSU has significant overlap between the abundance estimation areas of "trees" and "grass", and underestimates the abundance of "asphalt". Compared with the four traditional machine learning methods, 3D-CNN has some improvement in abundance estimation, but still faces the problem of misestimation; the Swin Transformer module lacks accuracy in area estimation, A2SAN underestimates the abundance of "roof", and SSABN overestimates the abundance of "grass". Reference Figure 5 From the electromagnetic curves shown, it can be seen that although all experimental methods perform relatively well in the extraction of "trees", there are significant deviations in the extraction of "asphalt", "grass" and "roofs", and some experimental methods even produce negative values. As shown in Table 2, the quantitative results of all experimental methods on the Urban4 dataset are shown. Obviously, the method proposed in this invention has achieved satisfactory results in terms of SAD and RMSE.

[0081] Table 2:

[0082]

[0083] like Figure 4As shown in the figure, the abundance estimation and endmember extraction of all experimental methods on the Urban5 dataset are shown. Among them, as the number of endmembers increases, the results of the first four experimental methods fail to achieve ideal abundance estimation; FCLSU cannot produce abundance estimates for "trees" and "roofs", while rNMF incorrectly estimates the abundance of "roofs". 3D-CNN successfully estimates the abundance of five endmembers, but errors still occur in the estimation of "grass" and "trees". SwinT and A2SAN underestimate the abundance of "asphalt" and "soil". Although SSABN and the method proposed in the present invention perform well in abundance estimation, both of them underestimate the newly added endmember "soil". In contrast, the deep learning method shows significant improvement. In the endmember extraction curve, the pure endmembers extracted by most experimental methods have different degrees of deviation, among which the difference in "soil" is the most significant, but the method proposed in the present invention can accurately extract the curve of each endmember. As shown in Table 3, the quantitative results of all experimental methods at different noise intensity levels of the Urban5 dataset are shown. Compared with the results of the Urban4 dataset, the accuracy of all experimental methods has decreased, which indicates that in the same scene, the more detailed the distinction between end members, the more difficult it is to extract. Nevertheless, among all experimental methods, the method proposed in the present invention still maintains a high accuracy rate, and its results are largely unaffected by the increase in noise level, which proves its robustness; the noise signal-to-noise ratio (SNR) used in the experiment is 20dB, 40dB, and 60dB respectively.

[0084] Table 3:

[0085]

[0086] like Figure 5 As shown, the abundance estimation and endmember extraction of all experimental methods on the Synthesis dataset are presented. The results show that most experimental methods have significant deviations in abundance estimation on this dataset. However, compared with deep learning methods, machine learning methods show better estimation results. Among deep learning methods, 3D-CNN's abundance estimation performance is relatively poor. In contrast, SSABN has certain advantages in abundance estimation, but there are serious errors in the estimation of "dust debris" and "glass fiber", which affect the 5th and 7th abundance estimation results, respectively.

[0087] In the above, although the method proposed in the present invention has some deviations in the abundance estimation of "cedar" and "cardboard", it can estimate the abundance estimation of most substances more accurately compared with other experimental methods, especially in the electromagnetic curves of end member extraction. The method proposed in the present invention shows higher accuracy, while the spectral curves of other experimental methods are mostly distorted. As shown in Table 4, the quantitative results of all experimental methods on the Synthesis dataset are shown. At low noise intensity, most machine learning methods have low SAD values ​​and perform better in end member extraction performance. However, as the noise intensity increases, the SAD value of the machine learning method deteriorates more significantly. In contrast, the performance of the deep learning method remains relatively stable, reflecting its strong robustness. In particular, the method proposed in the present invention maintains high quantitative results at all noise intensities, further demonstrating its ability to perform high-precision unmixing in complex scenes with multiple end members.

[0088] Table 4:

[0089]

[0090] In summary, the experimental results of the method proposed in the present invention on a variety of hyperspectral image data sets show that the method has significant advantages in effectiveness and accuracy in hyperspectral unmixing, and also demonstrates its significant advantages in unmixing precision and robustness. It is particularly suitable for multiple practical application scenarios such as remote sensing, environmental monitoring, and agricultural resource assessment, showing broad application prospects and good market potential.

