A snapshot compression imaging reconstruction method and system based on dual low-rank regularization

Abstract

The present invention relates to the field of image processing, and proposes a snapshot compression imaging reconstruction method and system based on dual low-rank regularization. By designing a dual low-rank regularization model, three-dimensional hyperspectral image data is two-dimensionalized, thereby ensuring the quality of original data and avoiding the problem of poor reconstruction effect caused by data loss or error. The dual low-rank regularization model combines a low-rank group sparse prior algorithm and a deep low-rank decomposition algorithm to form a dual low-rank regularization framework. The low-rank group sparse prior algorithm is used to extract prior information of global spectral correlation and local spatial smoothness, respectively, and the texture detail features of the reconstruction result are enhanced. The deep low-rank decomposition algorithm is used to extract spatial factors and spectral factors, respectively, and deep information is extracted quickly and effectively, further avoiding the problem of texture detail loss and spectral distortion in the reconstruction result. The present invention improves the reconstruction accuracy and authenticity of snapshot compression imaging.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to a snapshot compression imaging reconstruction method and system based on dual low-rank regularization. Background Art

[0002] Hyperspectral images contain dozens to hundreds of spectral bands and provide richer spectral information compared to traditional red, green and blue (RGB) images. With this significant advantage, hyperspectral images have been widely used in many fields such as remote sensing, face recognition and medical image processing.

[0003] In the existing technology, obtaining a complete hyperspectral image requires multiple exposure scans of the scene through a spectrometer. This process is not only time-consuming and labor-intensive, but also difficult to adapt to the needs of dynamic scenes. In recent years, many snapshot compression imaging methods based on compressed sensing theory have emerged to capture three-dimensional hyperspectral image data. Because they can achieve fast imaging at a lower cost and bandwidth, they have become key technologies for application in dynamic scenes. However, the existing snapshot compression imaging methods are prone to texture detail loss and spectral distortion during the compression and reconstruction process, resulting in poor reconstruction effects and the reconstruction results cannot meet application requirements.

[0004] Therefore, how to design a snapshot compression imaging reconstruction method to improve the reconstruction quality has become an urgent problem to be solved. Summary of the invention

[0005] Based on this, the present invention proposes a snapshot compression imaging reconstruction method and system based on dual low-rank regularization. By designing a dual low-rank regularization model, the three-dimensional hyperspectral image data is two-dimensionalized, thereby ensuring the quality of the original data and avoiding the problem of poor reconstruction effect due to data loss or error. The dual low-rank regularization model combines the low-rank group sparse prior algorithm and the deep low-rank decomposition algorithm to form a dual low-rank regularization framework. The low-rank group sparse prior algorithm is used to extract the prior information of the global spectral correlation and the local spatial smoothness, respectively, and the texture detail features of the reconstruction results are enhanced. The deep low-rank decomposition algorithm is used to extract the spatial factors and the spectral factors, respectively, and the deep information is extracted quickly and effectively, further avoiding the problems of texture detail loss and spectral distortion in the reconstruction results. The present invention improves the reconstruction accuracy and authenticity of snapshot compression imaging.

[0006] The present invention proposes a snapshot compression imaging reconstruction method based on dual low-rank regularization, comprising:

[0007] The three-dimensional hyperspectral image data is processed into two dimensions through a dual low-rank regularization model to obtain two-dimensional snapshot measurement data, wherein the two-dimensional processing is based on a coded aperture snapshot spectral imaging algorithm;

[0008] Extracting prior information of the two-dimensional snapshot measurement data according to a low-rank group sparse prior algorithm, wherein the low-rank group sparse prior algorithm includes a group sparse regularization term, and the prior information includes global spectral prior information and local spatial prior information;

[0009] Extracting low-dimensional subspace factors of the two-dimensional snapshot measurement data according to a deep low-rank decomposition algorithm, wherein the deep low-rank decomposition algorithm is used to decompose the two-dimensional snapshot measurement data, and the low-dimensional subspace factors include spatial factors and spectral factors;

[0010] Reconstructed target three-dimensional hyperspectral image data is obtained according to the prior information and the low-dimensional subspace factor.

