Multi-spatial resolution remote sensing simulation and error influence stripping method

Abstract

The invention discloses a multi-spatial resolution remote sensing simulation and error influence stripping method, and relates to the technical fields of remote sensing image processing, quantitative remote sensing, environment modeling and the like. The method comprises the following steps: firstly, simulating a multi-scale remote sensing image in a mode of combining a modulation transfer function (MTF) and mean interpolation; secondly, constructing a multiple regression model, explaining a model error by using an image quality index, thereby stripping the interference of the image quality on an analysis result, and obtaining a normalized residual error only caused by the change of a spatial resolution ontology; and finally, an RSI and Median + IQR scoring method based on a sliding window is provided, a resolution stable interval is identified, and next analysis of an experiment is performed so as to guide selection of an actual modeling scale. The method has high universality and scale adaptability, and is suitable for early-stage scale analysis and optimization modeling in various remote sensing quantitative analysis tasks such as land coverage, vegetation parameters, urban heat islands and water body monitoring.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of remote sensing image processing and quantitative remote sensing analysis, and particularly relates to a multi-spatial resolution remote sensing simulation and error influence stripping method. BACKGROUND

[0002] Remote sensing data, as an important means of spatial information acquisition, plays a key role in the fields of environmental monitoring, land use classification, vegetation coverage estimation and water quality inversion. However, the choice of spatial resolution of remote sensing images often faces a fundamental trade-off: high-resolution images can provide more detailed information about ground objects, but they are accompanied by huge data volume and computational burden; while low-resolution images inevitably lose image details and introduce mixed pixel errors, although they improve processing efficiency. This resolution-dependent effect can significantly affect the feature extraction, prediction accuracy and even the interpretability of the results of downstream analysis tasks. Therefore, in many research fields such as environmental monitoring and resource investigation, it is necessary to systematically analyze the influence of spatial scale on the research results to determine the optimal observation scale.

[0003] Currently, the research on the influence of spatial scale mainly adopts an empirical comparative analysis method, which evaluates the scale effect by directly comparing the differences in results under different resolutions. This method has obvious limitations: first, it cannot distinguish between the resolution change itself (i.e. scale effect) and the independent contribution of image quality degradation to the results; second, it lacks a standardized resolution simulation process, making it difficult to directly compare results between different studies; most importantly, the coupling effect of image quality degradation and modeling error can interfere with the evaluation of the true scale influence. SUMMARY

[0004] To solve the above technical problems, the present application provides a multi-spatial resolution remote sensing simulation and error influence stripping method, which first constructs a high-fidelity simulated multi-resolution image sequence, then strips the "image damage effect" in the task error with the help of image quality indicators, and finally forms an independent residual sequence reflecting the intrinsic effect of spatial resolution, thereby providing a unified and reliable scale research basis for various remote sensing analysis tasks.

[0005] The technical scheme of the present application is as follows:

[0006] A multi-spatial resolution remote sensing simulation and error influence stripping method, comprising the following steps:

[0007] S1: performing blur processing on the original high-resolution remote sensing image to construct a multi-spatial scale remote sensing image sequence;

[0008] S2: Apply the inversion model to the remote sensing image sequence and quantify its prediction error; At the same time, introduce an image quality index to depict the error source affecting the reconstruction, separate the degradation error in the image reconstruction process from the total prediction error, and obtain the influence of the scale on the inversion model;

[0009] S3: Set a sliding window, calculate the residual sensitivity index RSI in the window, identify the sensitivity change rule of the inversion variable of the inversion model to the resolution, and then determine the optimal spatial scale interval by combining the error minimization and stability principle.

[0010] Further, the step S1 specifically comprises the following steps: S11: Taking the remote sensing image with a spatial resolution of R0 as the original data, performing radiation correction, atmospheric correction and geometric correction; S12: According to the modulation transfer function MTF of the remote sensing sensor, the spatial degradation process of the image is simulated, and the image obtained in S11 is blurred by the equivalent Gaussian convolution kernel to obtain a blurred image. The formula of the Gaussian kernel standard deviation is as follows: ; Wherein, is the standard deviation of the equivalent Gaussian convolution kernel, is the modulation transfer function value of the used remote sensing sensor at the Nyquist frequency, is the spatial resolution of the original image; The formula of the Gaussian standard deviation is as follows: ; The formula of the blurred convolution is as follows: ; ; Wherein, is the value of the first pixel of the blurred image, is the value of the first pixel of the original image, is the weight of the Gaussian convolution kernel, k is the convolution kernel radius, and k is taken as , , , , respectively represent the convolution kernel index offset; S13: Resample the blurred image to multiple target resolutions to obtain a remote sensing image sequence of multiple spatial scales.

