An evaluation method of the influence of the design signal-to-noise ratio of a satellite remote sensing payload on inversion performance

Abstract

The application provides an evaluation method for the influence of the design signal-to-noise ratio of a satellite remote sensing load on inversion performance, takes a hyperspectral load and a high-radiation-precision reference load as a basic data source, associates the hyperspectral load predicted radiance with water environmental element quantitative inversion results after time-space, numerical value and spectrum matching, simulates the hyperspectral load predicted radiance according to the research load design index, obtains the simulation radiance of the research load, simulates the function relationship between the simulation radiance of the research load and the water environmental element quantitative inversion results through a neural network model, samples the test data set from the research load simulation radiance-water environmental element quantitative inversion result data set, adds noise to the test data, obtains the influence degree of different noise on the inversion of different water environmental element quantitative values, and determines the signal-to-noise ratio in combination with the design requirement. The application realizes the application efficiency evaluation of the index on the simulation image water environmental element inversion result.

Description

Technical Field

[0001] This invention relates to a method for evaluating the impact of signal-to-noise ratio on inversion performance in satellite remote sensing payload design, belonging to the field of remote sensing satellite design technology. Background Technology

[0002] Spaceborne remote sensing cameras are widely used in Earth observation, leveraging their high-precision radiometric performance to achieve quantitative monitoring of land, atmosphere, and ocean, serving multiple fields such as ecological and environmental protection, resource exploration, and emergency management, demonstrating immense application value. The signal-to-noise ratio (SNR) of satellite remote sensing payloads has a profound impact on the effectiveness of quantitative applications of on-orbit image data. However, during the design phase, it is difficult to directly obtain corresponding on-orbit images based on different SNRs, making it impossible to assess their impact on application effectiveness. This leads to a discrepancy between the camera's design specifications and actual applications, limiting the camera's future application potential.

[0003] The signal-to-noise ratio (SNR) directly affects the quantitative detection capability of satellite remote sensing payloads. Taking ocean color remote sensing as an example, the intrinsic signal reflecting the optical properties of water bodies accounts for a relatively low proportion of the total satellite observation signal and is highly susceptible to noise. This is especially true for Class II turbid water bodies such as nearshore shallow seas and inland rivers and lakes. A low SNR will result in the quantitative detection results of water environmental elements (generally including chlorophyll concentration, suspended solids concentration, and colored soluble organic matter concentration) failing to accurately reflect the regional water quality. Therefore, it is extremely important to conduct SNR assessment based on the typical regional observation requirements proposed by users. A reasonable SNR will improve the application effectiveness of the satellite and increase its future application value. Summary of the Invention

[0004] The technical problem solved by this invention is to overcome the shortcomings of the prior art and provide an evaluation method for the impact of the signal-to-noise ratio of satellite remote sensing payload design on inversion performance. This method evaluates the impact of the signal-to-noise ratio of satellite remote sensing payload design on the effectiveness of on-orbit quantitative applications. By establishing the coupling relationship between the currently orbiting hyperspectral payload, high radiometric reference payload, and their quantitative inversion results of water environment elements, it outputs simulated on-orbit images corresponding to different signal-to-noise ratios. Based on the inversion results of water environment elements from the simulated images, the application effectiveness of the indicators is evaluated.

[0005] The technical solution of this invention is:

[0006] This invention discloses a method for evaluating the impact of the signal-to-noise ratio of satellite remote sensing payload design on the inversion performance of water environment elements, comprising:

[0007] Step (1): Construct a basic dataset, which includes the predicted radiance of the hyperspectral payload and the quantitative inversion values ​​of water environment elements;

[0008] Step (2): Based on the basic dataset, the simulated radiance of the load under development is obtained through simulation, and a dataset of simulated radiance of the load under development and quantitative inversion values ​​of water environment elements is constructed.

[0009] Step (3): Randomly sample the dataset of simulated radiance of the under-construction load and quantitative inversion values ​​of water environment elements to obtain a training set and a test set. Train the neural network model based on the training set to obtain the trained neural network model. Input the simulated radiance of the under-construction load and calculate the quantitative inversion values ​​of water environment elements simulated by the model.

[0010] Step (4): Add a standard deviation σ to each quantitative inversion value of the water environment element simulated by the model in the test set. N After removing Gaussian noise, the data is input into the trained neural network model to obtain simulated quantitative inversion values.