[0011] In summary, according to the above-mentioned snapshot compression imaging reconstruction method based on dual low-rank regularization, by designing a dual low-rank regularization model, the three-dimensional hyperspectral image data is two-dimensionalized, the quality of the original data is guaranteed, and the problem of poor reconstruction effect caused by data loss or error is avoided. The dual low-rank regularization model combines the low-rank group sparse prior algorithm and the deep low-rank decomposition algorithm to form a dual low-rank regularization framework. The low-rank group sparse prior algorithm is used to extract the prior information of global spectral correlation and local spatial smoothness, respectively, and the texture detail features of the reconstruction results are enhanced. The deep low-rank decomposition algorithm is used to extract spatial factors and spectral factors, respectively, and the deep information is extracted quickly and effectively, further avoiding the problems of texture detail loss and spectral distortion in the reconstruction results. The present invention improves the reconstruction accuracy and authenticity of snapshot compression imaging. Specifically, the three-dimensional hyperspectral image data is two-dimensionalized through a dual low-rank regularization model to obtain two-dimensional snapshot measurement data. The two-dimensional processing is based on a coded aperture snapshot spectral imaging algorithm, which ensures the quality of the original data and avoids poor reconstruction effect due to data problems. The prior information of the two-dimensional snapshot measurement data is extracted according to a low-rank group sparse prior algorithm, which includes a group sparse regularization term. The prior information includes global spectral prior information and local spatial prior information, which enhances the overall texture detail features and further improves the accuracy of the reconstruction results. The deep low-rank decomposition algorithm extracts low-dimensional subspace factors of the two-dimensional snapshot measurement data. The deep low-rank decomposition algorithm is used to decompose the two-dimensional snapshot measurement data. The low-dimensional subspace factors include spatial factors and spectral factors. The low-dimensional subspace factors containing deep information are extracted quickly and effectively, further avoiding the problems of texture detail loss and spectral distortion in the reconstruction results. According to the prior information and the low-dimensional subspace factors, the reconstructed target three-dimensional hyperspectral image data is obtained, which further enhances the reliability of the model and the reconstruction quality. The present invention improves the reconstruction accuracy and authenticity of snapshot compression imaging.

[0012] Furthermore, the step of converting the three-dimensional hyperspectral image data into two-dimensional data by using a dual low-rank regularization model to obtain two-dimensional snapshot measurement data specifically includes:

[0013] Collect three-dimensional hyperspectral image data and input it into a dual low-rank regularization model to convert the three-dimensional hyperspectral image data into two-dimensional data to obtain two-dimensional snapshot measurement data;

[0014] The specific algorithm of the two-dimensional processing is as follows:

[0015] ,

[0016] in, represents the 2D snapshot measurement data, represents the modulation and dispersion operation, represents random noise, Represents pixel value.

[0017] Furthermore, the step of extracting the prior information of the two-dimensional snapshot measurement data according to the low-rank group sparse prior algorithm specifically includes:

[0018] The prior information of the two-dimensional snapshot measurement data is extracted according to the low-rank group sparse prior algorithm, and the low-rank group sparse prior algorithm is specifically as follows:

[0019] ,

[0020] ,

[0021] ,

[0022] in, represents the 2D snapshot measurement data, represents the pixel value, represents the regularization parameter, express The differential operator of and Represent two differential operators in the spatial dimension, Represent the coordinates of the pixel in three directions of space, Indicates the length of the image data. Indicates the width of the image data, represents the Frobenius norm.

[0023] Furthermore, the step of extracting the low-dimensional subspace factors of the two-dimensional snapshot measurement data according to the deep low-rank decomposition algorithm specifically includes:

[0024] Extracting low-dimensional subspace factors of the two-dimensional snapshot measurement data according to a deep low-rank decomposition algorithm, wherein the low-dimensional subspace factors include spatial factors and spectral factors;

[0025] The deep low-rank decomposition algorithm includes a spatial network and a spectral network;

[0026] The spatial network captures the spatial factors of the target three-dimensional hyperspectral image data according to the unsupervised guided deep decoding network;

[0027] The spectral network is an unsupervised one-dimensional convolutional network with a U-Net structure, which is used to capture the spectral factors of the target three-dimensional hyperspectral image data;

[0028] The deep low-rank decomposition algorithm is specifically as follows:

[0029] ,

[0030] in, represents the pixel value, represents the spatial representation coefficient, represents a spectral orthogonal basis, and represents the learning weight parameter, and denote spatial network and spectral network respectively, and represents random noise.