[0011] Further, the step S2 specifically comprises the following steps: S21: For each scale of remote sensing image, call the remote sensing analysis model to output the parameter estimation value; S22: According to the high-resolution remote sensing image The obtained output results are taken as pseudo-true values, and the error indicators corresponding to all prediction results at each spatial resolution are calculated, and the mean absolute percentage error (MAPE) is used to represent the error indicators: ; wherein, is the number of inversion values at a certain scale, is the i-th pseudo-true value, is the i-th inversion value; S23: The peak signal-to-noise ratio (PSNR), structural similarity (SSIM), and spectral angle (SA) values between the remote sensing image sequence at each scale and the original image sequence are calculated respectively, which are used to represent the image quality change: The calculation formula of PSNR is as follows: ; ; wherein, is the mean square error, represents the maximum possible pixel value in the image, and are the pixel values of the original image and the image at each scale at position , and are the row number and column number of the image; The calculation formula of SSIM is as follows: ; wherein, and represent the local mean values of the original image and the image at each scale, and represent the local standard deviations of the original image and the image at each scale, is the covariance, is a constant; The calculation formula of SA is as follows: ; wherein, , is the dot product of two vectors; is the length of the vector; and are spectral vectors, each element or represents the spectral reflectance or radiance value of a wave band; S24: A multiple regression model is constructed, and the residual error is extracted as an error component other than the image quality: ; wherein, This represents the expected error value predicted based on image quality metrics at the i-th resolution scale. , , , representing the peak signal-to-noise ratio, structural similarity, and spectral angle at the i-th resolution scale, respectively, and a, b, c, and d are the coefficients of the regression model; ; in, It is the true, total prediction error observed at the i-th resolution scale. It is the residual at the i-th resolution scale; S25: Normalize the residuals to the [0,1] interval and plot the residual sensitivity curve as a function of resolution; ; in, The normalized residuals, It is the minimum residual value across all scales. It is the maximum residual value across all scales.

[0012] Furthermore, step S3 specifically includes the following steps: S31: Using each scale point as the center, set a fixed-size sliding window, and calculate the residual sensitivity index (RSI) and the comprehensive score based on Median + IQR within the window: ; ; in, Representing the The overall score of the sliding window Represents the set of normalized residuals within the sliding window. This represents the median of the normalized residuals within the sliding window. The interquartile range represents the normalized residuals within the sliding window; S32: According to and Constructing a dual-criteria decision-making strategy, firstly based on The numerical values ​​determine a set of candidate windows such that the average error level of the candidate intervals is low, and then a selection is made from the candidate windows. The smallest window.

[0013] Furthermore, in step S21, the remote sensing analysis model is a regression, classification, or other estimation algorithm.

[0014] Further, in step S21, the remote sensing analysis model can be a prediction model of normalized vegetation index NDVI, leaf area index LAI, ground temperature Ts or water quality parameters such as chlorophyll a Chl-a.

[0015] Further, the residual sensitivity index RSI is defined as the mean value of the normalized residual curve in a window of a specified size, which is used to quantify the independent influence of resolution on model error.

[0016] Further, the optimal resolution interval refers to the RSI being low and the Score based on Median+IQR being the lowest, that is, a balance point that satisfies small deviation and small fluctuation at the same time is found.

[0017] Further, in S13, the resampling selects any interpolation method to reconstruct the original image into multiple spatial resolution scales.

[0018] Further, in S13, the target resolution is any equidistant resolution interval required for the study.

[0019] The beneficial effects of the present application are as follows:

[0020] (1) The "simulation-peeling-analysis" integrated process is proposed, the image quality peeling method is introduced to analyze the remote sensing modeling error source, and the theoretical mechanism is supported;

[0021] (2) The residual sensitivity index RSI and the score criterion based on "Median+IQR" are proposed, which has quantifiable and comparable resolution evaluation capability;

[0022] (3) The sliding window evaluation strategy realizes the automatic identification of the error smooth interval, which is suitable for multi-parameter and full-scale data;

[0023] (4) It supports flexible nesting of different remote sensing tasks, data types and model forms, and is suitable for promotion to most remote sensing quantitative analysis processes. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 is the technical process diagram of the present application;