[0011] Step (5): Repeat step (4) N times to obtain N simulated quantitative inversion values, and calculate the standard deviation of the N simulated quantitative inversion values;

[0012] Step (6): Change the signal-to-noise ratio of the Gaussian noise, repeat steps (4) to (5), and obtain the simulated quantitative inversion values ​​and standard deviations corresponding to different signal-to-noise ratios;

[0013] Step (7): Based on the signal-to-noise ratio, simulated quantitative inversion value and standard deviation in step (6), calculate the signal-to-noise ratio application effectiveness impact factor and evaluate the impact of the signal-to-noise ratio on simulated quantitative inversion values ​​with different values.

[0014] In the above evaluation method, step (1) involves constructing a basic dataset, specifically as follows:

[0015] Step S11: Select the hyperspectral payload and the high radiometric precision multispectral payload as reference payloads respectively. Based on the spatiotemporal and numerical matching principle, obtain the data of the hyperspectral payload and the reference payload synchronously or quasi-synchronously imaging in the same area.

[0016] Step S12: Perform data processing and data matching on the data of the hyperspectral payload and the reference payload to obtain quantitative inversion values ​​of water environment elements that match the numerical values ​​of the reference payload data;

[0017] Step S13: Perform band matching based on the hyperspectral load and the reference load to obtain the matching results;

[0018] Step S14: Based on the matching results and the atmospheric radiative transfer model, calculate the spectral matching factor between the hyperspectral load and the reference load;

[0019] Step S15: Based on the spectral matching factor, and combined with the atmospheric top radiance of a pixel obtained from the hyperspectral load observation, predict the predicted radiance that matches the observation conditions of the hyperspectral load and the reference load.

[0020] Step S16: Based on the predicted radiance and quantitative inversion values ​​of water environment elements, construct the basic dataset of predicted radiance and quantitative inversion values ​​of water environment elements for hyperspectral load.

[0021] In the above evaluation method, step S12 involves data processing and matching of the hyperspectral payload data and the reference payload data to obtain quantitative inversion values ​​of water environment elements that match the reference payload data. Specifically:

[0022] S21: The spatial resolution of the hyperspectral payload data is sampled to be the same as that of the reference payload data to obtain the sampled hyperspectral payload data;

[0023] S22: Using the operational algorithm for water environment elements with reference load, the quantitative inversion of regional water environment elements is realized, and the quantitative inversion values ​​of water environment elements are obtained.

[0024] S23: Based on the range of values ​​of the quantitative inversion values ​​of water environment elements, the sampled hyperspectral load data is matched with the corresponding quantitative inversion values ​​at that point.

[0025] In the above evaluation method, step S11, the spatiotemporal and numerical matching principle, is specifically as follows:

[0026] The zenith angle shall not exceed 60°, and the solar altitude angle shall not exceed 75°;

[0027] The reflectance of remote sensing in the visible light band is not negative;

[0028] No clouds or shadows obstructing the view;

[0029] The water environment element value is μ i , Where μ min μ min These represent the minimum and maximum values ​​of the quantitative inversion values ​​of regional water environment elements, respectively, N. Para The total number of water environment element values ​​selected for numerical matching. The sampling step size, i, is defined as the range of 1 to N within the quantitative inversion values ​​of regional water environment elements. Para .

[0030] In the above evaluation method, step (2) is based on the basic dataset, and the simulated radiance of the load under development is obtained through simulation. The dataset of simulated radiance of the load under development and quantitative inversion values ​​of water environment elements is constructed as follows:

[0031] Based on the band settings of the payloads under development, the predicted radiance of the hyperspectral payloads in the basic dataset in step (1) is convolved to obtain the equivalent band radiance.

[0032] Based on the equivalent band radiance, the simulated radiance of the payload under research is obtained after modulation transfer function simulation. This, together with the corresponding quantitative inversion values ​​of water environment elements in the basic dataset, constitutes the dataset of quantitative inversion values ​​of water environment elements for the simulated radiance of the payload under research.

[0033] The radiance of the load under research was obtained after simulation of the modulation transfer function, specifically as follows:

[0034] G(u,v)=F(u,v)*M(u,v)

[0035] In the formula, G(u,v) represents the simulated radiance, F(u,v) is the Fourier transform of the image composed of equivalent band radiance, and M(u,v) is the modulation transfer function.

[0036] In the above evaluation method, step (4) adds a standard deviation σ to each quantitative inversion value of the water environment element simulated by the model in the test set. N After removing Gaussian noise, the result is input into the trained neural network model to obtain the simulated quantitative inversion value, as shown in the formula:

[0037] WaterPara predict =f NN (L design,k +Gauss N )

[0038] Among them, WaterPara predict For the quantitative inversion results of the neural network model simulation, f NN For neural network models, Gauss N Gaussian noise; L design,k The equivalent band radiance of the k-th band of the payload under development

[0039] In the above evaluation method, step (7) calculates the signal-to-noise ratio application effectiveness impact factor, and the specific method is as follows:

[0040]

[0041] Where E is the signal-to-noise ratio application effectiveness impact factor, σ predict To simulate the standard deviation of the quantitative inversion values, σ N The standard deviation of Gaussian noise;

[0042]

[0043] Where N is the number of simulations; The mean of N simulated quantitative inversion values ​​of water environment elements; WaterPara predict This provides quantitative inversion results of simulated water environment elements from a neural network model.

