Hyperspectral image fusion method and system based on hyper-band aggregation and Dirichlet

Abstract

The invention relates to the technical field of hyperspectral image processing, and provides a hyperspectral image fusion method and system based on hyper-band aggregation and Dirichlet, and the method comprises the steps: obtaining an LRHS image through a remote sensing optical imaging system; fusing the LRHS image and the panchromatic image through a spatial spectrum data distribution fusion module to obtain a PANHS image; carrying out band aggregation on the LRHS through an aggregation algorithm, calculating a compression association graph according to a hyper-band, and extracting a current state feature according to a cyclic memory network; memory state characteristics are obtained through the Dirichlet auto-encoder module; fusion features are obtained through a cyclic memory network; integrating the fusion features into the PANHS image to obtain integrated features; and obtaining a hyperspectral image with high spatial resolution through the residual block. According to the method, the spatial resolution of the hyperspectral image is remarkably improved, meanwhile, the spectral consistency is kept, and technical support is provided for refined analysis and practical application of the hyperspectral image.

Description

Technical Field

[0001] The present invention relates to the technical field of hyperspectral image processing, and in particular, to a hyperspectral image fusion method and system based on super-band aggregation and Dirichlet. Background Art

[0002] Hyperspectral (HS) images are obtained by sampling hundreds of continuous narrow spectral bands using a spectral imaging system, which can provide rich spectral information and are commonly used in the study of spectral difference characteristics of materials. However, due to the physical design limitations of the optical imaging system, the imaging results need to be considered as a trade-off between spectral resolution and spatial resolution. A single-spectral imaging system cannot obtain high-resolution hyperspectral (HRHS) images, and its imaging results often result in low-resolution hyperspectral (LRHS) images, which limits the application of HS images in fields such as mineral detection, ecosystem monitoring, and agricultural detection. The panchromatic (PAN) imaging system only outputs a single-band PAN image and has high spatial resolution but low spectral resolution. The hyperspectral pansharpening task aims to make full use of the spectral information of HS images and the spatial information of PAN images to generate HRHS images. Due to the high demand for HRHS images in the market, this technology has received extensive attention at present and shows good performance in many downstream tasks of remote sensing images such as military surveillance, environmental monitoring, target recognition, and classification.

[0003] Traditional hyperspectral pansharpening methods can be further divided into four categories: methods based on component substitution (CS), methods based on multi-resolution analysis (MRA), methods based on Bayesian, and methods based on models. Although traditional methods can improve the spatial resolution of HS images, due to inappropriate prior knowledge modeling, unknown sensor characteristics, mismatched prior assumptions, and the fact that the features manually constructed from the dictionary do not match the reality, the quality of the fused images will seriously decline, and there are still significant deficiencies in aspects such as hyperspectral image sequence modeling and the interpretability of fusion networks. These problems restrict the popularization and performance improvement of hyperspectral image pansharpening technology in practical applications and urgently need to be broken through and innovated from the theoretical and technical levels. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the related art. To this end, the present invention provides a hyperspectral image fusion method and system based on hyperband aggregation and Dirichlet, which combines the hyperband aggregation method with a recurrent memory network to realize sequence space-spectral fusion modeling between PAN and LRHS images, achieve physical interpretability of hyperspectral decomposition, significantly improve the spatial resolution of hyperspectral images, and maintain the spectral consistency of hyperspectral images, providing technical support for the refined analysis and practical application of hyperspectral images.

[0005] The present invention provides a hyperspectral image fusion method based on hyperband aggregation and Dirichlet, including: S1: Obtain the LRHS image through a remote sensing optical imaging system; S2: Fusion the LRHS image and the panchromatic image through a spatial-spectral data distribution fusion module to obtain the PANHS image; S3: Aggregate the bands of the LRHS image through an aggregation algorithm to obtain hyperbands, calculate the compression correlation graph of the LRHS image according to the hyperbands, and calculate the compression correlation graph of the PANHS image according to the hyperbands; S4: Extract the spatial-spectral correlation of the compression correlation graph of the LRHS image according to the recurrent memory network to obtain the current state features of the LRHS image; Extract the spatial-spectral correlation of the compression correlation graph of the PANHS image according to the recurrent memory network to obtain the current state features of the PANHS image; S5: Fusion the spectral and spatial information of the current state features of the LRHS image and the current state features of the PANHS image through a Dirichlet autoencoder module to obtain memory state features; Extract candidate features through a recurrent memory network, and fuse the memory state features and the candidate features to obtain fusion features; S6: Integrate the fusion features into the PANHS image to obtain integrated features; S7: Repeat steps S3 to S6 to obtain multiple integrated features, and fuse the multiple integrated features through a residual block to obtain a hyperspectral image with high spatial resolution.

