Uncertainty evaluation method for atmospheric correction of hyperspectral remote sensing based on pixel-by-pixel and band-by-band

Abstract

In order to solve the quality assessment problem of hyperspectral remote sensing reflectance products, the present invention proposes a pixel-by-pixel and band-by-band hyperspectral remote sensing atmospheric correction uncertainty assessment method, conducts uncertainty assessment research on the atmospheric correction process of remote sensing data based on the radiation transmission model, sorts out the atmospheric correction uncertainty transmission chain, and establishes an uncertainty transfer model through the formula transfer method and the Monte Carlo method, so as to achieve quantitative evaluation of the uncertainty of the surface reflectance product with a confidence interval under a specified confidence level. The present invention can obtain the quantitative assessment results of the uncertainty of the surface reflectance on a pixel-by-pixel and band-by-band basis, provide users with quality information of the reflectance product, and support subsequent quantitative applications.

Description

Technical Field

[0001] The present invention relates to a pixel-by-pixel and band-by-band hyperspectral remote sensing atmospheric correction uncertainty assessment method, which can be used in the atmospheric correction process of airborne or satellite-borne hyperspectral remote sensing data, thereby serving the quality assessment of data obtained by remote sensors, and is a supporting technology in the field of earth observation. Background Art

[0002] Hyperspectral remote sensing can obtain a lot of very narrow and continuous spectral channel imaging data, thereby extracting rich spatial geometry and spectral information. The existing accuracy evaluation method in the field of quantitative remote sensing is generally to use the measured reflectance on the ground as a reference value, and to evaluate the accuracy level of the remote sensing inversion reflectance data by calculating the average error or root mean square error of multiple verification points. However, comparative verification with ground measured data is an a posteriori method. On the one hand, in most cases, there is no reference data of synchronous ground measurement for a posteriori verification; on the other hand, this method cannot propose effective improvement measures and optimization methods for the atmospheric correction method and input data accuracy. The accuracy of the surface reflectance inverted by atmospheric correction is determined by many factors, including the accuracy of input radiance data, the accuracy of atmospheric parameters, and the accuracy of inversion models and methods. In addition, hyperspectral data has high dimensionality, weak signal energy, long imaging links, and complex and variable influencing factors, resulting in unclear mechanisms for the instability of reflectance spectra. In addition, there is currently no complete theoretical explanation for the fluctuation of reflectance of the same object in multi-view data with overlapping imaging ranges. In practical applications, the phenomena of "same object, different spectrum" and "different objects, same spectrum" also show the uncertainty of spectral characteristics.

[0003] In order to evaluate the reliability of hyperspectral remote sensing reflectance results, it is necessary to conduct a predictive uncertainty assessment of reflectance products on the basis of traditional accuracy evaluation methods. Past studies have focused more on the acquisition, processing and application of hyperspectral remote sensing data, and less attention has been paid to the uncertainty assessment of remote sensing data. Limited by the support of a large number of computing resources and algorithms, the quality assessment of hyperspectral remote sensing data at home and abroad currently only stays on the qualitative quality identification of each pixel (such as cloud pixel identification, aerosol content, product quality status), and has not achieved quantitative uncertainty evaluation of each pixel and band, and has ignored the correlation between multiple parameters, making it difficult to meet users' needs for product quality. Therefore, in order to improve the accuracy, stability and reliability of hyperspectral remote sensing reflectance products, it is urgent to develop quantitative uncertainty assessment of remote sensing data on a pixel-by-pixel and band-by-band basis, provide users with quality information of reflectance products, and support algorithm improvement and subsequent quantitative application. Summary of the invention

[0004] In order to solve the problem of quality assessment of hyperspectral remote sensing reflectance products, the present invention proposes a pixel-by-pixel and band-by-band uncertainty assessment method for hyperspectral remote sensing atmospheric correction, conducts uncertainty analysis on the atmospheric correction process of hyperspectral remote sensing data based on the radiation transfer model, sorts out the uncertainty transfer chain of atmospheric correction, and establishes an uncertainty transfer model through the formula transfer method and the Monte Carlo method, so as to realize quantitative evaluation of the uncertainty of surface reflectance products with confidence intervals under specified confidence levels.

