Self-adaptive compressed spectrum ground remote sensing imaging method, system and device and storage medium

Abstract

The invention discloses a self-adaptive compressed spectrum ground remote sensing imaging method, system and device and a storage medium, belongs to the technical field of aerospace ground optical remote sensing, and solves the technical problem that the spectral resolution and the spatial resolution are difficult to synchronously improve in a traditional optical remote sensing method. Calculating surface feature reflectivity data; constructing a hyperspectral training set based on the ground object reflectivity data and a quantum efficiency curve of an image sensor of the compressed spectral imager; based on a hyperspectral training set deep learning result, obtaining an optimal response matrix of the optical filter, and aiming at different ground feature categories, respectively solving optimal optical filter response matrixes corresponding to the different ground feature categories; optical filter transmittance matrixes of the optimal optical filter response matrixes corresponding to different ground feature categories are obtained respectively, the optical filter transmittance matrixes are converted into voltage control matrixes of the K0-order liquid crystal adjustable optical filters, in-orbit configuration of the multi-spectral response curves is achieved, and the optimal compression spectral imaging performance is achieved.

Description

Technical Field

[0001] The present invention relates to the field of aerospace-to-earth optical remote sensing technology, and in particular to an adaptive compressed spectrum earth remote sensing imaging method, system, equipment and storage medium. Background Art

[0002] Spatial resolution and spectral resolution are two core technical specifications for optical remote sensing satellites. These typically use narrowband filters or grating spectrometry to separate the spectrum of ground objects into multiple independent spectral channels, capturing images for each. The greater the number of spectral channels, the narrower the spectral width and the weaker the energy. Simultaneously improving both spectral and spatial resolution requires a significant increase in optical aperture, which significantly increases the cost and makes hyperspectral imaging with a resolution better than 5 meters difficult to achieve. In recent years, the following public literature has reported on high-resolution hyperspectral imaging technology.

[0003] Chinese patent CN110211044A discloses a "Multispectral Imaging Method and Apparatus Based on Demosaicing and Principal Component Analysis." The method integrates a wide-band multispectral filter array (MSFA) with a preset number of channels in front of a detector to acquire data, where the acquired data is a mosaic image corresponding to the wide-band multispectral filter array (MSFA). The acquired data is processed using a demosaicing algorithm to reconstruct a complete spatial domain full-resolution image corresponding to the preset number of channels. Based on principal component analysis theory and utilizing the spectral modulation characteristics of the MSFA and the spectral modulation data of each spatial point, the spectrum of each spatial point is reconstructed and decoupled in parallel to obtain a multispectral image. This method leverages the excellent reconstruction results of the demosaicing algorithm when the number of channels is small and the dimensionality reduction effect of principal component analysis to achieve multispectral imaging. Chinese patent CN113447118A discloses a "multispectral imaging chip and color imaging method capable of color imaging." The chip comprises a multispectral chip body and a color restoration module. The multispectral chip body includes a spectral modulation module and a spectral inversion module. The spectral modulation module achieves spectral separation using a filter film formed from N materials with known and varying transmittances. A photoelectric conversion substrate beneath the filter film converts the separated optical signal into an electrical signal, which is then amplified and converted into a digital signal or encoded output after analog-to-digital conversion. The spectral inversion module inverts the incident spectrum based on the intensity of the optical signal output by the spectral modulation module and the corresponding pixel position information. The color restoration module converts the spectral information output by the multispectral chip body into an RGB image by multiplying the spectral information by a spectrum-to-color transformation matrix. This chip converts the spectral information into an RGB image, leveraging the large amount of information contained in spectral images to achieve high-fidelity color imaging. The above two Chinese patents are applicable to hyperspectral reconstruction of area array image sensors. However, in order to improve the system exposure capability, high-resolution aerospace optical cameras all adopt the time delay integration method and the TDI push-scan method to achieve image acquisition. The mosaic method cannot be used to achieve spectral band deployment. In addition, there are many types of ground objects in ground remote sensing, and its generalization and adaptability face severe challenges.