[0006] Further, step S2 includes: S21: Upsample the LRHS image to obtain an upsampled LRHS image; S22: Fusion the upsampled LRHS image and the panchromatic image through a spatial-spectral data distribution fusion module to obtain the PANHS image.

[0007] Further, step S22 includes: S221: Calculate the adaptive mean and the adaptive variance of the upsampled LRHS image through parallel convolution, S222: Assign the adaptive mean and the adaptive variance of the upsampled LRHS image to the panchromatic image to obtain the PANHS image.

[0008] Further, step S3 includes: S31: Through an aggregation algorithm, use the high correlation between the waves of the LRHS image for band aggregation to obtain a super band; S32: Calculate the correlation map of the LRHS image according to the super band; Select the first k bands in the correlation map to form a sub-compressed correlation map; S33: Perform channel selection on the LRHS image according to the sub-compressed correlation map, select one band within each super band to obtain the LRHS compressed correlation map, and perform channel selection on the PANHS image according to the sub-compressed correlation map to obtain the PANHS compressed correlation map.

[0009] Further, the aggregation algorithm is an improved K-mean classical aggregation algorithm, and only local iterative calculations are performed during each iteration.

[0010] Further, in step S4, the calculation expression of the current state feature of the LRHS image is: Where is the group of LRHS forget gate states, is the Sigmoid activation function, is the group of LRHS compressed correlation maps, is the group of PANHS forget gate states, is the group of PANHS compressed correlation maps, is the group of LRHS state features, is the Hadamard product, is the group of LRHS state features, is the Tanh activation function, is the group of LRHS observation features; The calculation expression of the current state feature of the PANHS image is: Where is the group of PANHS state features, is the The state characteristics of the i-th PANHS group For the Observation characteristics of the i-th PANHS group

[0011] Furthermore, in step S5, the calculation expression of the memory state characteristics is: Wherein, Is the memory state characteristic of the i-th PANHS group, Is the memory state characteristic of the j-th LRHS group, Is the Dirichlet autoencoder module, Is the State characteristic of the j-th LRHS group, Is the State characteristic of the i-th PANHS group.

[0012] Furthermore, the calculation expression of the fusion feature is: Wherein, Is the i-th reset matrix, Is the Sigmoid activation function, Is the i-th PANHS compression correlation graph, Is the j-th candidate feature, Is the Tanh activation function, Is the i-th fusion feature, Is the Memory state characteristic of the i-th PANHS group.

[0013] Furthermore, the recurrent memory network includes a forget gate layer, an observation feature layer, a reset gate layer, and a candidate feature layer.

[0014] The present invention also provides a hyperspectral image fusion system based on hyperband aggregation and Dirichlet, for executing the above-mentioned hyperspectral image fusion method based on hyperband aggregation and Dirichlet, including: An acquisition module, the acquisition module is used to acquire the LRHS image through a remote sensing optical imaging system; A first fusion module, the first fusion module is used to fuse the LRHS image and the panchromatic image through a spatial spectral data distribution fusion module to obtain the PANHS image; Aggregation module, which is used to perform band aggregation on the LRHS through an aggregation algorithm to obtain a super-band, calculate the LRHS image compression correlation graph based on the super-band, and calculate the PANHS image compression correlation graph based on the super-band; Feature extraction module, which is used to extract the spatial-spectral correlation of the LRHS image compression correlation graph according to the recurrent memory network to obtain the current state features of the LRHS image, and extract the spatial-spectral correlation of the PANHS image compression correlation graph according to the recurrent memory network; obtain the current state features of the PANHS image; Second fusion module, which is used to fuse the spectral and spatial information of the current state features of the LRHS image and the current state features of the PANHS image through the Dirichlet autoencoder module to obtain memory state features; extract candidate features through the recurrent memory network, and fuse the memory state features and the candidate features to obtain fused features; Integration module, which is used to integrate the fused features into the PANHS image to obtain integrated features; Third fusion module, which is used to obtain multiple integrated features and fuse the multiple integrated features through residual blocks to obtain a high-spatial-resolution hyperspectral image.