[0025] Figure 2 is the trend image of the PSNR of the pure water body of the embodiment of the present application with the change of spatial resolution;

[0026] Figure 3 is the trend image of the SSIM of the pure water body of the embodiment of the present application with the change of spatial resolution;

[0027] Figure 4 is the trend image of the SA of the pure water body of the embodiment of the present application with the change of spatial resolution;

[0028] Figure 5 is a trend image of MAPE of Chl-a of the embodiment of the application varying with spatial resolution;

[0029] Figure 6 is a trend image of normalized residual of Chl-a of the embodiment of the application varying with spatial resolution;

[0030] Figure 7 is a sliding window score map of the embodiment of the application based on "Median + IQR". DETAILED DESCRIPTION

[0031] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other. In order to achieve the above-mentioned purpose, the technical solutions adopted by the present application are as follows.

[0032] The present application provides a multi-spatial resolution remote sensing simulation and error influence stripping method, comprising the following steps:

[0033] S1: performing blur processing (which can be blur processing based on modulation transfer function (MTF), but is not limited thereto) on the original high-resolution remote sensing image, so that the output image is closer to the data obtained by the real remote sensing imaging system in terms of visual and spectral characteristics, and then constructing a remote sensing image sequence with different spatial resolutions;

[0034] S2: applying an existing inversion model (which can be a water quality parameter inversion model, but is not limited thereto) to the multi-spatial scale image, and quantifying the prediction error thereof; at the same time, introducing an image quality index to depict the error source of image reconstruction, stripping the degradation error in the image reconstruction process from the total prediction error, so as to obtain the influence of scale on the inversion model;

[0035] S3: setting a sliding window, calculating the residual sensitivity index RSI in the window, identifying the sensitivity change rule of the inversion variable to the resolution, and then determining the optimal spatial scale interval in combination with the error minimization and stability principle.

[0036] Further, the step S1 specifically comprises the following steps:

[0037] S11: taking the remote sensing image with spatial resolution R0 as the original data, performing radiation correction, atmospheric correction and geometric correction;

[0038] S12: According to the modulation transfer function (MTF) of the remote sensing sensor, a spatial degradation process for simulating the image is used to blur the image obtained in S11 through an equivalent Gaussian convolution kernel. The standard deviation of the Gaussian kernel can be determined according to the sensor MTF parameter and the original image resolution, so as to simulate the spatial degradation characteristics of the remote sensing imaging system. The MTF parameter can be obtained through the sensor design document, experimental measurement or on-orbit calibration, and is preferably the MTF value of the sensor at the Nyquist frequency. .

[0039] The formula of the Gaussian kernel standard deviation is as follows:

[0040] ;

[0041] wherein, is the standard deviation (unit: m) of the equivalent Gaussian convolution kernel, is the modulation transfer function value (unitless, 0~1) of the remote sensing sensor used at the Nyquist frequency, is the spatial resolution of the original image (unit: m).

[0042] The formula of the Gaussian standard deviation (pixel unit) is as follows:

[0043] ;

[0044] The formula of the blur convolution is as follows:

[0045] ;

[0046] ;

[0047] wherein, is the value of the pixel of the blurred image, is the value of the pixel of the original image, is the weight of the Gaussian convolution kernel, k is the convolution kernel radius, and is usually 3 , and m and n represent the convolution kernel index offset, respectively. S13: Resample the blurred image into multiple target resolutions to obtain a multi-scale remote sensing image sequence.

[0048] Further, the step S2 specifically includes the following steps:

[0049] S21: For each scale of remote sensing image, call the remote sensing analysis model to output parameter estimation;

[0050] S22: According to the parameter estimation of the high-resolution image

[0051] S22: According to the parameter estimation of the high-resolution image ​The obtained output results are taken as pseudo-true values, and the error indicators corresponding to all prediction results at each spatial resolution are calculated, and the mean absolute percentage error (MAPE) is used to represent the error indicators:

[0052] ;

[0053] wherein N is the number of inversion values at a certain scale, is the i-th pseudo-true value, is the i-th inversion value.

