Remote sensing image pixel and sampling point matching method and device based on Gaussian distribution

Abstract

The application discloses a kind of based on Gaussian distribution's remote sensing image pixel and sampling point matching method and device, it is related to remote sensing data processing and environmental monitoring technical field.The method first obtains ground sampling point longitude and latitude coordinate and the remote sensing image data of coverage area, and extracts image spatial resolution, the geographical reference information of pixel;Then according to image resolution to determine initial weighted window range, in combination with preset boundary weight threshold and maximum coverage distance, calculate Gaussian kernel standard deviation;With sampling point as center, construct two-dimensional Gaussian weight field, according to the Euclidean distance of sampling point and the center of pixel in image to calculate weight, and according to the pixel set of dynamic determination involved in calculation according to Gaussian kernel coverage range;Finally, the reflectivity of each pixel is multiplied by corresponding weight and is normalized, and obtains the estimated reflectivity at sampling point.The method can effectively fuse neighborhood pixel information when sampling point and pixel center do not coincide, improve reflectivity extraction precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing data processing and geoscience information analysis, and particularly relates to a remote sensing image pixel and sampling point matching method and device based on Gaussian distribution. BACKGROUND

[0002] In the fields of water quality monitoring, environmental investigation and ecological assessment, remote sensing images and ground sampling point data are often jointly analyzed. Since remote sensing images have fixed spatial resolution and pixel grid structure, the longitude and latitude positions of ground sampling points are often not completely coincided with the centers of image pixels, and directly extracting the reflectivity value of a single pixel may introduce large errors.

[0003] Traditional pixel-sampling point matching methods mainly include nearest neighbor method and simple average method, but these methods do not fully consider the spatial distance difference between sampling points and pixel centers, which easily leads to information loss and spatial heterogeneity distortion. In addition, when the sampling point falls near the pixel boundary, the spectral information of the adjacent pixels cannot be reasonably weighted to the actual sampling point position, thereby reducing the reflectivity estimation accuracy and affecting the reliability of subsequent water quality parameter inversion.

[0004] Therefore, there is an urgent need for a method that can dynamically determine the pixel set participating in the calculation according to the spatial relationship between the sampling point and the surrounding pixels, and reasonably weight and fuse the reflectivity, so as to improve the accuracy of matching remote sensing data and ground measured data. SUMMARY

[0005] To solve the above problems, the present application provides a remote sensing image pixel and sampling point matching method and device based on Gaussian distribution, which solves the reflectivity estimation error problem caused by the non-coincidence of the sampling point position and the pixel center by constructing a two-dimensional Gaussian weight field and weighting and fusing according to the spatial distance between the sampling point and the pixel center, thereby improving the data matching accuracy.

[0006] The technical scheme of the present application is as follows:

[0007] A remote sensing image pixel and sampling point matching method based on Gaussian distribution, comprising the following steps:

[0008] S1: Collecting the longitude and latitude coordinates of the ground sampling points and obtaining the remote sensing image data covering the sampling point area; extracting the spatial resolution, pixel row and column number and geographic reference information of the remote sensing image;

[0009] S2: Determining the spatial range of the initial weighting window according to the spatial resolution of the remote sensing image;

[0010] S3: Setting the weight threshold of the Gaussian function at the maximum coverage distance of the window , and calculating the Gaussian kernel standard deviation . ;

[0011] S4: Construct a two-dimensional Gaussian weight field centered on the sampling point position, and calculate the weight of each pixel ;

[0012] S5: Calculate the Euclidean distance according to the spatial position difference between the sampling point and the center of the adjacent pixel ;

[0013] S6: Determine the dynamic pixel set based on the coverage range of the Gaussian kernel, so that the coverage range is consistent with the spatial relationship of the sampling point position;

[0014] S7: Multiply the reflectivity of each pixel in the dynamic pixel set by the corresponding weight and sum; divide the resulting weighted sum by the sum of all weights to obtain the estimated reflectivity at the sampling point position;

[0015] S8: Use the estimated reflectivity as input data for remote sensing inversion modeling.

[0016] A remote sensing image pixel and sampling point matching device based on Gaussian distribution, comprising the following modules:

[0017] The sampling point module collects the latitude and longitude coordinates of the ground sampling points and obtains remote sensing image data covering the sampling point area; extracts the spatial resolution, pixel row and column number and geographic reference information of the remote sensing image;

[0018] The window module determines the spatial range of the initial weighting window according to the spatial resolution of the remote sensing image;

[0019] The standard deviation calculation module sets the weight threshold of the Gaussian function at the maximum coverage distance of the window , and calculates the Gaussian kernel standard deviation ;

[0020] The Gaussian weight field construction module constructs a two-dimensional Gaussian weight field centered on the sampling point position, and calculates the weight of each pixel ;

[0021] The Euclidean distance calculation module calculates the Euclidean distance according to the spatial position difference between the sampling point and the center of the adjacent pixel ;

[0022] The dynamic pixel set determination module determines the dynamic pixel set based on the coverage range of the Gaussian kernel, so that the coverage range is consistent with the spatial relationship of the sampling point position;

[0023] The reflectivity estimation module multiplies the reflectivity of each pixel in the dynamic pixel set by the corresponding weight multiplying and summing; dividing the resulting weighted sum by the sum of all weights to obtain the estimated reflectance at the sample point location;

[0024] an input module, the estimated reflectance being input data for remote sensing inversion modeling.