[0015] One or more of the above technical solutions in the embodiments of the present invention have at least one of the following technical effects: By combining the super-band aggregation method with the recurrent memory network, sequence spatial-spectral fusion modeling between the PAN and LRHS images is realized, the physical interpretability of hyperspectral decomposition is achieved, the spatial resolution of the hyperspectral image is significantly improved, and the spectral consistency of the hyperspectral image is maintained at the same time, providing technical support for the refined analysis and practical application of hyperspectral images.

[0016] The additional aspects and advantages of the present invention will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present invention. Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a schematic flowchart of a hyperspectral image fusion method based on super-band aggregation and Dirichlet provided by the present invention.

[0019] Figure 2It is a schematic diagram of the network structure of a hyperspectral image fusion method based on hyperband aggregation and Dirichlet provided by the present invention.

[0020] Figure 3 It is a schematic diagram of the structure of the spatial spectral data distribution fusion module provided by the present invention.

[0021] Figure 4 It is a schematic diagram of the structure of the cyclic memory fusion module provided by the present invention.

[0022] Figure 5 It is a schematic diagram of the structure of a hyperspectral image fusion system based on hyperband aggregation and Dirichlet provided by the present invention.

[0023] Figure 6 It is a schematic diagram of the experimental effect of the comparative experiment of the quantitative index results of the present invention on the Pavia center dataset.

[0024] Reference signs: 101, acquisition module; 102, first fusion module; 103, aggregation module; 104, feature extraction module; 105, second fusion module; 106, integration module; 107, third fusion module. Detailed implementation manners

[0025] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments in the present invention belong to the scope of protection of the present invention. The following embodiments are used to illustrate the present invention, but cannot be used to limit the scope of the present invention.

[0026] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the embodiments of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without conflict, those skilled in the art can combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0027] The following combines Figures 1 to 6 to describe the hyperspectral image fusion method and system based on hyperband aggregation and Dirichlet of the present invention.

[0028] As Figure 1 shown, a hyperspectral image fusion method based on super-band aggregation and Dirichlet includes: S1: Obtain the LRHS image through a remote sensing optical imaging system; Obtain images of remote sensing optical imaging systems in fields such as mineral detection, ecosystem monitoring, and agricultural detection.

[0029] S2: Fusion the LRHS image and the panchromatic image through a spatial spectral data distribution fusion module to obtain a PANHS image (panchromatic image with spectral information), S21: Upsample the LRHS image to obtain an upsampled LRHS image; In some specific embodiments of the present invention, the LRHS image is upsampled by 4 times using bicubic interpolation to obtain an upsampled LRHS image.

[0030] S22: Fusion the upsampled LRHS image and the panchromatic image through a spatial spectral data distribution fusion module to obtain a PANHS image; S221: Calculate the adaptive mean and adaptive variance of the upsampled LRHS image through parallel convolution; S222: Assign the adaptive mean and adaptive variance of the upsampled LRHS image to the panchromatic image to obtain a PANHS image; The calculation expression of the PANHS image is: Where, is the panchromatic image with spectral information, is the panchromatic image, is the variance calculation, is the LRHS image after being upsampled by 4 times using bicubic interpolation, is the convolution operation with a 3×3 convolution kernel, is the mean calculation.

[0031] The structure of the spatial spectral data distribution fusion module is as Figure 3 shown.

[0032] S3: Aggregate the bands of the LRHS image through an aggregation algorithm to obtain super-bands, calculate the compression correlation graph of the LRHS image based on the super-bands, and calculate the compression correlation graph of the PANHS image based on the super-bands; S31: Through the aggregation algorithm, use the high correlation between the waves of the LRHS image to perform band aggregation to obtain super-bands; The aggregation algorithm is an improved K-mean classical aggregation algorithm, and only local iterative calculations are performed during each iteration; S32: Calculate the correlation graph of the LRHS image based on the super-bands; select the first k bands in the correlation graph to form a sub-compressed correlation graph; In some specific embodiments of the present invention, assume the hyperspectral image , where C is the number of channels, H is the height of the image, and W is the width of the image, is a real number matrix. Assume the number of internal channels of the super-band is , then the number of super-bands , , assume the channel is a channel in the th super-band. belongs to . During each iteration process, each channel only calculates with the surrounding adjacent super-bands, ensuring high calculation efficiency. In the nd iteration, calculate the correlation graph of the LRHS image. The calculation expression is: where, is the value of the th row and th column, is the square of the Frobenius distance between the image of the channel in the th row and the image of the th column in the nd iteration of the super-band .