[0031] Furthermore, the spatial network specifically includes:

[0032] The spatial network includes an encoder-decoder network and a deep decoding network, wherein the encoder-decoder network has skip connections;

[0033] The encoder-decoder network is a two-dimensional convolutional network with a U-Net structure;

[0034] The deep decoding network consists of an upsampling refinement unit and a feature refinement unit;

[0035] The upsampling refinement unit is an attention gate structure, the original branch in the attention gate structure of the upsampling refinement unit includes an upsampling layer and a recursive normalization layer, and the attention weight branch in the attention gate structure of the upsampling refinement unit includes a two-dimensional convolution layer, a LeakyReLU activation function layer, and a Sigmoid activation function layer;

[0036] The feature refinement unit is an attention gate structure, the original branch in the feature refinement unit attention gate structure includes a multidimensional convolution layer and a two-dimensional convolution layer, and the attention weight branch in the feature refinement unit attention gate structure includes a two-dimensional convolution layer, a LeakyReLU activation function layer and a Sigmoid activation function layer;

[0037] The spatial network obtains multi-scale information of the guidance image according to the encoder-decoder network, and performs deep decoding feature weighting according to the channel attention gate to align spatial information and semantic information.

[0038] Furthermore, the step of obtaining the reconstructed target three-dimensional hyperspectral image data according to the prior information and the low-dimensional subspace factor further includes:

[0039] The dual low-rank regularization model is optimized according to the ADMM algorithm to obtain an optimized model through cyclic iteration. The optimized model is specifically as follows:

[0040] ,

[0041] in, represents the overall optimization model, represents the pixel value, and represents the learning weight parameter, and represents auxiliary variables, represents the Lagrange multiplier, represents the 2D snapshot measurement data, represents the modulation and dispersion operation, and denote spatial network and spectral network respectively, and represents random noise, represents the regularization parameter, express The difference operator of represents the penalty parameter, represents the Frobenius norm.

[0042] Furthermore, the step of optimizing the dual low-rank regularization model according to the ADMM algorithm further includes:

[0043] The Lagrange multiplier is updated according to the dual ascent method. The specific algorithm for updating the Lagrange multiplier is as follows:

[0044] ,

[0045] in, represents the Lagrange multiplier.

[0046] The present invention proposes a snapshot compression imaging reconstruction system based on dual low-rank regularization, comprising:

[0047] A two-dimensionalization module, used for performing two-dimensional processing on three-dimensional hyperspectral image data through a dual low-rank regularization model to obtain two-dimensional snapshot measurement data, wherein the two-dimensional processing is based on a coded aperture snapshot spectral imaging algorithm;

[0048] A priori module, used for extracting prior information of the two-dimensional snapshot measurement data according to a low-rank group sparse prior algorithm, wherein the low-rank group sparse prior algorithm includes a group sparse regularization term, and the prior information includes global spectral prior information and local spatial prior information;

[0049] A low-rank subspace decomposition module, used for extracting low-dimensional subspace factors of the two-dimensional snapshot measurement data according to a deep low-rank decomposition algorithm, wherein the deep low-rank decomposition algorithm is used for decomposing the two-dimensional snapshot measurement data, and the low-dimensional subspace factors include spatial factors and spectral factors;

[0050] The reconstruction module is used to obtain reconstructed target three-dimensional hyperspectral image data according to the prior information and the low-dimensional subspace factor.

[0051] The present invention also provides a storage medium, wherein the storage medium stores one or more programs, and when the programs are executed by a processor, the snapshot compression imaging reconstruction method based on double low-rank regularization as described above is implemented.

[0052] The present invention also provides a computer device, the computer device comprising a memory and a processor, wherein:

[0053] The memory is used to store computer programs;

[0054] When the processor is used to execute the computer program stored in the memory, the snapshot compression imaging reconstruction method based on double low-rank regularization as described above is implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A flowchart of a snapshot compression imaging reconstruction method based on dual low-rank regularization proposed in the first embodiment of the present invention;

[0056] Figure 2 A flowchart of a snapshot compression imaging reconstruction method based on dual low-rank regularization proposed in the second embodiment of the present invention;

[0057] Figure 3 A schematic diagram of the structure of a snapshot compression imaging reconstruction system based on dual low-rank regularization proposed in the third embodiment of the present invention;

[0058] Figure 4 A diagram showing the spectral low-rank properties and spatial smoothing structure of the snapshot compression imaging reconstruction method based on dual low-rank regularization proposed in the first embodiment of the present invention;

[0059] Figure 5 A subspace low-rank decomposition network structure diagram of the snapshot compression imaging reconstruction method based on double low-rank regularization proposed in the first embodiment of the present invention;

[0060] Figure 6 This is a comparison diagram of reconstruction results of the snapshot compression imaging reconstruction method based on dual low-rank regularization proposed in the second embodiment of the present invention.

