Satellite hyperspectral channel selection method and device

Abstract

The present invention provides a satellite hyperspectral channel selection method and device, the method comprising: calculating the cost function descent rate corresponding to each satellite hyperspectral channel, and taking the channel with the largest descent rate as the first selected channel; screening the alternative channels through the angle range of the Jacobian vectors of the unselected channels and the selected channels after linear transformation, so that the atmospheric information contained in them is significantly different from that of the selected channels; combining each alternative channel with the selected n-1 channels, and calculating the cost function descent rate thereof by a recursive method, and taking the channel with the largest descent rate as the nth selected channel; calculating the weight inverse matrix after the combination of the selected n channels by a recursive algorithm, and selecting the next channel, and repeating the above process until n is equal to a preset value. The present invention uses the maximum cost function descent rate as the channel selection criterion to ensure the reliability and accuracy of the variational method assimilation, and O(n 2 )'s computational complexity and parallel functionality make the present invention have higher computational efficiency.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and device for selecting a satellite hyperspectral channel. Background Art

[0002] With the rapid development of hyperspectral remote sensing technology, modern meteorological satellites are capable of acquiring observational data from thousands of spectral channels. For example, the Infrared Atmospheric Sounding Interferometer (IASI) instrument has 8,461 channels, and the GIIRS instrument has 1,690 channels. While this data provides a wealth of information for accurate inversion of atmospheric parameters, its sheer volume and complex correlations pose significant challenges to storage, transmission, and computation. In this context, it is crucial to eliminate redundant data and improve satellite data processing efficiency. Therefore, channel selection methods are a core issue in hyperspectral remote sensing applications.

[0003] The sensitivity of different spectral channels to atmospheric parameters varies significantly at different altitudes. Indiscriminately using all channels can lead to uneven distribution of spectral sensitivity or signal overlap. Furthermore, the Jacobians of adjacent channels can be highly correlated due to spectral overlap. Directly using these inversion methods can result in ill-conditioned matrices and difficulty in inverting them. Therefore, channel selection methods not only improve the accuracy of extracting effective information from the data but also play a crucial role in ensuring the numerical stability of the assimilation system.

[0004] The core of channel selection lies in defining a reasonable information quantification metric. The atmospheric retrievability index and entropy subtraction method comprehensively consider background field errors and observation errors and are widely used in current operations. Both methods essentially use the reduction ratio of the determinant of the error covariance matrix of the variational solution as an information quantification metric. Compared with the Jacobian matrix channel selection method, although these two algorithms are slightly less efficient, they can more accurately eliminate redundancy by dynamically updating the covariance matrix, significantly improving the DFS (Degrees of Freedom for Signal) value. When computing resources permit, the atmospheric retrievability index and entropy subtraction method are more suitable for scenarios requiring high precision.

[0005] Despite this, the flaws of the atmospheric retrievability index and entropy subtraction methods are also quite obvious. First, by using only the atmospheric background error covariance matrix and the Jacobian matrix of the satellite channels based on the atmospheric background field, the selected channels are likely unreliable when the background field error is large. Second, among all channel combinations, the atmospheric retrievability index and entropy subtraction methods yield the channel combination with the highest probability of minimizing the background error variance of the variational decomposition. However, this does not necessarily mean that the probability of this combination being the optimal combination is high. For example, when this probability is less than 0.5, it indicates that most of the optimal combinations are scattered among other channel combinations. The channel combinations obtained by these two methods are likely to prevent a significant decrease in the cost function, resulting in an inability to fully obtain satellite observation information. Summary of the Invention

[0006] This invention provides a method and device for selecting satellite hyperspectral channels, addressing the drawbacks of existing technologies, such as unreliable and suboptimal channel combinations. Targeting the variational assimilation of satellite data, this invention directly estimates each channel's contribution to the cost function and selects the satellite channel combination that maximizes the reduction in the cost function. This method considers real-time satellite observation information, enabling specific analysis of specific problems and dynamically selecting channels based on observation data. It is suitable for satellite channel selection in various weather conditions.