[0054] S23: The peak signal-to-noise ratio (PSNR), structural similarity (SSIM), and spectral angle (SA) values between each scale image and the original image are calculated respectively to represent the image quality change:

[0055] The calculation formula of PSNR is as follows:

[0056] ;

[0057] ;

[0058] wherein, is the mean square error, represents the maximum possible pixel value in the image, and are the pixel values of the original image and each scale image at position , m and n are the row number and column number of the image;

[0059] The calculation formula of SSIM is as follows:

[0060] ;

[0061] wherein, and represent the local mean values of the original image and each scale image, and represent the local standard deviations of the original image and each scale image, is the covariance, is a constant;

[0062] The calculation formula of SA is as follows:

[0063] ;

[0064] wherein, and are spectral vectors, each element or represents the spectral reflectance or radiance value of a wave band, is the dot product of the two vectors; the length of the vector; the range of 0° to 90°, the smaller the angle, the higher the spectral similarity;

[0065] S24: Construct a multiple regression model, and extract the residual error as an error component other than image quality:

[0066] ;

[0067] wherein, represents the expected error value predicted based on the image quality index at the i-th resolution scale, , , respectively represent the peak signal-to-noise ratio, the structural similarity and the spectral angle at the i-th resolution scale, and a, b, c, d are the coefficients of the regression model.

[0068] ;

[0069] wherein, is the real, total prediction error observed at the i-th resolution scale, is the residual error at the i-th resolution scale.

[0070] S25: Normalize the residual error to the interval [0, 1], and draw the residual sensitivity curve with respect to the resolution;

[0071] ;

[0072] wherein, is the normalized residual error, is the minimum residual error in all scales, is the maximum residual error in all scales.

[0073] Further, the step S3 specifically comprises the following steps:

[0074] S31: Set a fixed-size sliding window centered on each scale point, and calculate the residual sensitivity index RSI and the comprehensive score Score based on "Median + IQR" in the window:

[0075] ;

[0076] ;

[0077] wherein, represents the comprehensive score of the i-th sliding window, represents the normalized residual error set in the sliding window, ​Median + IQR IQR

[0078] S32: According to and the "Median + IQR" based The double-criteria decision strategy is constructed, first according to The numerical value of some candidate window set is determined, which guarantees that the average error level of the candidate interval is low, and then the The smallest window is selected from the candidate window. This criterion has the dual advantages of low average error and high stability.

[0079] Further, in step S21, the remote sensing analysis model is regression, classification or other estimation algorithm.

[0080] Further, in step S21, the remote sensing analysis model can be a prediction model of normalized vegetation index (NDVI), leaf area index (LAI), land surface temperature (Ts), or water quality parameters such as chlorophyll a (Chl-a).

[0081] Further, the residual sensitivity index RSI is defined as the mean of the normalized residual curve in a window of a specified size, which is used to quantify the independent influence of resolution on model error.

[0082] Further, the optimal resolution interval refers to the lowest RSI and the lowest "Median + IQR" based Score, that is, to find a balance point that satisfies "small deviation" and "small fluctuation" at the same time.

[0083] Further, in S13, the resampling can select any interpolation method to reconstruct the original image into multiple spatial resolution scales.

[0084] Further, in S13, the target resolution is any equidistant resolution interval required for the study.

[0085] As Figure 1 shown, the present application proposes a multi-spatial resolution remote sensing simulation and error influence stripping method, taking the Chl-a inversion task of small and medium-sized lakes as an example, the core idea of which is: on the basis of existing high-resolution remote sensing images, simulate to generate a remote sensing image sequence covering multiple spatial resolutions, through the construction of image quality and model error stripping mechanism, quantitatively identify the independent influence of spatial resolution on model performance, and then select the most suitable modeling resolution interval:

[0086] S1: Based on the system MTF characteristics, a spatial degradation model is constructed, and different spatial resolution remote sensing image sequences are generated through Gaussian blur and resampling method, to ensure that each scale image has the consistency of real imaging process.

[0087] S2: Construct a consistent Chl-a inversion model on multi-scale images, and quantify the prediction error, while introducing image quality indicators (such as PSNR, SSIM and SA) to model error sources, and using regression method to separate the "image degradation error" from the total error, and obtain a more pure "scale influence" component.

[0088] S3: Normalize and trend analysis of the residual error at each scale, identify the sensitivity change rule of different water quality parameters to the resolution, and then determine the optimal spatial scale interval by combining the error minimality and stability principle, to provide a basis for actual remote sensing monitoring resolution selection.