[0061] The following specific implementation manner will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0062] In order to facilitate understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Several embodiments of the present invention are provided in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0063] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0064] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0065] See also Figure 1 , see Figure 1 , which is a flowchart of a snapshot compression imaging reconstruction method based on dual low-rank regularization proposed in the first embodiment of the present invention, the snapshot compression imaging reconstruction method based on dual low-rank regularization comprises steps S01 to S04, wherein:

[0066] Step S01: converting three-dimensional hyperspectral image data into two-dimensional data by using a dual low-rank regularization model to obtain two-dimensional snapshot measurement data;

[0067] It should be noted that in this embodiment, three-dimensional hyperspectral image data is collected and input into a dual low-rank regularization model, and the three-dimensional hyperspectral image data is two-dimensionalized to obtain two-dimensional snapshot measurement data;

[0068] The specific algorithm of the two-dimensional processing is as follows:

[0069] ,

[0070] in, represents the 2D snapshot measurement data, represents the modulation and dispersion operation, represents random noise, Represents pixel value.

[0071] Step S02: extracting prior information of two-dimensional snapshot measurement data according to a low-rank group sparse prior algorithm;

[0072] It should be noted that in this embodiment, the prior information of the two-dimensional snapshot measurement data is extracted according to the low-rank group sparse prior algorithm, and the low-rank group sparse prior algorithm is specifically as follows:

[0073] ,

[0074] ,

[0075] ,

[0076] in, represents the 2D snapshot measurement data, represents the pixel value, represents the regularization parameter, express The differential operator of and Represent two differential operators in the spatial dimension, Represent the coordinates of the pixel in three directions of space, Indicates the length of the image data. Indicates the width of the image data, represents the Frobenius norm;

[0077] The spectral low-rank property and spatial smooth structure of this embodiment can be found in Figure 4 , Figure 4 It reflects the low-rank group sparse prior of three-dimensional hyperspectral image data, that is, the prior information of two-dimensional snapshot measurement data.

[0078] Step S03: extracting low-dimensional subspace factors of the two-dimensional snapshot measurement data according to a deep low-rank decomposition algorithm;

[0079] It should be noted that in this embodiment, low-dimensional subspace factors of two-dimensional snapshot measurement data are extracted according to a deep low-rank decomposition algorithm, and the low-dimensional subspace factors include spatial factors and spectral factors;

[0080] The deep low-rank decomposition algorithm includes a spatial network and a spectral network;

[0081] For the specific network structure of the subspace low-rank decomposition network in the deep low-rank decomposition algorithm, please refer to Figure 5 , it can be seen that the spatial factor and spectral factor are obtained through the spatial network and spectral network mapping;

[0082] The spatial network captures the spatial factors of the target three-dimensional hyperspectral image data according to the unsupervised guided deep decoding network;

[0083] The spectral network is an unsupervised one-dimensional convolutional network with a U-Net structure, which is used to capture the spectral factors of the target three-dimensional hyperspectral image data;

[0084] The deep low-rank decomposition algorithm is specifically as follows:

[0085] ,

[0086] in, represents the pixel value, represents the spatial representation coefficient, represents a spectral orthogonal basis, and represents the learning weight parameter, and denote spatial network and spectral network respectively, and represents random noise;

[0087] The spatial network includes an encoder-decoder network and a deep decoding network, wherein the encoder-decoder network has skip connections;

[0088] The encoder-decoder network is a two-dimensional convolutional network with a U-Net structure;

[0089] The deep decoding network consists of an upsampling refinement unit and a feature refinement unit;

[0090] The upsampling refinement unit is an attention gate structure, the original branch in the attention gate structure of the upsampling refinement unit includes an upsampling layer and a recursive normalization layer, and the attention weight branch in the attention gate structure of the upsampling refinement unit includes a two-dimensional convolution layer, a LeakyReLU activation function layer, and a Sigmoid activation function layer;

[0091] The feature refinement unit is an attention gate structure, the original branch in the feature refinement unit attention gate structure includes a multidimensional convolution layer and a two-dimensional convolution layer, and the attention weight branch in the feature refinement unit attention gate structure includes a two-dimensional convolution layer, a LeakyReLU activation function layer and a Sigmoid activation function layer;

[0092] The spatial network obtains multi-scale information of the guidance image according to the encoder-decoder network, and performs deep decoding feature weighting according to the channel attention gate to align spatial information and semantic information.