[0033] is an element in the correlation graph of the LRHS image, .

[0034] S33: Perform channel selection on the LRHS image according to the sub-compressed correlation graph, select one band within each super-band to obtain the LRHS compressed correlation graph, and perform channel selection on the PANHS image according to the sub-compressed correlation graph to obtain the PANHS compressed correlation graph.

[0035] S4: Extract the spatial-spectral correlation of the LRHS image compressed correlation graph according to the recurrent memory network to obtain the current state feature of the LRHS image, and extract the spatial-spectral correlation of the PANHS image compressed correlation graph according to the recurrent memory network; obtain the current state feature of the PANHS image; The calculation expression of the current state feature of the LRHS image is: where, is the The group of LRHS forgetting gate states, is the Sigmoid activation function, is the group of LRHS compressed association graphs, is the group of PANHS forgetting gate states, is the group of PANHS compressed association graphs, is the group of state features of LRHS, is the Hadamard product, is the group of state features of LRHS, is the Tanh activation function, is the group of LRHS observation features; The calculation expression of the current state feature of the PANHS image is: Among them, is the group of state features of PANHS, is the group of state features of PANHS, is the group of PANHS observation features.

[0036] S5: Fuse the spectral and spatial information of the current state features of the LRHS image and the current state features of the PANHS image through the Dirichlet autoencoder module to obtain the memory state features; extract the candidate features through the recurrent memory network, and fuse the memory state features and the candidate features to obtain the fused features; The calculation expression of the memory state features is: Among them, is the group of memory state features of PANHS, is the group of memory state features of LRHS, is the Dirichlet autoencoder module, is the group of state features of LRHS, is the group of state features of PANHS.

[0037] In the hyperspectral field, hyperspectral images can be unmixed and decomposed into the matrix product of an abundance matrix and an endmember matrix. Hyperspectral image unmixing is achieved by constructing an encoder-decoder structure based on the Dirichlet idea. The abundance matrix can be generated by the encoder, and then the decoder parameters serve as the endmember matrix. After the abundance matrix output by the encoder passes through the decoder, the unmixed hyperspectral image is obtained. Among them, the endmember matrix contains spectral information. The LRHS and HRHS images have the same endmember matrix, but the abundance matrix of LRHS contains less spatial information. Different encoders are used to decompose and calculate the abundance matrices of LRHS and PANHS. Given that the endmember matrix of LRHS contains complete spectral information, the fusion of spatial and spectral information is achieved by sharing the decoders of both, and finally it contains the rich spatial state memory in PANHS and the complete spectral state memory in LRHS. Then, the final information fusion is completed through the reset gate, and the calculation expression is: Among them, is the th group of reset matrices, is the Sigmoid activation function, is the th group of PANHS compressed correlation maps, is the th group of candidate features, is the Tanh activation function, is the th group of fused features, is the th group of memory state features of PANHS.

[0038] S6: According to the PANHS image compressed correlation map, integrate the fused features into the PANHS image with spectral information to obtain the integrated features; Based on the position index in the PANHS image compressed correlation map, add the fused features to the PANHS image. The calculation expression of the integrated features is: Among them, is the integrated feature, is the feature integration, is the first group of fused features, is the second group of fused features, is the th group of fused features, It is a panchromatic image.

[0039] S7: Repeat steps S3 to S6 to obtain multiple integrated features, and fuse the multiple integrated features through a residual block to obtain a hyperspectral image with high spatial resolution.

[0040] The recurrent memory network includes a forget gate layer, an observation feature layer, a reset gate layer, and a candidate feature layer.

[0041] The recurrent memory network and the Dirichlet autoencoder module form a recurrent memory fusion module, as shown in Figure 4 shown. The overall network structure of the present invention is as shown in Figure 2 shown.