[0005] A method for evaluating uncertainty of hyperspectral remote sensing atmospheric correction on a pixel-by-pixel and band-by-band basis, characterized in that it comprises the following steps:

[0006] Step 1: Establish an atmospheric correction model to complete the atmospheric correction value transfer from the remote sensor's hyperspectral remote sensing radiance image to the surface reflectance image. The core formula of atmospheric correction is:

[0007]

[0008] in, L is the entrance pupil radiance, is the target pixel reflectance, is the average reflectivity of the background objects, is the downward hemispheric albedo of the atmosphere, For the process radiation, A and B is the coefficient, the uncertainty introduced by the simplified calculation model, the error of the radiation transfer model and the flat surface Lambertian assumption is represented by “+ 0”. This formula assumes that the entrance pupil radiance is composed of three components, namely the part reflected by the target pixel directly entering the sensor, the part reflected by the background object entering the sensor, and the path radiation.

[0009] Step 2: Starting from the atmospheric correction model, trace the uncertainty of each variable in the model to the input variable and draw the uncertainty transfer link;

[0010] Step 3: For each input variable obtained by tracing back in step 2, uncertainty assessment is performed, and variable uncertainty is quantitatively evaluated by searching literature and user manuals;

[0011] Step 4: Use the uncertainty assessment results of the input variables obtained in step 3 to transfer and synthesize the uncertainty. For the links with analytical formulas, the uncertainty transfer of the formula transfer method is realized. For the links that call the radiation transfer model but have no analytical formula, the uncertainty transfer of the Monte Carlo method is realized. Considering the correlation between variables, the uncertainty assessment results of the surface reflectivity synthesis standard are calculated.

[0012] Step 5: Determine the expanded uncertainty. The expanded uncertainty is obtained by multiplying the composite standard uncertainty by the coverage factor, and finally the confidence interval of the reflectance product is given pixel by pixel and band by band.

[0013] Among them, the sources of uncertainty in step 3 include: input radiance data, atmospheric parameters of aerosol optical depth and water vapor content, radiation transfer model, central longitude and latitude, sensor altitude, observation zenith angle, azimuth, imaging time, image spatial resolution, and average ground elevation parameters of the image area.

[0014] Among them, in step 4, the uncertainty of the radiance data is The uncertainty is obtained by searching the literature related to the radiation accuracy of the hyperspectral payload and combining the formula transfer method with the Monte Carlo method. , , , , And uncertainty related items, and then calculate the formula for the composite standard uncertainty of reflectivity as follows:

[0015]

[0016] is the entrance pupil radiance uncertainty, is the uncertainty of the target pixel reflectance, is the uncertainty of the average reflectivity of the background objects, is the uncertainty in the downward hemispheric albedo of the atmosphere, is the uncertainty of the process radiation, and They are A and B uncertainty. For the Uncertainty Components and Uncertainty Components The correlation coefficient between .

[0017] In step 5, the expanded uncertainty The combined standard uncertainty Multiply the inclusion factor The calculation formula is as follows:

[0018]

[0019] The inclusion probability is p The inclusion factor at that time is finally obtained, and the confidence interval of the reflectance product is obtained pixel by pixel and band by band.

[0020] Beneficial effects achieved by the present invention:

[0021] (1) An uncertainty transfer model for atmospheric correction of hyperspectral remote sensing was established, and the uncertainty of surface reflectance was quantitatively evaluated by formula transfer method and Monte Carlo method. An uncertainty transfer analysis method for atmospheric correction was proposed.

[0022] (2) The quantitative uncertainty assessment results of the surface reflectance are obtained for each pixel and band, which can provide users with quality information of reflectance products, support algorithm improvement, locate the main influencing factors, guide the optimization of system indicators, thereby improving data accuracy and providing support for the quantitative application of remote sensing products. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 Flow chart for uncertainty assessment of atmospheric correction;

[0024] Figure 2 To correct the uncertainty transfer chain for the atmosphere;

[0025] Figure 3 The uncertainty evaluation results of the reflectance product of the Dunhuang calibration field and the pixel-by-pixel and band-by-band of the 5-meter 01 satellite hyperspectral image are given. . DETAILED DESCRIPTION