[0093] Step S04: obtaining reconstructed target three-dimensional hyperspectral image data according to prior information and low-dimensional subspace factors;

[0094] It should be noted that in this embodiment, the dual low-rank regularization model is optimized according to the ADMM algorithm to obtain an optimized model through cyclic iteration. The optimized model is specifically as follows:

[0095] ,

[0096] in, represents the overall optimization model, represents the pixel value, and represents the learning weight parameter, and represents auxiliary variables, represents the Lagrange multiplier, represents the 2D snapshot measurement data, represents the modulation and dispersion operation, and denote spatial network and spectral network respectively, and represents random noise, represents the regularization parameter, express The difference operator of represents the penalty parameter, represents the Frobenius norm;

[0097] The Lagrange multiplier is updated according to the dual ascent method. The specific algorithm for updating the Lagrange multiplier is as follows:

[0098] ,

[0099] in, represents the Lagrange multiplier.

[0100] In summary, according to the above-mentioned snapshot compression imaging reconstruction method based on dual low-rank regularization, by designing a dual low-rank regularization model, the three-dimensional hyperspectral image data is two-dimensionalized, the quality of the original data is guaranteed, and the problem of poor reconstruction effect caused by data loss or error is avoided. The dual low-rank regularization model combines the low-rank group sparse prior algorithm and the deep low-rank decomposition algorithm to form a dual low-rank regularization framework. The low-rank group sparse prior algorithm is used to extract the prior information of global spectral correlation and local spatial smoothness, respectively, and the texture detail features of the reconstruction results are enhanced. The deep low-rank decomposition algorithm is used to extract spatial factors and spectral factors, respectively, and the deep information is extracted quickly and effectively, further avoiding the problems of texture detail loss and spectral distortion in the reconstruction results. The present invention improves the reconstruction accuracy and authenticity of snapshot compression imaging. Specifically, the three-dimensional hyperspectral image data is two-dimensionalized through a dual low-rank regularization model to obtain two-dimensional snapshot measurement data. The two-dimensional processing is based on a coded aperture snapshot spectral imaging algorithm, which ensures the quality of the original data and avoids poor reconstruction effect due to data problems. The prior information of the two-dimensional snapshot measurement data is extracted according to a low-rank group sparse prior algorithm, which includes a group sparse regularization term. The prior information includes global spectral prior information and local spatial prior information, which enhances the overall texture detail features and further improves the accuracy of the reconstruction results. The deep low-rank decomposition algorithm extracts low-dimensional subspace factors of the two-dimensional snapshot measurement data. The deep low-rank decomposition algorithm is used to decompose the two-dimensional snapshot measurement data. The low-dimensional subspace factors include spatial factors and spectral factors. The low-dimensional subspace factors containing deep information are extracted quickly and effectively, further avoiding the problems of texture detail loss and spectral distortion in the reconstruction results. According to the prior information and the low-dimensional subspace factors, the reconstructed target three-dimensional hyperspectral image data is obtained, which further enhances the reliability of the model and the reconstruction quality. The present invention improves the reconstruction accuracy and authenticity of snapshot compression imaging.

[0101] See also Figure 2 , which is a flowchart of a snapshot compression imaging reconstruction method based on dual low-rank regularization proposed in the second embodiment of the present invention, the snapshot compression imaging reconstruction method based on dual low-rank regularization comprises steps S11 to S15, wherein:

[0102] Step S11: collecting three-dimensional hyperspectral image data and inputting it into a dual low-rank regularization model, converting the three-dimensional hyperspectral image data into two-dimensional data, and obtaining two-dimensional snapshot measurement data;

[0103] Step S12: extracting prior information of the two-dimensional snapshot measurement data according to a low-rank group sparse prior algorithm;

[0104] It should be noted that the low-rank group sparse prior algorithm in this embodiment includes a group sparse regularization term, which is the algorithm core of the low-rank group sparse prior algorithm and is used to extract prior information of two-dimensional snapshot measurement data.