[0007] The present invention provides a satellite hyperspectral channel selection method, comprising:

[0008] Calculate the drop rate of the value function corresponding to each channel in the satellite hyperspectral channel set, and select the channel with the largest drop rate as the first selected channel;

[0009] A vector is obtained by linearly transforming the Jacobian vectors of the unselected channels and the selected channels in the satellite hyperspectral channel set using a background error covariance matrix, and an angle between the vectors of the unselected channels and the selected channels is calculated. A subset of candidate channels is selected from the unselected channels by setting a range of angles, so that atmospheric vertical structure information observed by the channels in the subset of candidate channels is significantly different from that of the selected channels;

[0010] Combine each channel in the candidate channel subset with n-1 selected channels, where n≥2, and calculate the decrease rate of the value function of each combination using a recursive algorithm based on the (n-1)×(n-1)-dimensional weight inverse matrix of the selected channels. The channel corresponding to the combination with the largest decrease rate is taken as the nth selected channel.

[0011] After obtaining the nth selected channel, a recursive algorithm is used to calculate the n×n weight inverse matrix of the n selected channels to continue selecting the next channel.

[0012] According to a satellite hyperspectral channel selection method provided by the present invention, the satellite observation channel set is determined by the following formula: The rate of decrease of the value function corresponding to each of the M channels:

[0013]

[0014]

[0015]

[0016]

[0017]

[0018]

[0019]

[0020] in, Satellite observation channel sets The radiation observation value, radiation simulation value, radiation simulation Jacobian vector and radiation value error variance of the j-th channel; ,in represents the analytical solution of the atmospheric profile when using 1 channel, is its background vector, and m is the number of vertical layers of the atmospheric profile. The error covariance matrix is B; Represents the diagonal matrix of the error variance of the observed radiation values in n channels; Indicates the rate of decrease of the value function when there is only one channel; the superscript T and denote the matrix transpose and diagonal matrix respectively.

[0021] According to a satellite hyperspectral channel selection method provided by the present invention, a candidate channel set for screening the nth channel is constructed from the unselected channels by the following formula: :

[0022]

[0023]

[0024]

[0025] in, represents the angle between the Jacobian vectors of the jth unselected channel and the ith selected channel after linear transformation; Subsets The radiation observation value, radiation simulation value, radiation simulation Jacobian vector and observation error variance of the j-th channel; Represents the Jacobian vector of the i-th channel of the selected channel; is the adjustable angle increment parameter, Representing a subcollection The number of channels in the When it is an empty set, by adding Value to reconstruct the candidate set.

[0026] According to a satellite hyperspectral channel selection method provided by the present invention, the subset is divided into Each channel in is combined with the first n-1 selected channels, and the rate of decrease of the value function corresponding to each combination is calculated:

[0027]

[0028]

[0029]

[0030]

[0031]

[0032]

[0033]

[0034]

[0035]

[0036]

[0037]

[0038] in, Subsets The radiation observation value, radiation simulation value, radiation simulation Jacobian vector and observation error variance of the j-th channel; Representing a subcollection The number of channels; ,in represents the analytical solution of the atmospheric profile when n channels are used (n > 1), is its background vector, and the error covariance matrix is B; the subscript m is the number of vertical layers of the atmospheric profile; H nm Is the radiation simulation about Jacobian matrix; express The variance of the observation error on n channels is a diagonal matrix; Indicates the rate of decrease of the value function when n channels are used.

[0039] According to a satellite hyperspectral channel selection method provided by the present invention, the weight inverse matrix of the selected channel combination is determined by the following formula. and Calculate the weight inverse matrix according to the following recursive formula , the algorithm is as follows:

[0040]

[0041] The present invention also provides a satellite hyperspectral channel selection device, comprising:

[0042] The first channel selection module calculates the descent rate of the value function corresponding to each channel in the satellite hyperspectral channel set, and selects the channel with the largest descent rate as the first selected channel;

[0043] an alternative channel screening module, configured to linearly transform the Jacobian vectors of the unselected channels and the selected channels in the satellite hyperspectral channel set using a background error covariance matrix to obtain vectors, calculate the angle between the vectors of the unselected channels and the selected channels, and screen out an alternative channel subset from the unselected channels by setting an angle range so that the atmospheric vertical structure information observed by the channels in the alternative channel subset is significantly different from that of the selected channels;

[0044] The n-th channel selection module is used to combine each channel in the candidate channel subset with n-1 selected channels, where n≥2, and calculate the decrease rate of the value function of each combination using a recursive algorithm based on the (n-1)×(n-1)-dimensional weight inverse matrix of the selected channels, and select the channel corresponding to the combination with the largest decrease rate as the n-th selected channel;

[0045] The weight inverse matrix calculation module is used to calculate the n×n dimensional weight inverse matrix of the n selected channels using a recursive algorithm after obtaining the nth selected channel, so as to continue selecting the next channel.

