An adaptive spectrum-weighted active-passive fusion water depth inversion method

Abstract

The application discloses a kind of adaptive spectrum weighted active-passive fusion water depth inversion methods, comprising: the atmospheric correction of multispectral satellite image is carried out, and after extracting target area, land area and deep water area are removed;Obtain the seawater point cloud of spaceborne photon counting lidar in target area, extract sea surface and seabed information, and carry out refractive index correction and tide correction, obtain reference water depth;Each pixel in multispectral satellite image and reference water depth are matched, and divided into reference pixel and to-be-measured pixel;According to the spectral value of multispectral satellite image, the Euclidean distance of each to-be-measured pixel and all reference pixels is calculated, and the reciprocal of distance is used as the weight of each reference pixel;Weight is introduced into weighted least squares method, and the relationship between spectral value and water depth is fitted for each to-be-measured pixel, to realize the water depth inversion of to-be-measured pixel.The application can effectively improve the active-passive fusion water depth inversion precision when water depth distribution is uneven and shallow sea area environment is complex.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of ocean laser radar remote sensing detection, and particularly relates to a self-adaptive spectral weighting active-passive fusion water depth inversion method. BACKGROUND

[0002] High-precision topographic mapping is of great significance for marine shipping, coastal development, scientific research, etc. Accurate water depth data is the key to drawing navigation charts, ensuring the safety of marine shipping, and avoiding grounding accidents. It is also the basis for coastal development activities such as port construction and bridge foundation site selection. It also provides indispensable data for scientific research such as monitoring of coastline changes and assessment of the impact of sea level rise.

[0003] Although the traditional shipborne, airborne instruments or on-site water depth measurement method has high precision, it requires a large amount of manual and time cost, and the coverage is limited, which cannot realize the mapping of some sensitive areas. Satellite-based water depth detection technology provides an effective way for global coastal mapping. Synthetic aperture radar, multispectral satellite and laser radar have been applied to water depth mapping. Synthetic aperture radar can achieve water depth derivation by detecting sea wave changes through linear wave theory between sea wave period, wavelength and water depth. This method can measure turbid water, but it is an indirect measurement, so the precision is low. Multispectral satellite and laser radar can receive the light signal returned from the seabed. Multispectral satellite can achieve large-scale water depth inversion, but it depends on true value derivation. Spaceborne photon counting laser radar, especially ICESat-2 (The Ice, Cloud, and Land Elevation Satellite-2), provides unprecedented detailed mapping due to its high along-track and vertical resolution, which can directly extract sea surface and seabed point clouds from point clouds, thereby achieving accurate water depth inversion. ICESat-2 provides unprecedented spaceborne sounding precision, but because the ICESat-2 spot is small, it can only provide two-dimensional point clouds, and cannot achieve large-scale water depth measurement.

[0004] As a new satellite water depth detection method, the active-passive fusion water depth inversion can provide the water depth information of ICESat-2 to the passive satellite as a reference value, so as to simultaneously play the advantages of multispectral satellite and ICESat-2, and thus obtain wide-range and high-precision nearshore water depth information. The patents with publication numbers CN117346744A and CN113960625A disclose various active-passive fusion water depth detection methods based on the logarithmic ratio method and the multi-band linear fitting method. However, the linear or ratio relationship between the spectrum and the water depth can only be maintained within a certain depth, and the actual water depth distribution is seriously uneven, so the final fitting result is poor in the water depth range with less reference points. Meanwhile, the environment in the shallow sea area is complex, and the seabed, suspended silt and phytoplankton can all affect the spectral value of the multispectral satellite, thereby affecting the sounding accuracy. SUMMARY

[0005] The application provides an adaptive spectral weighting active-passive fusion water depth inversion method, which can effectively improve the active-passive fusion water depth inversion accuracy in the uneven water depth distribution and complex shallow sea environment.