[0105] Step S13: extracting low-dimensional subspace factors of the two-dimensional snapshot measurement data according to a deep low-rank decomposition algorithm;

[0106] Step S14: the spatial network captures the spatial factors of the target three-dimensional hyperspectral image data according to the unsupervised guided deep decoding network, the spectral network captures the spectral factors of the target three-dimensional hyperspectral image data, the spatial network obtains the multi-scale information of the guided image according to the encoder-decoder network, and performs deep decoding feature weighting according to the channel attention gate to align the spatial information and the semantic information;

[0107] Step S15: According to the prior information and the low-dimensional subspace factor, the reconstructed target three-dimensional hyperspectral image data is obtained, the double low-rank regularization model is optimized according to the ADMM algorithm, the optimized model is obtained by cyclic iteration, and the Lagrange multiplier is updated according to the dual ascent method;

[0108] It should be noted that in order to evaluate the effectiveness of the present invention, a comparative experiment was conducted using the KAIST data set in this embodiment to compare the present invention with the snapshot compression imaging reconstruction method in the prior art, including two model-based methods: GAP-TV and DeSCI, two model-driven deep learning methods: PnP-DIP and LRSDN, and five learning-based methods: λ-Net, TSA-Net, HDNet, MST++, and CST-L. The comparative results of the comparative experiment are shown in Table 1 below:

[0109] Table 1

[0110]

[0111] According to Table 1 above, the present invention is significantly superior to the snapshot compression imaging reconstruction method in the prior art in various evaluation indicators and experimental scenarios. From the average results, it can be seen that the present invention has significant progress compared with the prior art and has outstanding performance improvement. Compared with the self-supervised method LRSDN, the present invention improves the average peak signal-to-noise ratio and result similarity by 1.45 dB and 0.02 respectively. The method of the present invention does not require training data and can be flexibly applied to different data sets. Therefore, compared with the prior art, the present invention provides greater flexibility without affecting performance.

[0112] For a comprehensive evaluation of visual performance, see Figure 6 ,exist Figure 6The reconstruction results of scene 7 in the selected bands between compatible algorithms and ground truth are shown in the figure. GAP-TV and DeSCI rely heavily on hand-crafted priors to introduce artifacts and loss of texture details, and ignore the spectral low rank of three-dimensional hyperspectral images, resulting in color deviation. It can also be seen that LRSDN has lost a lot of textures. The present invention achieves excellent results in spatial spectral reconstruction by designing a low-rank decomposition network to capture deep low-rank priors and using group sparse priors to capture local spatial smoothness.

[0113] In summary, according to the above-mentioned snapshot compression imaging reconstruction method based on dual low-rank regularization, by designing a dual low-rank regularization model, the three-dimensional hyperspectral image data is two-dimensionalized, the quality of the original data is guaranteed, and the problem of poor reconstruction effect caused by data loss or error is avoided. The dual low-rank regularization model combines the low-rank group sparse prior algorithm and the deep low-rank decomposition algorithm to form a dual low-rank regularization framework. The low-rank group sparse prior algorithm is used to extract the prior information of global spectral correlation and local spatial smoothness, respectively, and the texture detail features of the reconstruction results are enhanced. The deep low-rank decomposition algorithm is used to extract spatial factors and spectral factors, respectively, and the deep information is extracted quickly and effectively, further avoiding the problems of texture detail loss and spectral distortion in the reconstruction results. The present invention improves the reconstruction accuracy and authenticity of snapshot compression imaging. Specifically, the three-dimensional hyperspectral image data is two-dimensionalized through a dual low-rank regularization model to obtain two-dimensional snapshot measurement data. The two-dimensional processing is based on a coded aperture snapshot spectral imaging algorithm, which ensures the quality of the original data and avoids poor reconstruction effect due to data problems. The prior information of the two-dimensional snapshot measurement data is extracted according to a low-rank group sparse prior algorithm, which includes a group sparse regularization term. The prior information includes global spectral prior information and local spatial prior information, which enhances the overall texture detail features and further improves the accuracy of the reconstruction results. The deep low-rank decomposition algorithm extracts low-dimensional subspace factors of the two-dimensional snapshot measurement data. The deep low-rank decomposition algorithm is used to decompose the two-dimensional snapshot measurement data. The low-dimensional subspace factors include spatial factors and spectral factors. The low-dimensional subspace factors containing deep information are extracted quickly and effectively, further avoiding the problems of texture detail loss and spectral distortion in the reconstruction results. According to the prior information and the low-dimensional subspace factors, the reconstructed target three-dimensional hyperspectral image data is obtained, which further enhances the reliability of the model and the reconstruction quality. The present invention improves the reconstruction accuracy and authenticity of snapshot compression imaging.