[0046] The present invention also provides an electronic device, comprising a memory, a processor, and a parallel computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any one of the satellite hyperspectral channel selection methods described above is implemented.

[0047] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for selecting a satellite hyperspectral channel as described above is implemented.

[0048] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned satellite hyperspectral channel selection methods.

[0049] The present invention provides a satellite hyperspectral channel selection method and device. By addressing the variational assimilation problem of satellite data, the method directly estimates the contribution of each channel to the cost function and selects the satellite channel combination that can maximize the reduction in the cost function. The method also considers real-time satellite observation information, analyzes specific weather conditions, and dynamically selects channels based on observation data. The method is applicable to satellite channel selection in various weather conditions, thereby improving the reliability and accuracy of channel selection. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] FIG1 is a schematic diagram of a flow chart of a satellite hyperspectral channel selection method provided by the present invention;

[0052] Figure 2 1. It is a schematic diagram comparing the channels selected by the satellite hyperspectral channel selection method and the entropy subtraction method provided by the present invention and the corresponding simulated brightness temperature errors;

[0053] Figure 3 1 is a schematic diagram comparing the relative values of the value functions of the channel combinations selected by the satellite hyperspectral channel selection method and the entropy subtraction method provided by the present invention;

[0054] Figure 4 1. It is a schematic diagram comparing the entropy reduction change rate of the satellite hyperspectral channel selection method and the entropy reduction method provided by the present invention;

[0055] Figure 5 It is a structural schematic diagram of the satellite hyperspectral channel selection device provided by the present invention;

[0056] Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0057] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0058] Satellite hyperspectral channel selection can be formulated as follows: the satellite has a total of M hyperspectral observation channels, and their radiation observation values, radiation simulation values, radiation simulation Jacobian vectors, and observation error variances are known, and are defined as follows:

[0059]

[0060] Satellite channel selection is how to select the satellite channel based on the above information. Select a subset Data assimilation by channel combination can extract as much effective information of atmospheric variables as possible from the observed radiation data, thereby improving the initial field of the numerical model.

[0061] For the sake of To distinguish the members of The member symbols in the table have had the “^” symbol removed to indicate the relevant information of the selected channel. The symbols are renumbered and defined as follows:

[0062]

[0063] Faced with thousands of satellite observation channels, it is necessary to develop efficient algorithms to solve the computational complexity of the cost function and its analytical solution. The cost function is a direct criterion for measuring the quality of satellite channel combinations, because a good channel combination can definitely significantly reduce the cost function. For a combination of n satellite channels, , its one-dimensional value function and value function decline rate are as follows:

[0064]

[0065] vector It is the atmospheric vertical structure information to be solved, such as the vertical temperature and humidity profiles of the atmosphere, which are divided into m layers; yes The background vector of is usually provided by the numerical model; B is The error covariance matrix of is obtained from historical data statistics; the vector yes The vector of radiation observation values y of n channels corresponds to represents the radiation analogy of this vector, represents the observation error variance of the vector, which is a diagonal matrix. Under the first-order approximation condition formula (4), we can get The variational decomposition formula (5) is as follows:

[0066]

[0067]

[0068] Among them, H nm yes The Jacobian matrix of E is n×m. The entropy reduction index of the error covariance matrix is consistent with the channel selection index of the "atmospheric reversibility index" and the "entropy reduction method". Where det represents the determinant of the matrix.

[0069] The key innovation of this invention is how to efficiently estimate the cost function and its analytical solution. This is also one of the key challenges it addresses. Currently, entropy reduction is widely used to select satellite channels. However, this method only yields the channel combination with the highest theoretical probability relative to other combinations, minimizing the background error variance of the variational decomposition.

[0070] This embodiment addresses the variational assimilation problem of satellite data, directly estimates the contribution of each channel to the cost function, and selects the satellite channel combination that can maximize the decrease in the cost function. It considers real-time satellite observation information, analyzes specific weather conditions, and dynamically selects channels based on observation data. This makes it suitable for satellite channel selection in various weather conditions, thereby improving the reliability and accuracy of channel selection.