[0004] Chinese patent CN113447118A discloses a "Wide-spectrum Modulation and Demodulation Imaging Spectral Chip and Production Method," addressing practical application challenges of existing imaging spectral chip technology: low energy efficiency, low spectral modulation, reduced spatial resolution, narrow field of view, complex manufacturing process, high process precision requirements, and inability to mass produce. The chip comprises, from top to bottom, a wide-spectrum modulation structure layer, including a wide-spectrum modulation structural unit, which, in an xz plane view, has two C-shaped openings arranged in mirror-image orientation to modulate incident light information; a photoelectric image sensor module, which collects the modulated incident light information and converts it into a digital electrical signal; and a spectral image demodulation module, which demodulates the digital electrical signal to generate target multispectral image information." This technology uses a dual-C-shaped wide-spectrum modulation structure to achieve spectral modulation, requiring semiconductor process support and high process complexity. Furthermore, the structure may be affected by the image sensor's temperature, posing risks to spectral transmittance stability and making it unsuitable for aerospace applications.

[0005] Chinese patent CN114779467A discloses a "method for selecting novel spectrometer film combinations based on detector characteristics," comprising the following steps: Step 1: Input an expression for filters containing detector characteristics and calculate discrete values; Step 2: Determine the number of clusters to be clustered, setting the number based on the spectral characteristics of the object; Step 3: Calculate an evaluation function and select the cluster with the smallest evaluation function; Step 4: Calculate new cluster centers and compare them with the last clustering results; Step 5: Optimize using the condition number Cond(A) as the objective function to find the optimal condition number. Using the kmean-particle swarm algorithm to select 100 film combinations containing detector characteristics, the results show that when the number of clusters is 5 or 6, the film combination with the smallest condition number can be optimized. Interpolation fitting of the 11 discrete values obtained within the wavelength range of 400-900 nm yields an average error percentage of 0.06, indicating good spectral reconstruction results. Chinese patent CN113188658A discloses a "tunable spectral reconstruction method." To reduce the amount of data, this spectral reconstruction method uses 10 film systems for spectral reconstruction. The specific reconstruction process is as follows: design the transmittance curve of the film system, and perform processing, retesting, and fitting; select a detector model suitable for the experiment; calibrate the monochromatic light source to obtain a standard spectral curve with a wavelength range of 400-900nm; obtain grayscale images of the 10 film systems and perform spectral reconstruction. The spectral reconstruction method provides a spectral reconstruction accuracy (ARE) of 0.0523. Compared with domestic and foreign research results, the reconstruction accuracy reaches the same level while using a smaller number of film systems, achieving data dimensionality reduction and facilitating engineering. The film system designs of the above two Chinese patents rely on the characteristics of the spectral data training set itself, which can achieve good spectral reconstruction results for a certain type of ground object. However, the variety of ground objects in ground remote sensing is large, and their generalization and adaptability face severe challenges.

[0006] In summary, in order to achieve wide-band spectral modulation, area array cameras must realize the processing of pixel-level spectral modulation units, which greatly limits its implementation process. The main methods used are colored material filling and photonic metasurfaces. However, aerospace optical remote sensing adopts push-broom imaging. The image of the ground object passes through different spectral bands in turn to obtain multiple wide-band image components for reconstructing hyperspectral data. This makes it possible to achieve spectral modulation through multiple wide-band strip filters. The implementation process can fully inherit the mature optical coating process, and its stability and spatial adaptability are guaranteed. Summary of the Invention

[0007] The present invention solves the technical problem that it is difficult to simultaneously improve the spectral resolution and spatial resolution in traditional optical remote sensing methods.

[0008] The present invention provides an adaptive compressed spectral earth remote sensing imaging method, comprising the following steps:

[0009] Step S1: Based on the hyperspectral data set, hyperspectral data subsets of different ground object categories are constructed respectively, and the spectral response function, atmospheric transmittance function and top-of-atmosphere solar spectrum data of the instrument are obtained according to the hyperspectral data subsets, and the ground object reflectance data is calculated;

[0010] Step S2, constructing a hyperspectral training set based on ground object reflectance data and a quantum efficiency curve of the compressed spectral imager image sensor;

[0011] Step S3: Based on the deep learning results of the hyperspectral training set, a compressed spectral band response function is constructed. After obtaining the optimal solution of the underdetermined equations based on the compressed spectral band response function, the optimal response matrix of the filter is obtained. The optimal filter response matrix corresponding to different ground object categories is obtained respectively.

[0012] Step S4 , obtaining the filter transmittance matrix of the optimal filter response matrix corresponding to different ground object categories respectively, and converting it into a voltage control matrix of the K0-order liquid crystal tunable filter.