[0089] Further preferred embodiments of the application are that the step S1 specifically comprises the following steps:

[0090] S11: Taking the airborne hyperspectral remote sensing image with a spatial resolution of 0.75m as the original data, performing radiation correction, atmospheric correction, geometric correction, band resampling and other data preprocessing;

[0091] S12: According to the Sentinel-2 MSI modulation transfer function (MTF) model, the spatial blur characteristics of the imaging system are simulated. The image is processed using a Gaussian convolution kernel, and the kernel function standard deviation According to the system response characteristics of Sentinel-2 MSI sensor at different bands, and through the MTF value of the sensor at Nyquist frequency , a two-dimensional Gaussian blur kernel is constructed, and the original image is convolved as a whole to realize spatial degradation processing, so as to obtain the blurred image.

[0092] ;

[0093] S13: The blurred image is resampled to multiple target resolutions (1.5-120m, interval 0.75m) using the mean interpolation method, to obtain a multi-scale image sequence.

[0094] Further preferred embodiments of the application are that the step S2 specifically comprises the following steps:

[0095] S21: For each scale image, call the established Chl-a remote sensing inversion model;

[0096] S22: Take the original high-resolution image model output result as the "pseudo true value", and calculate the average absolute percentage error (MAPE) corresponding to each scale;

[0097] S23: Calculate the PSNR, SSIM and SA values between each scale image and the original image, respectively, to represent the image quality change;

[0098] S24: Construct a multiple regression model (PSNR+SSIM+SA→MAPE), and extract the residual error as an error component other than image quality;

[0099] S25: Normalize the residual error to the interval [0, 1], and draw the residual sensitivity curve with the change of resolution.

[0100] A further preferred embodiment of the application is that the step S3 specifically comprises the following steps:

[0101] S31: Set a sliding window, and calculate the residual sensitivity index RSI and the comprehensive score Score based on "Median + IQR" in the window;

[0102] S32: Select the interval with the lowest score as the optimal modeling resolution recommendation interval under the analysis task in the set with a lower residual sensitivity index.

[0103] Embodiment:

[0104] In this embodiment, a general scale simulation and error stripping analysis method for quantitative remote sensing is used to select the best spatial resolution interval for the remote sensing inversion of Chl-a concentration in Dianshan Lake and Yuandang Lake region. First, the 0.75m resolution airborne hyperspectral image obtained in June 2022 is selected as the reference data. Then, based on the MTF characteristics of Sentinel-2 MSI, a Gaussian blur kernel is constructed to generate a simulation image sequence with 160 gradients from 0.75m to 120m (interval 0.75m). The experimental results show that: (1) Image quality indicators: PSNR and SSIM values decrease exponentially with the decrease of resolution, as shown in Figure 2 、 Figure 3 , and the SA value increases exponentially, as shown in Figure 4 , which is consistent with the expected spatial degradation theory; (2) Inversion error characteristics: the MAPE of Chl-a concentration shows a linear growth trend, as shown in Figure 5 , indicating that the decrease of resolution directly leads to the decrease of inversion accuracy; (3) Error sensitivity analysis: as shown in Figure 6 , the normalized residual curve shows significant fluctuations in the entire interval, revealing the important influence of scale effect in this interval. Finally, a sliding window with a window width of 20 meters is used for analysis. First, the candidate interval with low average error is selected according to the residual sensitivity index (RSI), and then the optimal spatial resolution interval of 18.2-38.2 meters is determined based on the minimum principle of comprehensive score (Score = Median + IQR), as shown in Figure 7 the black stripe window in the middle.

[0105] It should be noted that the technical solutions of the present application are not limited to the above specific embodiments, and any technical variations made according to the technical solutions of the present application fall within the scope of protection of the present application.

Claims

1. A multi-spatial resolution remote sensing simulation and error influence stripping method, characterized in that, The method comprises the following steps: S1: blurring the original high-resolution remote sensing image to construct a remote sensing image sequence of multiple spatial scales; S2: applying an inversion model to the remote sensing image sequence and quantifying the prediction error thereof; meanwhile, introducing an image quality index to depict the error source affecting the reconstruction, separating the degradation error in the image reconstruction process from the total prediction error, so as to obtain the influence of the scale on the inversion model; S3: setting a sliding window, calculating the residual sensitivity index RSI in the window, identifying the sensitivity change rule of the inversion variable of the inversion model to the resolution, and then determining the optimal spatial scale interval in combination with the error minimization and stability principles.