[0071] The following combination Figure 1 A satellite hyperspectral channel selection method of the present invention is described, comprising:

[0072] Step 101: Determine the satellite hyperspectral observation channel set The decrease rate of the value function corresponding to each channel in , a total of M, the channel with the largest decrease rate is selected as the first channel;

[0073] In this embodiment, the rate of decrease of the cost function corresponding to each satellite channel is determined by the following formula:

[0074]

[0075] in, Satellite observation channel sets The radiation observation value, radiation simulation value, radiation simulation Jacobian vector and radiation value error variance of the j-th channel; ,in represents the analytical solution of the atmospheric profile when using 1 channel, is its background vector, and m is the number of vertical layers of the atmospheric profile. The error covariance matrix is B; Represents the diagonal matrix of the error variance of the observed radiation values in n channels; Indicates the rate of decrease of the value function when there is only one channel; the superscript T and denote the matrix transpose and diagonal matrix respectively.

[0076] High-quality channel combination must make Significant decrease, therefore, The rate of decrease is used as the standard to measure the channel combination. Use enumeration to find The largest channel is selected as the first channel. When the first channel is selected, is a one-dimensional vector, is a number, so 、 and It is easy to calculate and get only one member The channel corresponding to 、 and , used for selecting the next channel.

[0077] Step 102: Calculate the angle between the Jacobian vectors of the unselected channel and the selected channel after linear transformation, and construct a subset of candidate channels for screening the nth selected channel from the unselected channels. , so that the sub-collection Contains multi-level atmospheric vertical structure information, where n>1;

[0078] This embodiment constructs a subset of candidate channels using the following formula: :

[0079]

[0080] in, represents the angle between the Jacobian vectors of the jth unselected channel and the ith selected channel after linear transformation; Subsets The radiation observation value, radiation simulation value, radiation simulation Jacobian vector and observation error variance of the j-th channel; Represents the Jacobian vector of the i-th channel of the selected channel; is the adjustable angle increment parameter, Representing a subcollection The number of channels in the When it is an empty set, by adding Value to reconstruct the candidate set.

[0081] How to make the channel combination contain as much high-precision information of atmospheric temperature and humidity at different altitudes as possible is another important and difficult problem that needs to be faced. Formula (5) shows that is the background vector Plus The linear combination of n column vectors of Therefore, whether or not to obtain multi-level atmospheric vertical structure information depends on Is there a certain angle between them so that each The peak values of are scattered at different height layers. Therefore, the Filter the nth channel in .

[0082] when When , it means that the two vectors are orthogonal and their vertical structural information is the most different. Complete orthogonality is the most ideal choice, but it is not satisfied in reality. Therefore, by adjusting the parameters Make Keep within a certain angle range. When it is an empty set, by adding Value to reconstruct the subcollection , the specific method can be found in the numerical experiment section.

[0083] Step 103: Subset Each channel in is combined with the first n-1 selected channels, and the drop rate of the cost function corresponding to each combination is calculated. The channel corresponding to the combination with the largest drop rate is taken as the nth selected channel. In this embodiment, the drop rate of the cost function is calculated using the following formula:

[0084]

[0085] in, Subsets The radiation observation value, radiation simulation value, radiation simulation Jacobian vector and observation error variance of the j-th channel in, Representing a subcollection The number of channels, H nm is a matrix consisting of n Jacobian vectors, H （n-1）m is the Jacobian matrix of the selected channel; ,in represents the analytical solution of the atmospheric profile when n channels are used, is its background vector, and m is the number of vertical layers of the atmospheric profile. The error covariance matrix is B; Represents the diagonal matrix of the observed radiation value error variance; It represents the rate of decrease of the value function when using n channel combinations.

[0086] Step 104: Get the corresponding 、 、 and , and then use the corresponding Calculated according to the recursive formula , and then set n=n+1 for the next channel selection process. The channel selection loops between step 102 and step 104 and ends when n=N. The recursive algorithm is as follows:

[0087]

[0088] One of the key challenges facing this invention is designing an efficient solution based on the matrix characteristics of the inverse matrix in the variational decomposition. The core idea behind selecting satellite channels is to choose a channel combination that significantly reduces the cost function. Therefore, it is necessary to find a variational analytical solution to the cost function, of which the inverse matrix is crucial.