[0026] A method for evaluating uncertainty of hyperspectral remote sensing atmospheric correction on a pixel-by-pixel and band-by-band basis, characterized in that it comprises the following steps:

[0027] Step 1: Establish an atmospheric correction model to complete the atmospheric correction value transfer from the remote sensor's hyperspectral remote sensing radiance image to the surface reflectance image;

[0028] Specifically, use the following formula:

[0029]

[0030] in, L is the entrance pupil radiance, is the target pixel reflectance, is the average reflectivity of the background objects, is the downward hemispheric albedo of the atmosphere, For the process radiation, A and BThe coefficient is denoted by "+ 0" for the uncertainty introduced by the simplified calculation model, the error of the radiation transfer model and the flat surface Lambertian assumption. In this formula, the entrance pupil radiance is considered to be composed of three components, namely, the part of the target pixel reflection directly entering the sensor, the part of the background object reflection radiation entering the sensor, and the path radiation. By using the radiation transfer model to calculate the atmospheric parameters of water vapor and aerosol content, and using the radiation transfer model to simulate the radiance data under known atmospheric parameters, and substituting it into the above formula, the radiance under different atmospheric parameters can be obtained. and , considering the proximity effect, we can find , you can get the value of surface reflectivity.

[0031] Step 2: Starting from the atmospheric correction model, trace the uncertainty of each variable in the model to the input variable and draw the uncertainty transfer link;

[0032] Specifically, the uncertainty is traced and the uncertainty transfer chain is drawn. Starting from the atmospheric correction model, the uncertainty of each variable in the model is traced back to the input variable, such as Figure 2 Shown by: Figure 2 The uncertainty transfer chain shows that the uncertainty of reflectivity can first be traced back to the radiance input data. , spatial average radiance , atmospheric parameters and . The radiance value of the image can be calculated by the point spread function (which describes the contribution of neighboring pixels to the radiance at the entrance pupil). Do convolution to get, and They are all calculated by the radiation transfer model, and then further traced back to the uncertainty of the radiance data, water vapor content, aerosol content, central longitude and latitude, sensor altitude, observation zenith angle and azimuth, imaging time, image spatial resolution, and average ground elevation parameters of the image area.

[0033] Step 3: For each input variable obtained by tracing back in step 2, uncertainty assessment is performed, and variable uncertainty is quantitatively evaluated by searching literature and user manuals;

[0034] The uncertainty quantification of input radiance and imaging geometry parameters is determined by searching the literature, the uncertainty level of the radiation transfer model simulation calculation is determined by the literature and user manual, and the uncertainty of the atmospheric parameters of aerosol optical depth and water vapor content is quantified by Monte Carlo simulation.

[0035] Step 4: Use the uncertainty assessment results of the input variables obtained in step 3 to transfer and synthesize the uncertainty. The formula transfer method is used to transfer the uncertainty of the links with analytical formulas, and the Monte Carlo method is used to transfer the uncertainty of the links that call the radiation transfer model but do not have analytical formulas. Considering the correlation between variables, the synthetic uncertainty assessment results of the surface reflectance are obtained pixel by pixel and band by band.

[0036] Specifically, the radiance uncertainty By searching the literature on the radiometric accuracy of the hyperspectral payload, the atmospheric parameters were determined. and The uncertainty calculation of the atmospheric parameters needs to be evaluated using the Monte Carlo method. By determining the input of the atmospheric parameter calculation process: aerosol content, water vapor content, central longitude and latitude, sensor height, observation zenith angle, azimuth, imaging time, image spatial resolution, and image area ground average elevation parameters, the Monte Carlo method is used to simulate the propagation. For example, if 10,000 atmospheric parameter calculations are set, the standard deviation of the 10,000 sets of atmospheric parameters obtained by the output is calculated separately to obtain the atmospheric parameter uncertainty. as well as . The uncertainty of The input and output calculation process is simulated and the obtained The standard deviation of multiple results is According to the uncertainty transfer formula, we can get Uncertainty :

[0037]

[0038] The final formula for calculating the composite standard uncertainty of reflectivity is as follows:

[0039]

[0040] is the entrance pupil radiance uncertainty, is the uncertainty of the target pixel reflectance, is the uncertainty of the average reflectivity of the background objects, is the uncertainty in the downward hemispheric albedo of the atmosphere, is the uncertainty of the process radiation, and They are A and B uncertainty. For the Uncertainty Components and Uncertainty Components The correlation coefficient between .