[0013] Furthermore, in one embodiment of the present invention, in step S1, the calculating of the ground object reflectivity data is specifically as follows:

[0014]

[0015] Among them, R(i,j,λ) is the reflectivity data of the ground object, h is the Planck constant, and its value is 6.63×10 -34 J·s, c is the speed of light in vacuum, and its value is 3×10 8 m / s, S0(i,j,λ) is the DN value of the original data of the hyperspectral dataset, λ is the wavelength of light, q0(λ) is the quantum efficiency of the hyperspectral data acquisition instrument, I(i,j,λ) is the light energy received by the optical camera detector during the exposure time of the spectral segment with the center wavelength of λ, without considering the influence of the atmosphere, and τ(i,j,λ) is the atmospheric transmittance of the optical camera detector pixel in the i-th row and j-th column, with the center wavelength of λ.

[0016] Furthermore, in one embodiment of the present invention, in step S2, the construction of the hyperspectral training set is specifically as follows:

[0017]

[0018] Among them, S1(i,j,λ) is the hyperspectral training set, q0(λ) is the quantum efficiency of the hyperspectral data acquisition instrument, S0(i,j,λ) is the DN value of the original data of the hyperspectral dataset, λ is the wavelength of light, and q1(λ) is the quantum efficiency of the image sensor for compressed spectral imaging.

[0019] Furthermore, in one embodiment of the present invention, in step S3, the compressed spectral band response function is constructed, and after obtaining the optimal solution of the underdetermined equations based on the compressed spectral band response function, the optimal response matrix of the filter is obtained, specifically:

[0020] A compressed spectral band response function is constructed, and compressed spectral imaging data is generated according to the imaging model. After obtaining the optimal solution of the underdetermined equations, hyperspectral reconstruction is performed. After multiple iterations, the optimal spectral band response matrix is obtained. After eliminating the influence of the sensor quantum efficiency, the optimal filter response matrix is obtained.

[0021] Furthermore, in one embodiment of the present invention, the underdetermined system of equations is specifically:

[0022] S2′(i,j,n)=Α -1 (i,j,n,λ)S2(i,j,n);

[0023] Among them, S2′(i,j,n) is an underdetermined equation system, S2 is the compressed spectral imaging data, λ is the wavelength of light, and Α -1 (i, j, n, λ) is the pseudo-inverse matrix of the compressed spectrum band response function.

[0024] Furthermore, in one embodiment of the present invention, in step S4, the filter transmittance matrix of the optimal filter response matrix corresponding to different ground object categories is converted into a voltage control matrix of the K0-order liquid crystal tunable filter, specifically:

[0025]

[0026] Among them, B m (n,λ) is the voltage control matrix of the tunable filter, k is the order number of the liquid crystal tunable filter, f m (n,k) is the voltage control matrix of the K0-order liquid crystal tunable filter, ρ k is the adjustment coefficient of the k-th order liquid crystal tunable filter.

[0027] The adaptive compressed spectral earth remote sensing imaging system described in the present invention includes the following modules:

[0028] Module S1, based on the hyperspectral data set, constructs hyperspectral data subsets for different ground object categories, obtains the instrument's spectral response function, atmospheric transmittance function, and top-of-atmosphere solar spectrum data based on the hyperspectral data subsets, and calculates the ground object reflectance data;

[0029] Module S2 constructs a hyperspectral training set based on ground reflectance data and the quantum efficiency curve of the compressed spectral imager image sensor;

[0030] Module S3 constructs a compressed spectral band response function based on the deep learning results of the hyperspectral training set. After finding the optimal solution of the underdetermined equations based on the compressed spectral band response function, the optimal filter response matrix is obtained. The optimal filter response matrix corresponding to different ground object categories is also obtained.

[0031] Module S4 obtains the filter transmittance matrix of the optimal filter response matrix corresponding to different ground object categories and converts it into a voltage control matrix of the K0-order liquid crystal tunable filter.

[0032] An electronic device according to the present invention is characterized in that it comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0033] Memory for storing computer programs;

[0034] The processor is used to implement the method steps described above when executing the program stored in the memory.

[0035] The computer-readable storage medium of the present invention stores a computer program, which implements the method steps described above when executed by a processor.