[0089] Taking the GIIRS instrument of the Fengyun-4B satellite (FY4B) as an example, there are 1690 channels. If n-1 satellite channels have been selected, it is necessary to select the optimal channel as the nth channel from the remaining channels by enumeration. Then, it is necessary to solve the inverse matrix 1691-n times. The computational complexity of the commonly used inversion algorithms such as LU decomposition or Cholesky decomposition is When the value of n is large, solving the inverse matrix requires a very large amount of calculation, which is difficult to implement in actual business. Therefore, the algorithm for solving the inverse matrix is a key issue.

[0090] Comprehensive analysis of the above recursive formula shows that the computational complexity of formula (8) is approximately , mainly from . Use LU decomposition to calculate Then calculate the value function, the computational complexity is about , the efficiency is very low.

[0091] The present invention only needs to calculate once when selecting the nth channel , its computational complexity is , so the calculation amount mainly comes from formula (8), which needs to be calculated times. Entropy subtraction and atmospheric reversibility index also select channels by enumeration, and the computational complexity is ,Although this algorithm is not as efficient as the two algorithms, its ,computational complexity is quadratic, and when calculating the value function ,descent rate, the calculations between different channels are ,unrelated, and suitable for parallel computing, so the ,computational amount is completely acceptable.

[0092] More importantly, the channel combination found by the present invention is the combination with the highest value function decrease rate, which is consistent with the goal of variational decomposition and can obtain more atmospheric information. The recursive algorithm is used to derive the variational solution. The recursive algorithm is used to obtain the value function descent rate, which reduces the computational complexity to , which is one of the main core technologies of this invention

[0093] In order to test whether the channel combination can reflect the complete atmospheric vertical structure information, this experiment considers the channel selection under clear sky conditions. The GIIRS observation instrument of the Fengyun-4B (FY4B) satellite is used as an example to test the method of the present invention. GIIRS has a total of M=1690 hyperspectral channels, of which 1 to 723 are long-wave channels and 724-1690 are medium-wave channels. In order to make the experiment more representative, 11 station samples were used for testing and compared with the results of the entropy subtraction method. Background field Contains two variables: temperature T and absolute humidity q, 41 layers in the vertical direction, that is, ,m=82; using RTTOV-4.0 in radiation mode, we can get The simulated background error covariance matrix B is the statistical result under clear sky conditions; the radiation observation value and the observation error variance Provided by the GIIRS instrument. Alternative channel space The parameter θ is set to , where the initial value of k is 0, when When it is empty, k=k+1.

[0094] Figure 2 These are the channels and corresponding simulated brightness temperature errors selected by the method of the present invention (right figure) and the entropy subtraction method (left figure) in 11 sample experiments. Figure 2 This shows that there is a significant difference between the channels selected by the present invention and the entropy subtraction method. The entropy subtraction method does not select any channels in the three channel intervals, especially in the 1350-1690 channel of the medium wave. The common point is that the channels selected in the medium wave channel (to the right of the green line) are all channels with smaller simulation errors.

[0095] Since the present invention uses the rate of decline of the value function as the criterion for channel selection, Figure 3The value function shows an upward fluctuation between the 1-100 and 150-300 channels. However, after reaching its peak, it steadily decreases and converges, with the maximum decrease reaching 0.75. In contrast, the value function obtained by entropy subtraction does not decrease significantly, concentrating around 0.1. This is because entropy subtraction only selects the most probable combination of channels that minimizes variance. When this "maximum probability value" is relatively small, for most samples, the selected channel does not meet the conditions for a significant value function decrease.

[0096] Figure 4 It shows that both channel selection algorithms can make entropy reduction (i.e., the E value of formula (5)) decrease monotonically with the increase of the number of channels. Although the entropy reduction method outperforms the results of the present invention, the advantage is only less than 0.1.

[0097] The significance of the present invention is:

[0098] 1) Variational assimilation obtains the assimilation solution by solving the minimum value of the cost function. The method proposed in this paper selects the channels that can maximize the reduction of the cost function to form a satellite channel combination. Therefore, this invention is of great significance for the variational assimilation of satellite data.

[0099] 2) The present invention takes into account real-time satellite observation data and conducts specific analysis based on specific weather conditions. Therefore, channels can be dynamically selected based on factors such as the quality of the observation data. This makes it suitable for satellite channel selection before data assimilation in various weather conditions. This is superior to satellite channel selection methods based on statistical concepts, such as entropy subtraction, which only considers background field information.