2. The method of claim 1, wherein, The step S1 specifically comprises the following steps: S11: taking the remote sensing image with a spatial resolution of R0 as the original data, and performing radiation correction, atmospheric correction and geometric correction; S12: blurring the image obtained in S11 by fuzzy convolution of an equivalent Gaussian kernel to obtain a blurred image, according to the modulation transfer function MTF of the remote sensing sensor for simulating the spatial degradation process of the image; the formula of the Gaussian standard deviation is as follows: ； wherein, is the standard deviation of the equivalent Gaussian kernel, is the modulation transfer function value of the used remote sensing sensor at the Nyquist frequency, is the original image spatial resolution; The formula of the Gaussian standard deviation is as follows: ； The formula of the fuzzy convolution is as follows: ； ； wherein, is the value of the pixel of the blurred image, is the value of the pixel of the original image, is the value of the pixel of the blurred image, is the value of the pixel of the original image, is the weight of the Gaussian kernel, k is the radius of the kernel, and k = 3 , , respectively represent the index offset of the kernel. S13: resampling the blurred image into multiple target resolutions to obtain a remote sensing image sequence of multiple spatial scales.

3. The method of claim 2, wherein, The step S2 specifically comprises the following steps: S21: calling a remote sensing analysis model for each scale of remote sensing image to output parameter estimates; S22: obtaining the high-resolution remote sensing image The output result is taken as pseudo true value, and the error index corresponding to all prediction results at each spatial resolution is calculated, and the mean absolute percentage error MAPE is used to represent: ； wherein, is the number of inversion values for a certain scale, is the ith pseudo true value, is the ith inversion value; S23: calculating the peak signal-to-noise ratio PSNR, the structural similarity SSIM and the spectral angle SA between the remote sensing image sequence of each scale and the original image sequence, respectively, to represent the image quality change: The formula for calculating the PSNR is as follows: ； ； wherein, is the mean square error, represents the maximum possible pixel value in the image, and are the pixel values of the original image and the image at scale s, respectively, at position , and are the number of rows and columns of the image, respectively. The formula for calculating the SSIM is as follows: ； wherein, and denote the local mean of the original image and the images at each scale, and denote the local standard deviation of the original image and the images at each scale, is the covariance, is a constant; The formula for calculating the SA is as follows: ； wherein, , is the dot product of two vectors; is the length of a vector; and are spectral vectors, each element or represents a spectral reflectance or radiance value for a band. S24: constructing a multiple regression model and extracting the residual error as an error component other than the image quality: ； wherein, represents the expected error value predicted based on the image quality index at the i-th resolution scale, , , respectively represent the peak signal-to-noise ratio, the structural similarity and the spectral angle at the i-th resolution scale, and a, b, c, d are the coefficients of the regression model. ； wherein, is the real, total prediction error observed at the i-th resolution scale, is the residual at the i-th resolution scale; S25: normalizing the residual error to the interval [0, 1] and drawing a residual sensitivity curve varying with the resolution; ； wherein, is the normalized residual, is the minimum residual value across all scales, is the maximum residual value across all scales.

4. The multi-spatial resolution remote sensing simulation and error influence stripping method according to claim 3, characterized in that, The step S3 specifically comprises the following steps: S31: setting a fixed-size sliding window centered on each scale point, and calculating the residual sensitivity index RSI and the comprehensive score Score based on Median + IQR in the window: ； ； wherein, representing the integrated score of the representing the set of normalized residuals within the sliding window, representing the median of the normalized residuals within the sliding window, representing the interquartile range of the normalized residuals within the sliding window;​ S32: According to and a double-criteria decision strategy is constructed, first according to the numerical value of the error level of the candidate interval is determined, so that the average error level of the candidate interval is low, and then the window with the smallest window is selected from the candidate window.

5. The method of claim 3, wherein, In step S21, the remote sensing analysis model is a regression, classification or other estimation algorithm.

6. The method of claim 5, wherein, In step S21, the remote sensing analysis model can be a prediction model of the normalized vegetation index NDVI, the leaf area index LAI, the ground temperature Ts or the water quality parameter such as chlorophyll a Chl-a.

7. The method of claim 3, wherein, The residual sensitivity index RSI is defined as the mean value of the normalized residual curve in a window of a specified size, for quantifying the independent influence of the resolution on the model error.

8. The method of claim 3, wherein, The optimal resolution interval refers to a balance point that simultaneously satisfies a small deviation and a small fluctuation, i.e., a low RSI and a low Score based on Median + IQR.

9. The method of claim 2, wherein, In S13, the resampling selects any interpolation method to reconstruct the original image into multiple spatial resolution scales.

10. The method of claim 2, wherein, In S13, the target resolution is any equidistant resolution interval required for the research.

Images