[0036] This invention solves the technical problem of the difficulty in simultaneously improving spectral resolution and spatial resolution in traditional optical remote sensing methods. Specific beneficial effects include:

[0037] The adaptive compressed spectral earth remote sensing imaging method described in the present invention uses a liquid crystal electrically controlled variable retardation plate as the main component to form an adjustable filter. To improve the accuracy of spectral reconstruction, group training is performed for different ground object types to obtain the optimal spectral band combination for each ground object type. Through on-orbit AI ground object type recognition and dynamic configuration of filter transmittance curves, multi-spectral band data is collected to reconstruct hyperspectral, high-resolution remote sensing images.

[0038] The invention discloses an adaptive compressed spectral earth remote sensing imaging method, which is used for high spatial resolution and high spectral resolution optical remote sensing images. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0040] Figure 1 This is the overall flow chart of the adaptive compressed spectral imaging described in the first embodiment;

[0041] Figure 2This is a schematic diagram of the spectrum band configuration described in the first embodiment;

[0042] Figure 3 This is a schematic diagram of a single spectrum segment of hyperspectral data according to the first embodiment;

[0043] Figure 4 is a graph of the filter response matrix response curve (single channel) described in embodiment 1;

[0044] Figure 5 This is a graph of the test results (single pixel) of compressed spectrum reconstruction of different ground objects described in Implementation Method 1;

[0045] Figure 6 This is a diagram of the test results (single spectrum segment) of compressed spectrum reconstruction of different ground objects described in the first embodiment. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe various embodiments of the present invention in conjunction with the accompanying drawings. The embodiments described with reference to the accompanying drawings are exemplary and intended to be used to explain the present invention, but should not be understood as limiting the present invention.

[0047] Embodiment 1: The adaptive compressed spectral earth remote sensing imaging method described in this embodiment includes the following steps:

[0048] Step S1: Based on the hyperspectral data set, hyperspectral data subsets of different ground object categories are constructed respectively, and the spectral response function, atmospheric transmittance function and top-of-atmosphere solar spectrum data of the instrument are obtained according to the hyperspectral data subsets, and the ground object reflectance data is calculated;

[0049] Step S2, constructing a hyperspectral training set based on ground object reflectance data and a quantum efficiency curve of the compressed spectral imager image sensor;

[0050] Step S3: Based on the deep learning results of the hyperspectral training set, a compressed spectral band response function is constructed. After obtaining the optimal solution of the underdetermined equations based on the compressed spectral band response function, the optimal response matrix of the filter is obtained. The optimal filter response matrix corresponding to different ground object categories is obtained respectively.

[0051] Step S4 , obtaining the filter transmittance matrix of the optimal filter response matrix corresponding to different ground object categories respectively, and converting it into a voltage control matrix of the K0-order liquid crystal tunable filter.

[0052] In this embodiment, in step S1, the ground object reflectivity data is calculated as follows:

[0053]

[0054] Among them, R(i,j,λ) is the reflectivity data of the ground object, h is the Planck constant, and its value is 6.63×10 -34 J·s, c is the speed of light in vacuum, and its value is 3×10 8 m / s, S0(i,j,λ) is the DN value of the original data of the hyperspectral dataset, λ is the wavelength of light, q0(λ) is the quantum efficiency of the hyperspectral data acquisition instrument, I(i,j,λ) is the light energy received by the optical camera detector during the exposure time of the spectral segment with the center wavelength of λ, without considering the influence of the atmosphere, and τ(i,j,λ) is the atmospheric transmittance of the optical camera detector pixel in the i-th row and j-th column, with the center wavelength of λ.

[0055] In this embodiment, in step S2, the construction of the hyperspectral training set is specifically as follows:

[0056]

[0057] Among them, S1(i,j,λ) is the hyperspectral training set, q0(λ) is the quantum efficiency of the hyperspectral data acquisition instrument, S0(i,j,λ) is the DN value of the original data of the hyperspectral dataset, λ is the wavelength of light, and q1(λ) is the quantum efficiency of the image sensor for compressed spectral imaging.

[0058] In this embodiment, in step S3, the compressed spectral band response function is constructed, and the optimal solution of the underdetermined equations is obtained based on the compressed spectral band response function to obtain the optimal response matrix of the filter, which is specifically:

[0059] A compressed spectral band response function is constructed, and compressed spectral imaging data is generated according to the imaging model. After obtaining the optimal solution of the underdetermined equations, hyperspectral reconstruction is performed. After multiple iterations, the optimal spectral band response matrix is obtained. After eliminating the influence of the sensor quantum efficiency, the optimal filter response matrix is obtained.