[0100] 3) The method of the present invention has effectively solved the problem of inverse matrix calculation in covariance matrix update, and its computational complexity is Therefore, it can be integrated with other satellite channel selection algorithms. For example, entropy reduction is used before (after) the 50th channel, and the method of the present invention is used thereafter (before), so that the value function can drop rapidly.

[0101] The satellite hyperspectral channel selection device provided by the present invention is described below. The satellite hyperspectral channel selection device described below and the satellite hyperspectral channel selection method described above can be referenced to each other.

[0102] like Figure 5 As shown, the device includes a first channel selection module 501, an alternative channel screening module 502, an nth channel selection module 503 and a weight inverse matrix calculation module 504, wherein:

[0103] The first channel selection module 501 is used to determine the satellite hyperspectral channel set The rate of decrease of the value function corresponding to each channel in M channels, the channel with the largest rate of decrease is selected as the first channel;

[0104] The candidate channel screening module 502 is used to construct a candidate channel subset for screening the nth channel. , so that it contains atmospheric vertical structure information that is significantly different from the selected channel, where n>1 is an integer;

[0105] The nth channel selection module 503 is used to select the subset Each channel in is combined with the first n-1 selected channels respectively, and the recursive algorithm of the invention is used to calculate the decline rate of the value function corresponding to each combination, and the channel corresponding to the combination with the largest decline rate is taken as the nth selected channel.

[0106] The weight inverse matrix calculation module 504 calculates the n×n dimensional weight inverse matrix using a recursive formula based on the Jacobian vector corresponding to the selected n-th channel and the known n-1×n-1 dimensional weight inverse matrix, thereby reducing the computational complexity of solving the inverse matrix.

[0107] This embodiment directly estimates the contribution of each channel to the cost function by addressing the variational assimilation problem of satellite data, and selects the satellite channel combination that can maximize the reduction in the cost function. It considers real-time satellite observation information, analyzes specific weather conditions, and dynamically selects channels based on observation data. This makes it suitable for satellite channel selection in various weather conditions, thereby improving the reliability and accuracy of channel selection.

[0108] Figure 6 An example of a physical structure diagram of an electronic device is shown below. Figure 6 As shown, the electronic device may include: a processor 610, a communications interface 620, a memory 630, and a communications bus 640. The processor 610, the communications interface 620, and the memory 630 communicate with each other via the communications bus 640. The processor 610 can call the logic instructions in the memory 630 to execute the satellite hyperspectral channel selection method. The device functions include: 1) reading hyperspectral satellite observation data, radiation pattern simulation data, background covariance matrix information, etc.; 2) calling the first channel selection module 501, the candidate channel screening module 502, the nth channel selection module 503, and the weight inverse matrix calculation module 504 of the computer program to implement parallel operations of the hyperspectral channel selection algorithm; and 3) storing the hyperspectral channel selection results on a medium.

[0109] Furthermore, the logic instructions in the aforementioned memory 630, when implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0110] On the other hand, the present invention also provides a computer program product based on Fortran language, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, it can execute the above-mentioned satellite hyperspectral channel selection method. The method sequentially calls the first channel selection module 501, the alternative channel screening module 502, the nth channel selection module 503 and the weight inverse matrix calculation module 504, then adjusts n=n+1, and then returns to calling the alternative channel screening module 502 to loop until n is equal to a preset value and stops the loop.

[0111] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0112] Through the description of the above embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A satellite hyperspectral channel selection method comprising: Calculate the drop rate of the value function corresponding to each channel in the satellite hyperspectral channel set, and select the channel with the largest drop rate as the first selected channel; A vector is obtained by linearly transforming the Jacobian vectors of the unselected channels and the selected channels in the satellite hyperspectral channel set using a background error covariance matrix, and an angle between the vectors of the unselected channels and the selected channels is calculated. A subset of candidate channels is selected from the unselected channels by setting a range of angles, so that atmospheric vertical structure information observed by the channels in the subset of candidate channels is different from that of the selected channels; Combine each channel in the candidate channel subset with n-1 selected channels, where n≥2, and calculate the decrease rate of the value function of each combination using a recursive algorithm based on the (n-1)×(n-1)-dimensional weight inverse matrix of the selected channels. The channel corresponding to the combination with the largest decrease rate is taken as the nth selected channel. After obtaining the nth selected channel, a recursive algorithm is used to calculate the n×n dimensional weight inverse matrix of the selected channel to continue selecting the next channel.