[0042] As shown in Figure 5 shown, a hyperspectral image fusion system based on superband aggregation and Dirichlet is used to execute a hyperspectral image fusion method based on superband aggregation and Dirichlet, including: The acquisition module 101 is used to acquire the LRHS image through a remote sensing optical imaging system; The first fusion module 102 is used to fuse the LRHS image and the panchromatic image through a spatial spectral data distribution fusion module to obtain the PANHS image; The aggregation module 103 is used to perform band aggregation on the LRHS image through an aggregation algorithm to obtain a superband, calculate the LRHS image compression correlation graph according to the superband, and calculate the PANHS image compression correlation graph according to the superband; The feature extraction module 104 is used to extract the spatial spectral correlation of the LRHS image compression correlation graph according to the recurrent memory network to obtain the current state feature of the LRHS image, and extract the spatial spectral correlation of the PANHS image compression correlation graph according to the recurrent memory network; obtain the current state feature of the PANHS image; The second fusion module 105 is used to fuse the spectral and spatial information of the current state feature of the LRHS image and the current state feature of the PANHS image through the Dirichlet autoencoder module to obtain a memory state feature; extract candidate features through the recurrent memory network, and fuse the memory state feature and the candidate features to obtain a fusion feature; The integration module 106 is used to integrate the fusion feature into the PANHS image to obtain an integrated feature; The third fusion module 107 is used to obtain multiple integrated features, and fuse the multiple integrated features through a residual block to obtain a hyperspectral image with high spatial resolution.

[0043] Through the collaborative work of the above-mentioned modules, by combining the hyperband aggregation method with the recurrent memory network, the sequential spatial-spectral fusion modeling between PAN and LRHS images is realized, the physical interpretability of hyperspectral decomposition is achieved, the spatial resolution of hyperspectral images is significantly improved, and the spectral consistency of hyperspectral images is maintained simultaneously, providing technical support for the refined analysis and practical application of hyperspectral images.

[0044] In the present invention, quantitative index result comparison experiments and subjective comparison experiments are carried out with a variety of internationally advanced algorithms on the public dataset Pavia center. As Figure 6 shown, in order to better observe the differences between the fusion results, the mean absolute error (MAE) map between the fusion results and the ground truth (GT) is calculated and displayed, and a local area is magnified and shown in the lower right corner of the result, and the corresponding RGB image is shown in the lower left corner. These images indicate that there are more dark blue areas in the MAE map of the method of the present invention, indicating the smallest fusion error. Figure 6 Among them, the internationally advanced algorithms are FPFNet (2023), HyperRefiner (2023), TreeSNet (2024), DFCFN(2025). FPFNet (2023) is the 2023 version of the feature pyramid fusion network, HyperRefiner (2023) is the 2023 version of the refined hyperspectral panspectral network, TreeSNet (2024) is the 2024 version of the tree-structured neural network, and DFCFN (2025) is the 2025 version of the two-stage feature correction fusion network. As shown in Table 1, it can also be seen from the quantitative index results that the quantitative result indexes of the algorithm of the present invention are better than those of the internationally advanced algorithms in recent years.

[0045] Table 1 Comparison experiment of quantitative index results for Pavia center dataset Among them, CC is the correlation coefficient, SAM is the spectral angle mapping, RMSE is the root mean square error, ERGAS is the dimensionless global relative error, PSNR is the peak signal-to-noise ratio, "↑" indicates that the higher the value, the better, and "↓" indicates that the lower the value, the better.

[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A hyperspectral image fusion method based on superband aggregation and Dirichlet, characterized in that: include: S1: Acquire LRHS images through remote sensing optical imaging system; S2: The LRHS image and the panchromatic image are fused through the spatial spectral data distribution fusion module to obtain the PANHS image; S3: perform band aggregation on the LRHS image through an aggregation algorithm to obtain a super-band, calculate the compression association map of the LRHS image based on the super-band, and calculate the compression association map of the PANHS image based on the super-band; S4: extracting the spatial spectral correlation of the LRHS image compression association graph according to the recurrent memory network to obtain the current state characteristics of the LRHS image; extracting the spatial spectral correlation of the PANHS image compression association graph according to the recurrent memory network to obtain the current state characteristics of the PANHS image; S5: The memory state feature is obtained by fusing the spectral and spatial information of the current state feature of the LRHS image and the current state feature of the PANHS image through the Dirichlet autoencoder module; Extract candidate features through a recurrent memory network, and fuse the memory state features with the candidate features to obtain fused features; S6: Integrate the fused features into the PANHS image to obtain the integrated features; S7: Repeat steps S3 to S6 to obtain multiple integrated features, and fuse the multiple integrated features through the residual block to obtain a hyperspectral image with high spatial resolution.

2. The hyperspectral image fusion method based on superband aggregation and Dirichlet according to claim 1 is characterized in that: The S2 step includes: S21: obtaining an upsampled LRHS image by upsampling the LRHS image; S22: The upsampled LRHS image and the panchromatic image are fused through the spatial spectral data distribution fusion module to obtain a PANHS image.