[0060] In this embodiment, the underdetermined equations are specifically:

[0061] S2′(i,j,n)=Α -1 (i,j,n,λ)S2(i,j,n);

[0062] Among them, S2′(i,j,n) is an underdetermined equation system, S2 is the compressed spectral imaging data, λ is the wavelength of light, and Α -1 (i, j, n, λ) is the pseudo-inverse matrix of the compressed spectrum band response function.

[0063] In this embodiment, in step S4, the filter transmittance matrix of the optimal filter response matrix corresponding to different ground object categories is converted into a voltage control matrix of the K0-order liquid crystal tunable filter, specifically:

[0064]

[0065] Among them, B m (n,λ) is the voltage control matrix of the tunable filter, k is the order number of the liquid crystal tunable filter, f m (n,k) is the voltage control matrix of the K0-order liquid crystal tunable filter, ρ k is the adjustment coefficient of the k-th order liquid crystal tunable filter.

[0066] like Figure 1 As shown, this embodiment proposes an adaptive compressed spectral earth remote sensing imaging method, including the following steps:

[0067] In step S1, the existing hyperspectral data set is used to construct hyperspectral data subsets of different ground object categories. The spectral response function of the instrument, the atmospheric transmittance function and the solar spectrum data of the top of the atmosphere are obtained based on the hyperspectral data to calculate the ground object reflectance data.

[0068]

[0069] Where S0(i, j, λ) is the dimensionless DN value of the original data of the hyperspectral dataset, and h is the Planck constant, which is 6.63×10 -34 J·s, c is the speed of light in vacuum, and its value is 3×10 8 m / s, λ is the wavelength of the light wave, with the dimension of m, q0(λ) is the quantum efficiency of the hyperspectral data acquisition instrument, which is dimensionless, I(i,j,λ) is the light energy received during the exposure time of the optical camera detector pixel in the i-th row and j-th column, with the central wavelength of λ, without considering the influence of the atmosphere, with the dimension of J, τ(i,j,λ) is the atmospheric transmittance of the optical camera detector pixel in the i-th row and j-th column, with the central wavelength of λ, which is dimensionless, R(i,j,λ) is the ground reflectance data, which is dimensionless.

[0070] Step S2, such as Figure 2 As shown in Figure 2, using formula (1) and the quantum efficiency curve of the compressed spectral imager image sensor, formula (2) is derived to construct a hyperspectral training set.

[0071]

[0072] Where S1(i, j, λ) is the hyperspectral training set, and q1(λ) is the quantum efficiency of the image sensor for compressed spectral imaging.

[0073] Step S3, using the deep learning results of the spectral training subset, under the constraints of continuity and non-negativity, construct a compressed spectral band response function A, and generate compressed spectral imaging data S2 based on the imaging model. Using the artificial intelligence algorithm to obtain the optimal solution of the underdetermined equations shown in formula (4), perform hyperspectral reconstruction, and after multiple iterations, obtain the optimal spectral band response matrix. After eliminating the influence of the sensor quantum efficiency, the optimal filter response matrix is obtained. For m different types of land features such as deserts, grasslands, farmlands, and cities, the optimal filter response matrices B1~B m .

[0074] Α(i,j,n,λ)S1(i,j,λ)=S2(i,j,n); (3)

[0075] S2′(i,j,n)=Α -1 (i,j,n,λ)S2(i,j,n);(4)

[0076]

[0077] Step S4, B1~B m The filter transmittance matrix is converted into the voltage control matrix f of the K0-order liquid crystal tunable filter according to the formula m (n,k), stored in the onboard AI system.

[0078]

[0079] Where, ρ k is the adjustment coefficient of the k-th order liquid crystal tunable filter.

[0080] In order to better illustrate the adaptive compressed spectral earth remote sensing imaging method described in this embodiment, the following examples are described in detail:

[0081] Step S1, original data set construction:

[0082] Hyperspectral image data are collected and organized, and divided into six subsets according to the types of land features, namely forest, grassland, farmland, city, ocean, and desert. Hyperspectral image data from different remote sensors are standardized to expand the size of the available data set. Interspectral interpolation is used to achieve spectral segment unification. According to formula (2), the quantum efficiency curve of the target image sensor Gmax3265 is used to perform standardization conversion to eliminate the influence of quantum efficiency differences. Then, the standardized hyperspectral data cube is used to construct a training data set in the range of 380nm to 2500nm, as shown in the following figure: Figure 3 shown.