[0114] See also Figure 3 , which is a schematic diagram of the structure of a snapshot compression imaging reconstruction system based on dual low-rank regularization proposed in the third embodiment of the present invention, the system comprises:

[0115] A two-dimensionalization module 10 is used to perform two-dimensional processing on the three-dimensional hyperspectral image data through a dual low-rank regularization model to obtain two-dimensional snapshot measurement data, wherein the two-dimensional processing is based on a coded aperture snapshot spectral imaging algorithm;

[0116] A priori module 20, used for extracting the priori information of the two-dimensional snapshot measurement data according to a low-rank group sparse prior algorithm, wherein the low-rank group sparse prior algorithm includes a group sparse regularization term, and the prior information includes global spectral prior information and local spatial prior information;

[0117] A low-rank subspace decomposition module 30, used to extract low-dimensional subspace factors of the two-dimensional snapshot measurement data according to a deep low-rank decomposition algorithm, wherein the deep low-rank decomposition algorithm is used to decompose the two-dimensional snapshot measurement data, and the low-dimensional subspace factors include spatial factors and spectral factors;

[0118] The reconstruction module 40 is used to obtain reconstructed target three-dimensional hyperspectral image data according to the prior information and the low-dimensional subspace factor.

[0119] The present invention also proposes a computer storage medium on which one or more programs are stored, and when the programs are executed by a processor, the snapshot compression imaging reconstruction method based on double low-rank regularization is implemented.

[0120] The present invention also proposes a computer device, comprising a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the above-mentioned snapshot compression imaging reconstruction method based on dual low-rank regularization.

[0121] Those skilled in the art will appreciate that the logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be specifically implemented in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For purposes of this specification, "computer-readable medium" may be any device that can contain storage, communication, propagation or transmission of a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0122] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0123] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or a combination thereof: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0124] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0125] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A snapshot compression imaging reconstruction method based on dual low-rank regularization, characterized in that: include: The three-dimensional hyperspectral image data is processed into two dimensions through a dual low-rank regularization model to obtain two-dimensional snapshot measurement data, wherein the two-dimensional processing is based on a coded aperture snapshot spectral imaging algorithm; Extracting prior information of the two-dimensional snapshot measurement data according to a low-rank group sparse prior algorithm, wherein the low-rank group sparse prior algorithm includes a group sparse regularization term, and the prior information includes global spectral prior information and local spatial prior information; Extracting low-dimensional subspace factors of the two-dimensional snapshot measurement data according to a deep low-rank decomposition algorithm, wherein the deep low-rank decomposition algorithm is used to decompose the two-dimensional snapshot measurement data, and the low-dimensional subspace factors include spatial factors and spectral factors; Reconstructed target three-dimensional hyperspectral image data is obtained according to the prior information and the low-dimensional subspace factor.

2. The snapshot compression imaging reconstruction method based on dual low-rank regularization according to claim 1, characterized in that: The step of converting the three-dimensional hyperspectral image data into two-dimensional data by using a dual low-rank regularization model to obtain two-dimensional snapshot measurement data specifically includes: Collect three-dimensional hyperspectral image data and input it into a dual low-rank regularization model to convert the three-dimensional hyperspectral image data into two-dimensional data to obtain two-dimensional snapshot measurement data; The specific algorithm of the two-dimensional processing is as follows: ， in, represents the 2D snapshot measurement data, represents the modulation and dispersion operation, represents random noise, Represents pixel value.

3. The snapshot compression imaging reconstruction method based on dual low-rank regularization according to claim 1, characterized in that: The step of extracting the prior information of the two-dimensional snapshot measurement data according to the low-rank group sparse prior algorithm specifically includes: The prior information of the two-dimensional snapshot measurement data is extracted according to the low-rank group sparse prior algorithm, and the low-rank group sparse prior algorithm is specifically as follows: ， ， ， in, represents the 2D snapshot measurement data, represents the pixel value, represents the regularization parameter, express The differential operator of and Represent two differential operators in the spatial dimension, Represent the coordinates of the pixel in three directions of space, Indicates the length of the image data. Indicates the width of the image data, represents the Frobenius norm.