[0044] In the above evaluation method, step S15, predicting the predicted radiance that matches the observation conditions of the hyperspectral load with the reference load, specifically involves:

[0045]

[0046]

[0047] in, The predicted radiance of the i-th band of the hyperspectral load, is the sum of the predicted radiance of the i-th, i+1, ..., i+M-1 bands of the hyperspectral load, and f is the spectral matching factor.

[0048] In the above evaluation method, step S14, the spectral matching factor, is specifically as follows:

[0049]

[0050] L Hyspec,sum =∑a i L Hyspec,i

[0051]

[0052] Among them, L Hyspec,sum L is the equivalent top atmospheric radiance of the weighted superposition of bands i, i+1, ..., i+M-1 of the hyperspectral payloads, where i is the band number of the hyperspectral payload, M is the total number of bands of the hyperspectral payloads that satisfy the band matching principle, and L is the equivalent top atmospheric radiance of the hyperspectral payloads. refer,j For the atmospheric top radiance of the reference payload in band j, a i R represents the weighting coefficients for each band. Hyspec,i (λ) represents the relative spectral response of the hyperspectral load in the i-th band at wavelength λ. R is the sum of the relative spectral responses in that band. refer,j (λ) represents the relative spectral response of the reference load in the j-th band at wavelength λ. This is the sum of the relative spectral responses in that band.

[0053] This invention provides a technical solution for evaluating the impact of satellite remote sensing payload design signal-to-noise ratio on the inversion performance of water environment elements. It enables the analysis of the impact of on-orbit data on the inversion performance of payloads under development, simplifying the simulation process for such data. Furthermore, it offers a method for constructing simulation datasets based on spatiotemporal, numerical, and spectral matching methods, reducing the human, material, financial, and time costs associated with constructing datasets through in-situ experimental measurements. Through simulation, the impact of design signal-to-noise ratio on the inversion performance of water environment elements is evaluated, providing a pathway for the rational planning of signal-to-noise ratio indicators.

[0054] (1) This invention aims to provide a reference for signal-to-noise ratio evaluation during the design phase of satellite remote sensing payloads. Attached Figure Description

[0055] Figure 1 This is a flowchart of the evaluation method of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this invention clearer and easier to understand, the technical solutions of this invention will be described in detail below with reference to the accompanying drawings.

[0057] This invention discloses a method for evaluating the impact of the signal-to-noise ratio of satellite remote sensing payload design on the inversion performance of water environment elements, comprising:

[0058] Step (1): Construct a basic dataset, which includes the predicted radiance of the hyperspectral payload and the quantitative inversion values ​​of water environment elements;

[0059] Step (2): Based on the basic dataset, the simulated radiance of the load under development is obtained through simulation, and a dataset of simulated radiance of the load under development and quantitative inversion values ​​of water environment elements is constructed.

[0060] Step (3): Randomly sample the dataset of simulated radiance of the under-construction load and quantitative inversion values ​​of water environment elements to obtain a training set and a test set. Train the neural network model based on the training set to obtain the trained neural network model. Input the simulated radiance of the under-construction load and calculate the quantitative inversion values ​​of water environment elements simulated by the model.

[0061] Step (4): Add a standard deviation σ to each quantitative inversion value of the water environment element simulated by the model in the test set. N After removing Gaussian noise, the data is input into the trained neural network model to obtain simulated quantitative inversion values.

[0062] Step (5): Repeat step (4) N times to obtain N simulated quantitative inversion values, and calculate the standard deviation of the N simulated quantitative inversion values;

[0063] Step (6): Change the signal-to-noise ratio of the Gaussian noise, repeat steps (4) to (5), and obtain the simulated quantitative inversion values ​​and standard deviations corresponding to different signal-to-noise ratios;

[0064] Step (7): Based on the signal-to-noise ratio, simulated quantitative inversion value and standard deviation in step (6), calculate the signal-to-noise ratio application effectiveness impact factor and evaluate the impact of the signal-to-noise ratio on simulated quantitative inversion values ​​with different values.