Claims

1. A hyperspectral blind unmixing method, characterized in that: The method comprises the following steps: Constructing a hyperspectral blind unmixing model, wherein the hyperspectral blind unmixing model has a shallow encoder, an endmember extraction branch, and an abundance estimation branch; The hyperspectral image data is input into the hyperspectral blind unmixing model, and the spatial features and spectral features of the mixed spectrum of each pixel in the hyperspectral image data are encoded by a shallow encoder to generate a joint feature, and the joint feature is synchronously transmitted to the end member extraction branch and the abundance estimation branch through a feature sharing mechanism; The endmember extraction branch receives the joint features output by the shallow encoder, extracts hidden endmember features from the joint features through a feature extraction module, and then extracts pure endmembers from the joint features using a Kolmogorov-Arnold network; The abundance estimation branch receives the joint features output by the shallow encoder, and uses a windowed multi-head self-attention mechanism to extract an abundance map corresponding to the pure endmember from the joint features to obtain an unmixing result of the mixed spectrum.

2. A hyperspectral blind unmixing method according to claim 1, characterized in that: The method further comprises linearly combining the extracted pure end members and the abundance map to reconstruct a hyperspectral image, and evaluating the accuracy of an unmixing result of the hyperspectral blind unmixing model by comparing the hyperspectral image with the consistency of the hyperspectral image data input by the hyperspectral blind unmixing model; If the hyperspectral image is consistent with the hyperspectral image data input by the hyperspectral blind unmixing model, it means that the unmixing result of the hyperspectral blind unmixing model has high accuracy; If the hyperspectral image is inconsistent with the hyperspectral image data input by the hyperspectral blind unmixing model, it means that the accuracy of the unmixing result of the hyperspectral blind unmixing model is low.

3. A hyperspectral blind unmixing method according to claim 1, characterized in that: The method further comprises using a joint loss function to respectively constrain the endmember extraction branch and the abundance estimation branch of the hyperspectral blind unmixing model so as to iteratively optimize the hyperspectral blind unmixing model during training, wherein; The joint loss function includes a spectral angle distance loss function, a mean square error loss function, a total variation loss function and a regularization loss function; The spectral angle distance loss and the mean square error loss are used to ensure that the hyperspectral image reconstructed according to the extracted pure end members and abundance map can accurately approximate the hyperspectral image data input by the hyperspectral blind unmixing model; The total variation loss and the regularization loss are used to enhance the perception of subtle texture information and structural information in the hyperspectral image data, so as to ensure that the abundance map can capture edge features and local features in the hyperspectral image data.

4. The hyperspectral blind unmixing method according to claim 1, characterized in that: Each pixel of the hyperspectral image data corresponds to a mixed spectrum of different substances, and the mixed spectrum includes spatial features and spectral features; The spectral feature has several bands, and each band has a different wavelength range.

5. The hyperspectral blind unmixing method according to claim 1, characterized in that: The convolutional neural network structure of the shallow encoder is 1×1 convolution.

6. A hyperspectral blind unmixing method according to claim 1, characterized in that: The method of extracting pure end members from the joint features of the Kolmogorov-Arnold network is: The joint features are separated by a nonlinear multivariable function, and the spectral features separated from the joint features are clustered using a clustering technique to obtain the pure end members.

7. A hyperspectral blind unmixing method according to claim 1, characterized in that: The neural network of the endmember extraction branch has an activation function layer to constrain the pure endmembers extracted by the Kolmogorov-Arnold network so that the pure endmembers output by the endmember extraction branch are non-negative values.

8. A hyperspectral blind unmixing method according to claim 1, characterized in that: The abundance estimation branch is a Swin Transformer module.

9. A hyperspectral blind unmixing method according to claim 8, characterized in that: Smoothing the output features of the SwinTransformer module, wherein the smoothing is performed through a smoothing layer to constrain the sum of abundance values ​​of all pixels in the target abundance matrix output by the abundance estimation branch to be 1; The smoothing layer includes a 3×3 convolutional layer and a softmax function.

Images