[0041] Step 5: Determine the expanded uncertainty. The expanded uncertainty is obtained by multiplying the composite standard uncertainty by the coverage factor, and finally gives the confidence interval of the reflectance product pixel by pixel and band by band. The expanded uncertainty is obtained by multiplying the composite uncertainty by the coverage factor. The combined uncertainty Multiply the inclusion factor The calculation formula is as follows:

[0042]

[0043] The inclusion probability is p The inclusion factor when .

[0044] In order to illustrate the proposed uncertainty transfer assessment method for atmospheric correction, the Dunhuang calibration site in Gansu Province was selected as the test area to carry out uncertainty transfer assessment. Figure 3 The uncertainty assessment results of the reflectance product of the Dunhuang calibration field of the 5-meter 01-satellite hyperspectral image and the pixel-by-pixel and band-by-band are given.

Claims

1. A pixel-by-pixel and band-by-band hyperspectral remote sensing atmospheric correction uncertainty assessment method, characterized in that: The following steps are involved: Step 1: Establish an atmospheric correction model to complete the atmospheric correction value transfer from the remote sensor's hyperspectral remote sensing radiance image to the surface reflectance image. The core formula of atmospheric correction is: ； in, L is the entrance pupil radiance, is the target pixel reflectance, is the average reflectivity of the background objects, is the downward hemispheric albedo of the atmosphere, For the process radiation, and is the coefficient, and the uncertainty introduced by the simplified calculation model, the error of the radiation transfer model and the flat surface Lambertian assumption is represented by "+ 0". This formula assumes that the entrance pupil radiance consists of three components, namely, the part reflected by the target pixel directly entering the sensor, the part reflected by the background object entering the sensor, and the path radiation; Step 2: Starting from the atmospheric correction model, trace the uncertainty of each variable in the model to the input variable and draw the uncertainty transfer link; Step 3: For each input variable obtained by tracing back in step 2, uncertainty assessment is performed, and variable uncertainty is quantitatively evaluated by searching literature and user manuals; Step 4: Use the uncertainty assessment results of the input variables obtained in step 3 to transfer and synthesize the uncertainty. The uncertainty transfer of the formula transfer method is realized for the links with analytical formulas, and the uncertainty transfer of the Monte Carlo method is realized for the links that call the radiation transfer model but have no analytical formulas. Considering the correlation between variables, the uncertainty assessment results of the surface reflectivity synthesis standard are calculated, specifically the uncertainty of the radiance data. The uncertainty is obtained by searching the literature related to the radiation accuracy of the hyperspectral payload and combining the formula transfer method with the Monte Carlo method. , , , , And uncertainty related items, and then calculate the formula for the composite standard uncertainty of reflectivity as follows: ； is the entrance pupil radiance uncertainty, is the uncertainty of the target pixel reflectance, is the uncertainty of the average reflectivity of the background objects, is the uncertainty in the downward hemispheric albedo of the atmosphere, is the uncertainty of the process radiation, and They are and The uncertainty of For the Uncertainty Components and Uncertainty Components The correlation coefficient between Step 5: Determine the expanded uncertainty. The expanded uncertainty is obtained by multiplying the composite standard uncertainty by the coverage factor, and finally the confidence interval of the reflectance product is given pixel by pixel and band by band.

2. The method according to claim 1, characterized in that The sources of uncertainty in step 3 include: input radiance data, atmospheric parameters of aerosol optical depth and water vapor content, radiation transfer model, central longitude and latitude, sensor altitude, observation zenith angle, azimuth, imaging time, image spatial resolution, and average ground elevation parameters of the image area.

3. The method according to claim 1, characterized in that In step 5, the expanded uncertainty The combined standard uncertainty Multiply the inclusion factor The calculation formula is as follows: ； The inclusion probability is p The inclusion factor at that time is finally obtained, and the confidence interval of the reflectance product is obtained pixel by pixel and band by band.

Images