[0025] An electronic device, comprising: one or more processors; a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method.

[0026] A computer-readable storage medium having stored thereon executable instructions that, when executed by a processor, cause the processor to implement the method.

[0027] The present application has the following beneficial effects:

[0028] The present application can effectively fuse the spectral information of neighboring pixels under the condition that the sample point and the pixel center do not coincide, and can perform smoothing weighting according to the spatial distance to reduce the estimation error introduced by the position deviation; the pixel set is dynamically determined, so that invalid calculation caused by the fixed window method is avoided; the present application is suitable for multispectral and hyperspectral remote sensing data, and provides high-precision input data for environmental monitoring research. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 is a flowchart of a pixel and sample point matching method based on Gaussian distribution of a remote sensing image;

[0030] Figure 2 is a comparison graph of error band curves of three reflectance extraction methods on 251 spectral bands;

[0031] Figure 3 is a p-value distribution histogram of three reflectance extraction methods and ground measured values. DETAILED DESCRIPTION

[0032] The present application will be further described below in conjunction with the accompanying drawings and specific embodiments, which are illustrative examples of the present application and serve as a description of the present application. The present application is only explained, but not limited.

[0033] As shown in the figure: the present application proposes a pixel and sample point matching method based on Gaussian distribution of a remote sensing image, comprising the following steps: Figure 1

[0034] S1: the latitude and longitude coordinates of 10 sample points are obtained, and 0.75m airborne hyperspectral remote sensing image data covering the sample point area on the same day is obtained, and the sample point position is mapped to a specific position of the remote sensing image according to the latitude and longitude of the sample point;

[0035] ​S2: Taking the satellite image with a spatial resolution of 20m as the reference scale, the pixel window within a distance of 20m from the center pixel is taken as the reference window scale according to the same physical coverage range;

[0036] S3: The weight threshold of the Gaussian function at the maximum coverage distance of the window is set to 20m, and the formula is:

[0037] ;

[0038] wherein, the value range of . The Gaussian kernel standard deviation is calculated to ensure that different resolution images have consistent weight decay characteristics in the same geographical range;

[0039] S4: Taking the actual geographical position of the sampling point as the center, a two-dimensional Gaussian weight field is constructed in its neighborhood range, and the weight calculation formula is:

[0040] ;

[0041] wherein, is the Euclidean distance between the th pixel center and the sampling point;

[0042] S5: The spatial distance between the sampling point and each adjacent pixel center is calculated , and the pixel weight is obtained according to the above formula, the spatial distance between the sampling point and each adjacent pixel center is calculated by the Euclidean distance formula:

[0043] ;

[0044] S6: When the sampling point position deviates from the pixel center, the pixel set participating in the weighting is dynamically selected according to the coverage range of the Gaussian kernel;

[0045] S7: The reflectivity of each participating pixel is multiplied by the corresponding weight , and then summed and divided by the sum of all weights to obtain the estimated reflectivity at the sampling point:

[0046] ;

[0047] S8: The estimated reflectivity obtained is taken as the input data for subsequent remote sensing inversion modeling.

[0048] Step S1 includes: ​​​

[0049] S11: Obtain the longitude and latitude coordinates of the sampling point using GPS or other positioning devices;

[0050] S12: Retrieve image data covering the sampling point area from the remote sensing database;

[0051] S13: Georeference the image data so that the pixels are consistent with the geographic coordinates.

[0052] The determination of the dynamic pixel set is based on the actual landing position of the sampling point within the pixel, and is automatically determined by calculating the intersection area of the Gaussian kernel coverage range and the grid system.

[0053] Embodiment:

[0054] This embodiment uses the pixel matching reflectance extraction method based on Gaussian distribution (Method C) proposed in the present application, and compares and evaluates it with the traditional single-pixel extraction method (Method A) and the pixel window mean method (Method B). The evaluation indicators include reflectance extraction accuracy and statistical significance test. Figure 1 The steps of the pixel matching reflectance extraction method based on Gaussian distribution proposed in the present application are shown, Figure 2 The reflectance extraction accuracy error band curves of the three methods at 251 spectral bands are shown, Figure 3 The p-value distribution histograms of the three methods and the ground measured reflectance are shown.