[0065] In step (1), the basic dataset is constructed, specifically as follows:

[0066] Step S11: Select the hyperspectral payload and the high radiometric precision multispectral payload as reference payloads respectively. Based on the spatiotemporal and numerical matching principle, obtain the data of the hyperspectral payload and the reference payload synchronously or quasi-synchronously imaging in the same area.

[0067] Step S12: Perform data processing and data matching on the data of the hyperspectral payload and the reference payload to obtain quantitative inversion values ​​of water environment elements that match the numerical values ​​of the reference payload data;

[0068] Step S13: Perform band matching based on the relative spectral response function of the hyperspectral load and the reference load to obtain the matching result;

[0069] Step S14: Based on the matching results and the atmospheric radiative transfer model, calculate the spectral matching factor between the hyperspectral load and the reference load;

[0070] Step S15: Based on the spectral matching factor, and combined with the atmospheric top radiance of a pixel obtained from the hyperspectral load observation, predict the predicted radiance that matches the observation conditions of the hyperspectral load and the reference load.

[0071] Step S16: Based on the predicted radiance and quantitative inversion values ​​of water environment elements, construct the basic dataset of predicted radiance and quantitative inversion values ​​of water environment elements for hyperspectral load.

[0072] In step S12, data processing and matching are performed on the data from the hyperspectral payload and the reference payload to obtain quantitative inversion values ​​of water environment elements that match the reference payload data. Specifically:

[0073] S21: The spatial resolution of the hyperspectral payload data is sampled to be the same as that of the reference payload data to obtain the sampled hyperspectral payload data;

[0074] S22: Using the operational algorithm for water environment elements with reference load, the quantitative inversion of regional water environment elements is realized, and the quantitative inversion values ​​of water environment elements are obtained.

[0075] S23: Based on the range of values ​​of the quantitative inversion values ​​of water environment elements, the sampled hyperspectral load data is matched with the corresponding quantitative inversion values ​​at that point.

[0076] In step S11, the spatiotemporal and numerical matching principles are as follows:

[0077] The zenith angle shall not exceed 60°, and the solar altitude angle shall not exceed 75°;

[0078] The reflectance of remote sensing in the visible light band is not negative;

[0079] No clouds or shadows obstructing the view;

[0080] The water environment element value is μ i , Where μ min μ min These represent the minimum and maximum values ​​of the quantitative inversion values ​​of regional water environment elements, respectively, N. Para The total number of water environment element values ​​selected for numerical matching. The sampling step size, i, is defined as the range of 1 to N within the quantitative inversion values ​​of regional water environment elements. Para .

[0081] In step (2), based on the basic dataset, the simulated radiance of the load under development is obtained through simulation. A dataset of simulated radiance of the load under development and quantitative inversion values ​​of water environment elements is then constructed, specifically as follows:

[0082] Based on the band settings of the payloads under development, the predicted radiance of the hyperspectral payloads in the basic dataset in step (1) is convolved to obtain the equivalent band radiance.

[0083] Based on the equivalent band radiance, the simulated radiance of the payload under research is obtained after modulation transfer function simulation. This, together with the corresponding quantitative inversion values ​​of water environment elements in the basic dataset, constitutes the dataset of quantitative inversion values ​​of water environment elements for the simulated radiance of the payload under research.

[0084] The simulated radiance of the load under research was obtained after simulation of the modulation transfer function, specifically as follows:

[0085] G(u,v)=F(u,v)*M(u,v)

[0086] In the formula, G(u,v) represents the simulated radiance, F(u,v) is the Fourier transform of the image composed of equivalent band radiance, and M(u,v) is the modulation transfer function.

[0087] In step (4), a standard deviation σ is added to each quantitative inversion value of the water environment element simulated by the model in the test set. N After removing Gaussian noise, the result is input into the trained neural network model to obtain the simulated quantitative inversion value, as shown in the formula:

[0088] WaterPara predict =f NN (L design,k +Gauss N )

[0089] Among them, WaterPara predict For the quantitative inversion results of the neural network model simulation, f NN For neural network models, Gauss N Gaussian noise; L design,k The equivalent band radiance of the k-th band of the payload under development;

[0090] In step (7), the signal-to-noise ratio application effectiveness impact factor is calculated using the following method:

[0091]

[0092] Where E is the signal-to-noise ratio application effectiveness impact factor, σ predict To simulate the standard deviation of the quantitative inversion values, σ N The standard deviation of Gaussian noise;

[0093]

[0094] Where N is the number of simulations; The mean of N simulated quantitative inversion values ​​of water environment elements; WaterPara predict This provides quantitative inversion results of simulated water environment elements from a neural network model.