3. The hyperspectral image fusion method based on superband aggregation and Dirichlet according to claim 2 is characterized in that: Step S22 includes: S221: Calculate the adaptive mean of the upsampled LRHS image and the adaptive variance of the upsampled LRHS image by parallel convolution, S222: Assign the adaptive mean of the upsampled LRHS image and the adaptive variance of the upsampled LRHS image to the panchromatic image to obtain a PANHS image.

4. The hyperspectral image fusion method based on superband aggregation and Dirichlet according to claim 1 is characterized in that: The S3 steps include: S31: Through the aggregation algorithm, the high correlation between LRHS image waves is used to aggregate the bands and obtain the super band; S32: calculating a correlation map of the LRHS image according to the superband; selecting the first k bands in the correlation map to form a sub-compression correlation map; S33: performing channel selection on the LRHS image according to the secondary compression association map, selecting a band in each superband to obtain the LRHS compression association map, performing channel selection on the PANHS image according to the secondary compression association map to obtain the PANHS compression association map.

5. The hyperspectral image fusion method based on superband aggregation and Dirichlet according to claim 1, characterized in that: The aggregation algorithm is an improved K-mean classical aggregation algorithm, and only local iterative calculations are performed in each iteration.

6. The hyperspectral image fusion method based on superband aggregation and Dirichlet according to claim 1, characterized in that: In step S4, the calculation expression of the current state feature of the LRHS image is: in, For the Group LRHS forget gate state, is the Sigmoid activation function, For the Group LRHS compressed association graph, For the Set the PANHS forget gate state, For the Group PANHS compressed association graph, For the Status characteristics of group LRHS, is the Hadamard product, For the Status characteristics of group LRHS, is the Tanh activation function, For the Group LRHS observation characteristics; The calculation expression of the current state feature of the PANHS image is: in, For the Status characteristics of the PANHS group, For the Status characteristics of the PANHS group, For the Observational characteristics of the PANHS group.

7. The hyperspectral image fusion method based on superband aggregation and Dirichlet according to claim 1 is characterized in that: In step S5, the calculation expression of the memory state feature is: in, For the Memory status characteristics of group PANHS, For the The memory state characteristics of group LRHS, is the Dirichlet autoencoder module, For the Status characteristics of group LRHS, For the Status characteristics of the PANHS group.

8. The hyperspectral image fusion method based on superband aggregation and Dirichlet according to claim 1, characterized in that: The calculation expression of fusion features is: in, For the Group reset matrix, is the Sigmoid activation function, For the Group PANHS compressed association graph, For the Set of candidate features, is the Tanh activation function, For the Group fusion features, For the Memory status characteristics of group PANHS.

9. The hyperspectral image fusion method based on superband aggregation and Dirichlet according to claim 1, characterized in that: The recurrent memory network includes a forget gate layer, an observation feature layer, a reset gate layer and a candidate feature layer.

10. A hyperspectral image fusion system based on superband aggregation and Dirichlet, characterized in that: The method for performing a hyperspectral image fusion method based on superband aggregation and Dirichlet as claimed in any one of claims 1 to 9 comprises: An acquisition module, wherein the acquisition module is used to acquire a LRHS image through a remote sensing optical imaging system; A first fusion module, the first fusion module is used to fuse the LRHS image and the panchromatic image through the spatial spectral data distribution fusion module to obtain a PANHS image; An aggregation module, wherein the aggregation module is used to perform band aggregation on the LRHS image through an aggregation algorithm to obtain a super-band, calculate a compression association graph of the LRHS image according to the super-band, and calculate a compression association graph of the PANHS image according to the super-band; A feature extraction module, wherein the feature extraction module is used to extract the spatial spectral correlation of the LRHS image compression association graph according to the recurrent memory network to obtain the current state feature of the LRHS image, and to extract the spatial spectral correlation of the PANHS image compression association graph according to the recurrent memory network to obtain the current state feature of the PANHS image; A second fusion module, wherein the second fusion module is used to fuse the spectral and spatial information of the current state features of the LRHS image and the current state features of the PANHS image through the Dirichlet autoencoder module to obtain a memory state feature; extract candidate features through a recurrent memory network, and fuse the memory state features and the candidate features to obtain a fused feature; An integration module, wherein the integration module is used to integrate the fusion features into the PANHS image to obtain the integration features; The third fusion module is used to obtain multiple integrated features, and fuse the multiple integrated features through the residual block to obtain a hyperspectral image with high spatial resolution.

Images