[0055] From Figure 2 It can be seen that the error band curve of Method C is below Methods A and B at most waveband positions, showing better extraction accuracy. Statistical calculation shows that the mean absolute error (MAE) of the three methods at all wavebands is 0.0025, 0.0023 and 0.0022 respectively, among which the MAE of Method C is the lowest, proving its obvious advantage in reflectance extraction accuracy. The width of the error band reflects the volatility of the extraction error, i.e. the stability of the method. The analysis results show that the error band width of Method C is generally narrower, especially at wavebands of 500-630 nm and after 680 nm, the standard deviation of its MAE is significantly lower than that of Methods A and B, indicating that Method C has higher stability and smaller error fluctuation.

[0056] Figure 3 It is shown that the p-value of Method A is mainly concentrated in the lower interval (0-0.05) and is relatively dispersed, indicating that the extraction results at multiple wavebands have significant differences with the ground measured reflectance. In contrast, the p-value distribution of Method C is more concentrated than that of Methods A and B, and compared with Method B, its distribution is more concentrated, reflecting obvious statistical significance advantage. Therefore, the pixel matching method based on Gaussian distribution is proved to be the preferred solution for image reflectance extraction.

Claims

1. A method for matching pixels and sampling points in remote sensing images based on Gaussian distribution, characterized in that, Includes the following steps: S1: Collect the latitude and longitude coordinates of ground sampling points and obtain remote sensing image data covering the sampling point area; extract the spatial resolution, pixel row and column numbers and their geographic reference information of the remote sensing images; S2: Determine the spatial range of the initial weighted window based on the spatial resolution of the remote sensing image; S3: Set the maximum coverage distance of the Gaussian function within the window. Weight threshold at And calculate the Gaussian kernel standard deviation. ; S4: Construct a two-dimensional Gaussian weight field centered on the sampling point location, and calculate the weight of each pixel. ; S5: Calculate the Euclidean distance based on the spatial difference between the sampling point and the center of the neighboring pixel. ; S6: Determine the dynamic pixel set based on the coverage of the Gaussian kernel, so that the spatial relationship between the coverage and the sampling point location is consistent; S7: Calculate the reflectance of each pixel in the dynamic pixel set. With corresponding weights Multiply and sum; divide the weighted sum by the sum of all weights to obtain the estimated reflectance at the sampling point location; S8: Use the estimated reflectance as input data for remote sensing inversion modeling.

2. The method according to claim 1, characterized in that, Step S1 includes: S11: Use GPS or other positioning devices to obtain the latitude and longitude coordinates of the sampling points; S12: Retrieve image data covering the sampling point area from the remote sensing database; S13: Georegister the image data to make the pixels consistent with the geographic coordinates.

3. The method according to claim 1, characterized in that, The Gaussian kernel standard deviation The maximum coverage distance of the window With weight threshold The constraints are determined and satisfy the formula: ； in, The value range is 0 < <1.

4. The method according to claim 1, characterized in that, The weight Calculate using the following formula: ； in, For the first The Euclidean distance between the center of each pixel and the sampling point This is the Gaussian kernel standard deviation calculated in step S3.

5. The method according to claim 1, characterized in that, The dynamic pixel set is determined based on the actual location of the sampling point within the pixel, and is automatically determined by calculating the intersection area between the Gaussian kernel coverage and the grid system.

6. The method according to claim 1, characterized in that, The formula for estimating reflectance is: ； in, For the first The reflectance value of each pixel. This represents the weight value for the corresponding pixel.

7. The method according to claim 1, characterized in that, Sampling points With the center of each neighboring pixel spatial distance Calculated using the Euclidean distance formula: 。 8. A remote sensing image pixel and sampling point matching device based on Gaussian distribution, characterized in that, Includes the following modules: The sampling point module collects the latitude and longitude coordinates of ground sampling points and acquires remote sensing image data covering the sampling point area; it also extracts the spatial resolution, pixel row and column numbers, and geographic reference information of the remote sensing images. The window module determines the spatial range of the initial weighted window based on the spatial resolution of the remote sensing image. The standard deviation calculation module sets the maximum coverage distance of the Gaussian function within the window. Weight threshold at And calculate the Gaussian kernel standard deviation. ; The Gaussian weight field construction module constructs a two-dimensional Gaussian weight field centered on the sampling point location and calculates the weight of each pixel. ; The Euclidean distance calculation module calculates the Euclidean distance based on the spatial difference between the sampling point and the center of the neighboring pixel. ; The dynamic pixel set determination module determines the dynamic pixel set based on the Gaussian kernel coverage area, so that the spatial relationship between the coverage area and the sampling point location is consistent; The reflectance estimation module calculates the reflectance of each pixel in the dynamic pixel set. With corresponding weights Multiply and sum; divide the weighted sum by the sum of all weights to obtain the estimated reflectance at the sampling point location; The input module takes the estimated reflectance as input data for remote sensing inversion modeling.

9. An electronic device, characterized in that, include: One or more processors; A memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed by a processor, cause the processor to perform the method described in any one of claims 1 to 7.

Images