[0095] In step S15, the predicted radiance that matches the observation conditions of the hyperspectral payload with those of the reference payload is specifically as follows:

[0096]

[0097]

[0098] in, The predicted radiance of the i-th band of the hyperspectral load, is the sum of the predicted radiance of the i-th, i+1, ..., i+M-1 bands of the hyperspectral load, and f is the spectral matching factor.

[0099] In step S14, the spectral matching factor is specifically:

[0100]

[0101] L Hyspec,sum =∑a i L Hyspec,i

[0102]

[0103] Among them, L Hyspec,sum L is the equivalent top atmospheric radiance of the weighted superposition of bands i, i+1, ..., i+M-1 of the hyperspectral payloads, where i is the band number of the hyperspectral payload, M is the total number of bands of the hyperspectral payloads that satisfy the band matching principle, and L is the equivalent top atmospheric radiance of the hyperspectral payloads. refer,j For the atmospheric top radiance of the reference payload in band j, a i R represents the weighting coefficients for each band. Hyspec,i (λ) represents the relative spectral response of the hyperspectral load in the i-th band at wavelength λ. R is the sum of the relative spectral responses in that band. refer,j (λ) represents the relative spectral response of the reference load in the j-th band at wavelength λ. This is the sum of the relative spectral responses in that band.

[0104] Example

[0105] This embodiment provides a method for evaluating the impact of the signal-to-noise ratio of satellite remote sensing payload design on the inversion performance of water environment elements. The operation process of this method is as follows: Figure 1 As shown, the specific steps include:

[0106] Step 1: Building the basic dataset

[0107] Step 2: Simulation of under-development load data

[0108] Step 3: Training the Neural Network Model

[0109] Step 4: Obtain simulation data and inversion results for different signal-to-noise ratios.

[0110] Step 5: Evaluate the impact of signal-to-noise ratio on the quantitative inversion results of different values.

[0111] Specifically, the implementation method for step 1 is as follows:

[0112] (1) Select a hyperspectral payload that is currently operating stably in orbit. The Gaofen-5 hyperspectral camera is used as an example below. A high-radioprecision multispectral payload is selected as a reference, with MODIS Terra / Aqua used as an example below. Based on the principle of spatiotemporal matching, acquire data from synchronous or quasi-synchronous imaging of the hyperspectral payload and the reference payload in the same area. Quasi-synchronous imaging requires that the imaging time difference between the two payloads does not exceed 48 hours. It also requires that the weather conditions be good during imaging, with cloud cover below 15% and the images not contaminated by solar glare from the sea surface. The data processing and matching process is as follows:

[0113] (1a) The two types of payload data were preprocessed separately, including radiometric calibration and geometric coarse correction. The spatial resolution of the two types of payload data was sampled to the same value to eliminate the influence of scale differences, and the hyperspectral payload was image registered with the reference payload as a reference.

[0114] (1b) Quantitative inversion of regional water environment elements is achieved using an operational algorithm based on reference load. The following section uses the inversion of chlorophyll concentration in marine water bodies as an example, employing the MODIS OC3 operational algorithm to achieve chlorophyll concentration inversion, expressed as follows:

[0115]

[0116] X = lg(max(R) rs (443),R rs (488)) / R rs (547))

[0117] In the formula, CHLa represents the chlorophyll concentration, and R... rs The values ​​represent remote sensing reflectance, 443, 488, and 547 represent the center wavelengths of the bands involved in the calculation, and a, b, c, d, and e are undetermined coefficients that can be fitted using measured data. The default values ​​can be a = 0.2424, b = -2.7430, c = 1.8017, d = 0.0015, and e = -1.2280.

[0118] Numerical matching of hyperspectral loading, reference loading, and quantitative inversion values ​​was performed based on the following principles:

[0119] The zenith angle and the solar altitude angle shall not exceed 60° and 75°, respectively.

[0120] The reflectance of remote sensing in the visible light band is not negative;

[0121] No clouds or shadows obstructing the view;

[0122] Chlorophyll concentration is taken as μ i , Where μ min μ min These represent the minimum and maximum values ​​of the quantitative inversion values ​​of regional water environment elements, respectively, N. Para The total number of water environment element values ​​selected for numerical matching. This refers to the sampling step size within the range of quantitative inversion values ​​of regional water environment elements, where i ranges from 1 to N. Para ;

[0123] Based on this, corresponding points of the hyperspectral payload and reference payload images are selected, and the inversion results of the hyperspectral payload, reference payload images and the corresponding water environment elements of the points are matched in time, space and numerically.

[0124] (1c) Calculate the spectral matching factor between the hyperspectral payload and the reference payload based on the atmospheric radiative transfer model. Considering the difference in spectral sampling intervals between the hyperspectral payload and the reference payload, it is assumed that the ranges of the i-th, i+1, ..., i+M bands of the hyperspectral payload overlap with the range of the j-th band of the reference payload, i.e.