4. The snapshot compression imaging reconstruction method based on dual low-rank regularization according to claim 1, characterized in that: The step of extracting the low-dimensional subspace factors of the two-dimensional snapshot measurement data according to the deep low-rank decomposition algorithm specifically includes: Extracting low-dimensional subspace factors of the two-dimensional snapshot measurement data according to a deep low-rank decomposition algorithm, wherein the low-dimensional subspace factors include spatial factors and spectral factors; The deep low-rank decomposition algorithm includes a spatial network and a spectral network; The spatial network captures the spatial factors of the target three-dimensional hyperspectral image data according to the unsupervised guided deep decoding network; The spectral network is an unsupervised one-dimensional convolutional network with a U-Net structure, which is used to capture the spectral factors of the target three-dimensional hyperspectral image data; The deep low-rank decomposition algorithm is specifically as follows: ， in, represents the pixel value, represents the spatial representation coefficient, represents a spectral orthogonal basis, and represents the learning weight parameter, and denote spatial network and spectral network respectively, and represents random noise.

5. The snapshot compression imaging reconstruction method based on dual low-rank regularization according to claim 4, characterized in that: The spatial network specifically includes: The spatial network includes an encoder-decoder network and a deep decoding network, wherein the encoder-decoder network has skip connections; The encoder-decoder network is a two-dimensional convolutional network with a U-Net structure; The deep decoding network consists of an upsampling refinement unit and a feature refinement unit; The upsampling refinement unit is an attention gate structure, the original branch in the attention gate structure of the upsampling refinement unit includes an upsampling layer and a recursive normalization layer, and the attention weight branch in the attention gate structure of the upsampling refinement unit includes a two-dimensional convolution layer, a LeakyReLU activation function layer, and a Sigmoid activation function layer; The feature refinement unit is an attention gate structure, the original branch in the feature refinement unit attention gate structure includes a multidimensional convolution layer and a two-dimensional convolution layer, and the attention weight branch in the feature refinement unit attention gate structure includes a two-dimensional convolution layer, a LeakyReLU activation function layer and a Sigmoid activation function layer; The spatial network obtains multi-scale information of the guidance image according to the encoder-decoder network, and performs deep decoding feature weighting according to the channel attention gate to align spatial information and semantic information.

6. The snapshot compression imaging reconstruction method based on dual low-rank regularization according to claim 1, characterized in that: The step of obtaining the reconstructed target three-dimensional hyperspectral image data according to the prior information and the low-dimensional subspace factor further includes: The dual low-rank regularization model is optimized according to the ADMM algorithm to obtain an optimized model through cyclic iteration. The optimized model is specifically as follows: ， in, represents the overall optimization model, represents the pixel value, and represents the learning weight parameter, and represents auxiliary variables, represents the Lagrange multiplier, represents the 2D snapshot measurement data, represents the modulation and dispersion operation, and denote spatial network and spectral network respectively, and represents random noise, represents the regularization parameter, express The difference operator of represents the penalty parameter, represents the Frobenius norm.

7. The snapshot compression imaging reconstruction method based on dual low-rank regularization according to claim 6, characterized in that: The step of optimizing the dual low-rank regularization model according to the ADMM algorithm also includes: The Lagrange multiplier is updated according to the dual ascent method. The specific algorithm for updating the Lagrange multiplier is as follows: ， in, represents the Lagrange multiplier.

8. A snapshot compression imaging reconstruction system based on dual low-rank regularization, characterized in that: include: A two-dimensionalization module, used for performing two-dimensional processing on three-dimensional hyperspectral image data through a dual low-rank regularization model to obtain two-dimensional snapshot measurement data, wherein the two-dimensional processing is based on a coded aperture snapshot spectral imaging algorithm; A priori module, used for extracting prior information of the two-dimensional snapshot measurement data according to a low-rank group sparse prior algorithm, wherein the low-rank group sparse prior algorithm includes a group sparse regularization term, and the prior information includes global spectral prior information and local spatial prior information; A low-rank subspace decomposition module, used for extracting low-dimensional subspace factors of the two-dimensional snapshot measurement data according to a deep low-rank decomposition algorithm, wherein the deep low-rank decomposition algorithm is used for decomposing the two-dimensional snapshot measurement data, and the low-dimensional subspace factors include spatial factors and spectral factors; The reconstruction module is used to obtain reconstructed target three-dimensional hyperspectral image data according to the prior information and the low-dimensional subspace factor.

9. A storage medium, characterized in that: The storage medium stores one or more programs, which, when executed by a processor, implement the snapshot compression imaging reconstruction method based on dual low-rank regularization as described in any one of claims 1 to 7.

10. A computer device, characterized in that: The computer device comprises a memory and a processor, wherein: The memory is used to store computer programs; When the processor is used to execute the computer program stored in the memory, the snapshot compression imaging reconstruction method based on dual low-rank regularization as described in any one of claims 1 to 7 is implemented.

Images