[0083] Step S2, solve the filter response matrix:

[0084] After obtaining the original hyperspectral data of different objects, they are split into training set, test set and cross-validation set in a ratio of 6:2:2 for training. During the training process, randomly generated Gaussian matrix, orthogonal matrix, sparse matrix and covariance matrix and regularization matrix obtained by principal component decomposition are used as the initial spectral response matrix A. According to formula (3) and the original hyperspectral data S1, the compressed spectral data S2 is calculated. The pseudo-inverse A of the spectral response matrix is solved. -1 , calculate the reconstructed hyperspectral data S2′ according to formula (4); use the mean square error loss of the original hyperspectral data S1 and the reconstructed hyperspectral data S2′ to construct the objective function, judge the error loss caused by A, guide the model to perform multiple iterative optimizations, and obtain the optimal spectral response matrix; according to formula (5), substitute the quantum efficiency curve of Gmax3265 for matrix transformation to obtain the optimal filter response matrix under different ground objects. For different ground objects, the transmittance curve of the filter response matrix is as follows Figure 4 shown.

[0085] Step S3, collection and reconstruction of compressed spectrum:

[0086] After obtaining the response matrix of the filter under different ground objects, the response matrix is converted into the voltage control matrix of the liquid crystal tunable filter according to formula (6), and the voltage control matrix is saved to the onboard AI system; during the imaging process, the onboard AI is first used to identify the main type of the ground object, and the ground object tag number is sent back to the main control. The main control loads the corresponding voltage control matrix according to the ground object index number, controls the filter to perform spectral imaging according to the modulated transmittance curve, and obtains compressed spectral data; after the imaging is completed, the ground object tag number and the compressed spectral data are packaged and sent back to the data center in a digital transmission mode. The compressed spectral data is demodulated according to formula (4) to obtain the reconstructed ground object hyperspectral data. The reconstruction result of the compressed hyperspectral data is as follows: Figure 5-Figure 6 shown.

[0087] Embodiment 2: The adaptive compressed spectral earth remote sensing imaging system described in this embodiment includes the following modules:

[0088] Module S1, based on the hyperspectral data set, constructs hyperspectral data subsets for different ground object categories, obtains the instrument's spectral response function, atmospheric transmittance function, and top-of-atmosphere solar spectrum data based on the hyperspectral data subsets, and calculates the ground object reflectance data;

[0089] Module S2 constructs a hyperspectral training set based on ground reflectance data and the quantum efficiency curve of the compressed spectral imager image sensor;

[0090] Module S3 constructs a compressed spectral band response function based on the deep learning results of the hyperspectral training set. After finding the optimal solution of the underdetermined equations based on the compressed spectral band response function, the optimal filter response matrix is obtained. The optimal filter response matrix corresponding to different ground object categories is also obtained.

[0091] Module S4 obtains the filter transmittance matrix of the optimal filter response matrix corresponding to different ground object categories and converts it into a voltage control matrix of the K0-order liquid crystal tunable filter.

[0092] Embodiment 3: An electronic device described in this embodiment includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus;

[0093] Memory for storing computer programs;

[0094] The processor is configured to implement the method steps described in the first embodiment when executing the program stored in the memory.

[0095] Implementation method 4: This implementation method describes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps described in implementation method 1 are implemented.

[0096] The above is a detailed introduction to the adaptive compressed spectral earth remote sensing imaging method, system, device and storage medium proposed in the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. An adaptive compressed spectral earth remote sensing imaging method, characterized in that: The following steps are involved: Step S1: Based on the hyperspectral data set, hyperspectral data subsets of different ground object categories are constructed respectively, and the spectral response function, atmospheric transmittance function and top-of-atmosphere solar spectrum data of the instrument are obtained according to the hyperspectral data subsets, and the ground object reflectance data is calculated; Step S2, constructing a hyperspectral training set based on ground object reflectance data and a quantum efficiency curve of the compressed spectral imager image sensor; Step S3: Based on the deep learning results of the hyperspectral training set, a compressed spectral band response function is constructed. After obtaining the optimal solution of the underdetermined equations based on the compressed spectral band response function, the optimal response matrix of the filter is obtained. The optimal filter response matrix corresponding to different ground object categories is obtained respectively. Step S4 , obtaining the filter transmittance matrix of the optimal filter response matrix corresponding to different ground object categories respectively, and converting it into a voltage control matrix of the K0-order liquid crystal tunable filter.