[0125]

[0126] In the formula, This represents the lower limit of the range of the i-th band of the hyperspectral payload. This represents the upper limit of the range of the (i+M-1)th band of the hyperspectral payload. These represent the lower and upper limits of the j-th band of the reference load, respectively, and M is the total number of bands of the hyperspectral load that satisfy the band matching principle.

[0127] In this embodiment, the band matching process is based on the band range. Assuming that the ranges of the second and third bands of the hyperspectral payload are [601, 602] (unit: nm) and [603, 604] respectively, and the range of the second band of the reference payload is [601, 604], then it is considered that the second and third bands of the hyperspectral payload match the second band of the reference payload.

[0128] By inputting the surface reflectance, atmospheric parameters, and observation conditions of the corresponding channel into the atmospheric radiative transfer model, the atmospheric top radiance of the i-th, i+1, ..., i+M-1 bands of the hyperspectral payload and the j-th band of the reference payload can be obtained.

[0129] Let L be the equivalent top atmospheric radiance of the weighted superposition of the i-th, i+1, ..., i+M-1 bands of the hyperspectral payload. Hyspec,sum The atmospheric top radiance of the reference payload in band j is L. refer,j The spectral matching factor f is

[0130]

[0131] The weighting process for the entrance pupil radiance of the hyperspectral loading is as follows:

[0132] L Hyspec,sum =∑a i L Hyspec,i

[0133]

[0134] In the formula a i R represents the weighting coefficient for each band. Hyspec,i (λ) represents the relative spectral response of the hyperspectral load in the i-th band at wavelength λ. R represents the sum of the relative spectral responses in that band. refer,j(λ) represents the relative spectral response of the reference load in the j-th band at wavelength λ. This represents the sum of the relative spectral responses for that band. The surface reflectance and atmospheric parameters required for the atmospheric radiative transfer model can be obtained by atmospheric correction of the reference payload or by field measurements, and the observation conditions can be obtained from the image data files.

[0135] For the hyperspectral load top radiance of the matching result in (1b), based on the spectral matching factor f and the weighting coefficient a i The process for calculating the top-of-atmosphere radiance, i.e., the predicted radiance, that matches the observation conditions of the reference payload is as follows:

[0136]

[0137]

[0138] In the formula This represents the predicted radiance of the i-th band of the hyperspectral load. This represents the sum of predicted radiance from multiple hyperspectral loads.

[0139] Specifically, the implementation method for step 2 is as follows:

[0140] (2) Based on the band settings of the payload under development, the predicted radiance of the hyperspectral payload obtained in step 1 is convolved to obtain the equivalent band radiance σ of the k-th band of the payload under development. predict :

[0141]

[0142] In the formula R design,k This represents the relative spectral response of the k-th band of the payload under development.

[0143] Simulation of the modulation transfer function (MTF) of the image after Fourier transform:

[0144] G(u,v)=F(u,v)*M(u,v)

[0145] In the formula, G(u,v) represents the simulation result, F(u,v) is the Fourier transform of the image composed of equivalent band radiance, and M(u,v) is the modulation transfer function. The spatial domain simulation result is obtained by inverse transforming the frequency domain simulation result.

[0146] Specifically, the implementation method for step 3 is as follows:

[0147] (3) For the simulation radiance-quantitative inversion value dataset of the under-construction load obtained in steps 1 and 2, a training set and a test set are randomly sampled, and the proportion of the number of training sets to the total number of data is set to 0.7.

[0148] A neural network model was trained based on the training set. The input was the simulated radiance of the under-construction load after MTF simulation, and the output was the corresponding simulated quantitative inversion value. A feedforward neural network was selected as the model, with the Sigmoid function chosen as the activation function for the neurons. After adjustment, the model contained one hidden layer with 10 basic units. The Levenberg-Marquardt algorithm was used for model training, with mean squared error as the training metric.

[0149] Specifically, the implementation method for step 4 is as follows:

[0150] (4) Add a standard deviation of σ to each data point in the test set. N The Gaussian noise is input into the neural network model obtained in step 3 to obtain the corresponding simulated quantitative inversion value:

[0151] CHLa predict =f NN (L design,k +Gauss N )

[0152] In the formula CHLa predict f represents the simulated quantitative inversion value of the neural network model. NN This represents the neural network model obtained in step 3, Gaussian. N This represents Gaussian noise, which follows a Gaussian distribution and has a standard deviation of σ. N The noise mean is the chlorophyll concentration value in μ in the test set. i .