2. The satellite hyperspectral channel selection method according to claim 1, characterized in that: The satellite hyperspectral observation channel set is calculated by the following formula The rate of decrease of the value function corresponding to the M channels in total: ； ； ； ； ； ； ； in, They are The radiation observation value, radiation simulation value, radiation simulation Jacobian vector and observation error variance of the j-th channel; ,in represents the analytical solution of the atmospheric profile when using 1 channel; is its background vector, and the error covariance matrix is B, which is obtained from historical data statistics; the subscript m is the vertical layer number of the atmospheric profile; Indicates the rate of decrease of the value function when there is only one channel; the superscript T and denote the matrix transpose and diagonal matrix respectively.

3. The satellite hyperspectral channel selection method according to claim 2, characterized in that: The Jacobian vectors of the unselected channels and the selected channels in the satellite hyperspectral channel set are linearly transformed using the background error covariance matrix using the following formula to obtain a vector, and the angle between the vectors of the unselected channels and the selected channels is calculated. The candidate channel subset is screened from the unselected channels by setting the angle range. , making The atmospheric vertical structure information observed in the middle channel is different from that in the selected n-1 channels: ； ； ； in, Represents the angle between the linearly transformed Jacobian vectors of the j-th unselected channel and the i-th selected channel; Alternative channel subsets The radiation observation value, radiation simulation value, radiation simulation Jacobian vector and observation error variance of the j-th channel; Represents the Jacobian vector of the i-th channel of the selected channel; is the adjustable angle increment parameter, Indicates a subset of candidate channels The number of channels in When it is an empty set, by adding Value to reconstruct the subcollection .

4. The satellite hyperspectral channel selection method according to claim 3, characterized in that: The alternative channel subset is used by the following formula Each channel in is combined with n-1 selected channels and the (n-1)×(n-1) dimensional inverse weight matrix of the selected channels is calculated. Use the recursive algorithm to calculate the decline rate of the value function of each combination: ； ； ； ； ； ； ； ； ； ； ； in, Alternative channel subsets The radiation observation value, radiation simulation value, radiation simulation Jacobian vector and observation error variance of the j-th channel; Representing a subcollection The number of channels; ,in represents the analytical solution of the atmospheric profile when n channels are used, n > 1, is its background vector, and the error covariance matrix is B; the subscript m is the number of vertical layers of the atmospheric profile; H nm Is the radiation simulation about Jacobian matrix; Represents the observation error variance of n channels, which is a diagonal matrix; Indicates the rate of decrease of the value function when n channels are used.

5. The satellite hyperspectral channel selection method according to claim 4, characterized in that: The inverse weight matrix of n selected channels According to the selected nth channel corresponding and Calculated by the following recursive formula:

6. A satellite hyperspectral channel selection device, characterized by: The first channel selection module is used to calculate the descent rate of the value function corresponding to each channel in the satellite hyperspectral channel set, and select the channel with the largest descent rate as the first selected channel; an alternative channel screening module, configured to linearly transform the Jacobian vectors of the unselected channels and the selected channels in the satellite hyperspectral channel set using a background error covariance matrix to obtain vectors, calculate the angle between the vectors of the unselected channels and the selected channels, and screen out an alternative channel subset from the unselected channels by setting an angle range so that the atmospheric vertical structure information observed by the channels in the alternative channel subset is significantly different from that of the selected channels; The n-th channel selection module is used to combine each channel in the candidate channel subset with n-1 selected channels, where n≥2, and calculate the decrease rate of the value function of each combination using a recursive algorithm based on the (n-1)×(n-1)-dimensional weight inverse matrix of the selected channels, and select the channel corresponding to the combination with the largest decrease rate as the n-th selected channel; The weight inverse matrix calculation module is used to calculate the n×n weight inverse matrix of the n selected channels using a recursive algorithm after obtaining the nth selected channel, so as to continue selecting the next channel.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the satellite hyperspectral channel selection method according to any one of claims 1 to 5 is implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the satellite hyperspectral channel selection method according to any one of claims 1 to 5 is implemented.

9. A computer program product comprising a computer program, characterized in that: When the computer program is executed by a processor, the satellite hyperspectral channel selection method according to any one of claims 1 to 5 is implemented.

Images