2. The adaptive compressed spectral earth remote sensing imaging method according to claim 1, characterized in that: In the step S1, the ground object reflectivity data is calculated as follows: Among them, R(i,j,λ) is the reflectivity data of the ground object, h is the Planck constant, and its value is 6.63×10 -34 J·s, c is the speed of light in vacuum, and its value is 3×10 8 m / s, S0(i,j,λ) is the DN value of the original data of the hyperspectral dataset, λ is the wavelength of light, q0(λ) is the quantum efficiency of the hyperspectral data acquisition instrument, I(i,j,λ) is the light energy received by the optical camera detector during the exposure time of the spectral segment with the center wavelength of λ, without considering the influence of the atmosphere, and τ(i,j,λ) is the atmospheric transmittance of the optical camera detector pixel in the i-th row and j-th column, with the center wavelength of λ.

3. The adaptive compressed spectral earth remote sensing imaging method according to claim 1, characterized in that: In step S2, the construction of the hyperspectral training set is specifically as follows: Among them, S1(i,j,λ) is the hyperspectral training set, q0(λ) is the quantum efficiency of the hyperspectral data acquisition instrument, S0(i,j,λ) is the DN value of the original data of the hyperspectral dataset, λ is the wavelength of light, and q1(λ) is the quantum efficiency of the image sensor for compressed spectral imaging.

4. The adaptive compressed spectrum earth remote sensing imaging method according to claim 1, characterized in that: In step S3, the compressed spectral band response function is constructed, and the optimal solution of the underdetermined equations is obtained based on the compressed spectral band response function to obtain the optimal response matrix of the filter, which is specifically: A compressed spectral band response function is constructed, and compressed spectral imaging data is generated according to the imaging model. After obtaining the optimal solution of the underdetermined equations, hyperspectral reconstruction is performed. After multiple iterations, the optimal spectral band response matrix is obtained. After eliminating the influence of the sensor quantum efficiency, the optimal filter response matrix is obtained.

5. The adaptive compressed spectrum earth remote sensing imaging method according to claim 1 or 4, characterized in that: The underdetermined system of equations is specifically: S2′(i,j,n)=Α -1 (i,j,n,λ)S2(i,j,n); Among them, S2′(i,j,n) is an underdetermined equation system, S2 is the compressed spectral imaging data, λ is the wavelength of light, and Α -1 (i, j, n, λ) is the pseudo-inverse matrix of the compressed spectral band response function.

6. The adaptive compressed spectrum earth remote sensing imaging method according to claim 1, characterized in that: In step S4, the filter transmittance matrix of the optimal filter response matrix corresponding to different ground object categories is obtained and converted into the voltage control matrix of the K0-order liquid crystal tunable filter, specifically: Among them, B m (n,λ) is the voltage control matrix of the tunable filter, k is the order number of the liquid crystal tunable filter, f m (n,k) is the voltage control matrix of the K0-order liquid crystal tunable filter, ρ k is the adjustment coefficient of the k-th order liquid crystal tunable filter.

7. An adaptive compressed spectrum earth remote sensing imaging system, characterized in that: Includes the following modules: Module S1, based on the hyperspectral data set, constructs hyperspectral data subsets for different ground object categories, obtains the instrument's spectral response function, atmospheric transmittance function, and top-of-atmosphere solar spectrum data based on the hyperspectral data subsets, and calculates the ground object reflectance data; Module S2 constructs a hyperspectral training set based on ground reflectance data and the quantum efficiency curve of the compressed spectral imager image sensor; Module S3 constructs a compressed spectral band response function based on the deep learning results of the hyperspectral training set. After finding the optimal solution of the underdetermined equations based on the compressed spectral band response function, the optimal filter response matrix is obtained. The optimal filter response matrix corresponding to different ground object categories is also obtained. Module S4 obtains the filter transmittance matrix of the optimal filter response matrix corresponding to different ground object categories and converts it into a voltage control matrix of the K0-order liquid crystal tunable filter.

8. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory for storing computer programs; A processor, configured to implement the method steps described in any one of claims 1 to 6 when executing a program stored in a memory.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps of any one of claims 1 to 6 are implemented.

Images