[0153] The repeatable process is described as follows:

[0154] (4a) For each data point in the test set, repeat the above process a certain number of times N, and then calculate the standard deviation σ of the chlorophyll concentration quantification results from multiple simulations. predict This value reflects the impact of noise on the dispersion of the quantitative inversion results.

[0155]

[0156] N can be 2000.

[0157] (4b) Change the standard deviation σ of Gaussian noise N Repeat the above process, and in this paper, we take σ. N The numerical range is 1 to 30, with a step size of 1.

[0158] Specifically, the implementation method for step 5 is as follows:

[0159] (5) The signal-to-noise ratio is defined as:

[0160]

[0161] For a given signal-to-noise ratio level and its noise standard deviation σ N The standard deviation σ is calculated by taking simulated quantitative inversion values ​​of different magnitudes. predict The ratio of the two is used as the signal-to-noise ratio application effectiveness factor E:

[0162]

[0163] By repeating the calculation in step 4, we obtain a data table of three values: SNR (reflecting the magnitude of the signal-to-noise ratio), μ (reflecting the range of values ​​of the quantitative inversion result), and E (reflecting the application effectiveness of the signal-to-noise ratio), and curves can be plotted for each value.

[0164] Based on the design requirements of the payload under development, namely the range of signal-to-noise ratio (SNR), the range of quantitative parameters of water environment elements in typical observation areas proposed by the user, as well as the overall system design principles, cost, and design difficulty, a reasonable SNR value was finally determined.

[0165] The above description is only the best specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

[0166] The contents not described in detail in this specification are common knowledge to those skilled in the art.

Claims

1. A method for evaluating the impact of signal-to-noise ratio in satellite remote sensing payload design on inversion performance, characterized in that, include: Step (1): Construct the basic dataset, which includes the predicted radiance of the hyperspectral payload and the quantitative inversion values ​​of water environment elements; Step (2): Based on the basic dataset, the simulated radiance of the load under development is obtained through simulation, and a dataset of simulated radiance of the load under development and quantitative inversion values ​​of water environment elements is constructed. Step (3): Randomly sample the dataset of simulated radiance-quantitative inversion values ​​of water environment elements of the under-research load to obtain a training set and a test set. Train the neural network model based on the training set to obtain the trained neural network model. Step (4): Add a standard deviation of to each simulated radiance of the under-development load in the test set. After removing Gaussian noise, the data is input into the trained neural network model to obtain simulated quantitative inversion values. Step (5): Repeat step (4) N times to obtain N simulated quantitative inversion values, and calculate the standard deviation of the N simulated quantitative inversion values; Step (6): Change the signal-to-noise ratio of the Gaussian noise, repeat steps (4) to (5), and obtain the simulated quantitative inversion values ​​and standard deviations corresponding to different signal-to-noise ratios; Step (7): Based on the signal-to-noise ratio, simulated quantitative inversion value and standard deviation in step (6), calculate the signal-to-noise ratio application effectiveness impact factor and evaluate the impact of the signal-to-noise ratio on simulated quantitative inversion values ​​with different values; The specific method for calculating the signal-to-noise ratio application effectiveness impact factor in step (7) is as follows: Where E is the signal-to-noise ratio application effectiveness factor. To simulate the standard deviation of the quantitative inversion values, The standard deviation of Gaussian noise; Where N is the number of simulations; The mean of the quantitative inversion values ​​of water environment elements from N simulations; This provides the simulated quantitative inversion value for the neural network model.

2. The method for evaluating the impact of signal-to-noise ratio on inversion performance of satellite remote sensing payload design according to claim 1, characterized in that: The method for constructing the basic dataset in step (1) is as follows: S11. Select a hyperspectral payload and a high radiometric precision multispectral payload as reference payloads respectively. Based on the spatiotemporal and numerical matching principle, obtain data of synchronous or quasi-synchronous imaging of the hyperspectral payload and the reference payload in the same area. S12. Perform data processing and data matching on the data of the hyperspectral load and the data of the reference load to obtain quantitative inversion values ​​of water environment elements that match the numerical values ​​of the reference load data. S13. Based on the hyperspectral load and the reference load, perform band matching to obtain the matching result; S14. Based on the matching results and the atmospheric radiative transfer model, calculate the spectral matching factor between the hyperspectral load and the reference load; S15. Based on the spectral matching factor, and combined with the atmospheric top radiance of a pixel obtained from hyperspectral load observation, predict the predicted radiance that matches the observation conditions of the hyperspectral load and the reference load. S16. Based on the predicted radiance and the quantitative inversion values ​​of water environment elements, construct a basic dataset of predicted radiance and quantitative inversion values ​​of water environment elements for hyperspectral load.

3. The method for evaluating the impact of signal-to-noise ratio on inversion performance of satellite remote sensing payload design according to claim 2, characterized in that: In step S12, data processing and matching are performed on the data of the hyperspectral payload and the data of the reference payload to obtain quantitative inversion values ​​of water environment elements that match the numerical values ​​of the reference payload data. Specifically: S21: The spatial resolution of the hyperspectral payload data is sampled to be the same as that of the reference payload data to obtain the sampled hyperspectral payload data; S22: Using the operational algorithm for water environment elements with reference load, the quantitative inversion of regional water environment elements is realized, and the quantitative inversion values ​​of water environment elements are obtained. S23: Based on the range of values ​​of the quantitative inversion values ​​of water environment elements, the sampled hyperspectral load data is matched with the corresponding quantitative inversion values ​​at that point.

4. The method for evaluating the impact of signal-to-noise ratio on inversion performance of satellite remote sensing payload design according to claim 2, characterized in that: The spatiotemporal and numerical matching principles are as follows: The zenith angle shall not exceed 60°, and the solar altitude angle shall not exceed 75°; The reflectance of remote sensing in the visible light band is not negative; No clouds or shadows obstructing the view; Water environment element values ​​are , ,in , These represent the minimum and maximum values ​​of the quantitative inversion values ​​of regional water environment elements, respectively. The total number of water environment element values ​​selected for numerical matching. The sampling step size within the range of quantitative inversion values ​​of regional water environment elements. i The value range is 1~ .

5. The method for evaluating the impact of signal-to-noise ratio on inversion performance of satellite remote sensing payload design according to claim 1, characterized in that: In step (2), based on the basic dataset, the simulated radiance of the load under development is obtained through simulation, and a dataset of simulated radiance of the load under development and quantitative inversion values ​​of water environment elements is constructed, specifically as follows: Based on the band settings of the payloads under development, the predicted radiance of the hyperspectral payloads in the basic dataset mentioned in step (1) is convolved to obtain the equivalent band radiance. Based on the equivalent band radiance, the simulated radiance of the payload under development is obtained after modulation transfer function simulation. This simulated radiance, along with the corresponding quantitative inversion values ​​of water environment elements in the basic dataset, constitutes the dataset of simulated radiance of the payload under development - quantitative inversion values ​​of water environment elements.

6. The method for evaluating the impact of signal-to-noise ratio on inversion performance of satellite remote sensing payload design according to claim 5, characterized in that: The simulated radiance of the load under research, obtained after modulation transfer function simulation, is specifically as follows: In the formula, Indicates the simulated radiance. The Fourier transform of an image composed of equivalent band radiance. This is the modulation transfer function.

7. The method for evaluating the impact of signal-to-noise ratio on inversion performance of satellite remote sensing payload design according to claim 1, characterized in that: In step (4), a standard deviation is added to the simulated radiance of each under-development load in the test set. After removing Gaussian noise, the result is input into the trained neural network model to obtain the simulated quantitative inversion value, as shown in the formula: in, For the simulated quantitative inversion value of the neural network model, For neural network models, It is Gaussian noise; This represents the equivalent band radiance of the k-th band of the payload under development.

8. The method for evaluating the impact of signal-to-noise ratio on inversion performance of satellite remote sensing payload design according to claim 2, characterized in that: In step S15, the predicted radiance that matches the observation conditions of the hyperspectral payload with those of the reference payload is specifically as follows: in, The predicted radiance of the i-th band of the hyperspectral load, Let f be the sum of the predicted radiance of the i-th, i+1, ..., i+M-1 bands of the hyperspectral payload, and f be the spectral matching factor. The equivalent top atmospheric radiance is the weighted superposition of the i-th, i+1, ..., i+M-1 bands of the hyperspectral payload. These are the weighting coefficients for each band.

9. The method for evaluating the impact of signal-to-noise ratio on inversion performance of satellite remote sensing payload design according to claim 2, characterized in that: The spectral matching factor in step S14 is specifically as follows: in, This represents the equivalent top atmospheric radiance of the weighted superposition of bands i, i+1, ..., i+M-1 of the hyperspectral payloads, where i is the band number of the hyperspectral payload and M is the total number of bands of the hyperspectral payloads that satisfy the band matching principle. The reference load is the atmospheric top radiance in band j. These are the weighting coefficients for each band. For the i-th band of the hyperspectral payload, the wavelength is The relative spectral response, This is the sum of the relative spectral responses in that band; For the reference payload, the j-th band has a wavelength of The relative spectral response, The sum of the relative spectral responses in this band is given by f, where f is the spectral matching factor.

Images