[0006] An adaptive spectral weighting active-passive fusion water depth inversion method comprises the following steps:

[0007] (1) Perform atmospheric correction on the multispectral satellite image, extract the target area, and remove the land area and deep water area;

[0008] (2) Obtain the seawater point cloud of the spaceborne photon counting lidar in the target area, extract the sea surface and seabed information from the seawater point cloud, and perform refractive index correction and tidal correction to obtain the reference water depth;

[0009] (3) Match each pixel in the multispectral satellite image and the reference water depth according to the latitude and longitude information, and divide them into reference pixels and to-be-measured pixels, wherein each reference pixel takes the reference water depth matched therewith as the water depth;

[0010] (4) Calculate the Euclidean distance between each to-be-measured pixel and all reference pixels according to the spectral value of the multispectral satellite image, and take the reciprocal of the distance as the weight of each reference pixel;

[0011] (5) Introduce the weight to the weighted least squares method, fit the relationship between the spectral value and the water depth for each to-be-measured pixel, and realize the water depth inversion of the to-be-measured pixel.

[0012] In step (1), the specific process of removing the land area and the deep water area is as follows:

[0013] Take the grid coordinates of the GEBCO global DEM data in the range of 0 to -100 m as a mask, and coarsely remove the land area and the deep water area in the multispectral satellite image;

[0014] The multispectral satellite image after rough elimination is segmented according to the normalized water body difference index, and the formula is as follows:

[0015] ;

[0016] wherein, is the green band reflectivity, is the near-infrared band reflectivity, and the fine elimination of land area and deep water area is completed by setting the NDWI threshold value to 0.

[0017] In step (2), the sea surface and seabed information is extracted from the seawater point cloud, and the specific process is as follows:

[0018] The seawater point cloud of the target area of the spaceborne photon counting laser radar is screened out according to the latitude and longitude, the seawater point cloud is divided into multiple segments along the track direction, for each segment of seawater point cloud, the vertical interval is set to generate a histogram of photon height distribution, and the histogram is fitted with a Gaussian function to calculate the average

[0019] of the Gaussian parameters of the sea surface photon distribution and the standard , and the sea surface photon range is determined as ; the elevation average value of the sea surface photon in the determined range is calculated every 10 pulses as the instantaneous sea surface elevation, and the photons less than are water body and seabed photons.

[0020] The DBSCAN algorithm is used to realize seabed extraction, and the uncorrected water depth is obtained by subtracting the instantaneous sea surface elevation from the seabed elevation.

[0021] In step (2), the refractive index correction and tide correction are performed, and the specific process is as follows:

[0022] The water depth after refractive index correction is:

[0023] ;

[0024] wherein, is the laser incidence angle, and are the refractive indices of air and water body respectively;

[0025] Then, according to the track time of the spaceborne photon counting laser radar and the time of the multispectral satellite image, the TPXO9 tide model is used to correct the to obtain the reference water depth .

[0026] The specific process of step (3) is as follows:

[0027] If there is no reference water depth in the range of a certain pixel, the pixel is classified into the pixel to be measured;

[0028] If there is one reference water depth in the range of a certain pixel, the pixel is classified into the reference pixel, and the one reference water depth is taken as the water depth of the pixel;

[0029] If there are multiple reference water depths in the range of a certain pixel, the pixel is classified into the reference pixel, and the average of the multiple reference water depths is taken as the water depth of the pixel.

[0030] In step (4), the Euclidean distance between each pixel to be measured and all reference pixels is calculated, and the formula is as follows:

[0031] ;

[0032] Wherein, represents the Euclidean distance between the pixel to be measured and the i-th reference pixel; , , , respectively represent the spectral values of the pixel to be measured in the red, green and blue three bands, , , respectively represent the spectral values of the reference pixel in the red, green and blue three bands, , is the number of reference pixels.

[0033] In step (4), the reciprocal of the distance is taken as the weight of each reference pixel, and the formula is as follows:

[0034] ;

[0035] Wherein, is a very small value used to avoid zero division, and for each pixel to be measured, a weight matrix is constructed. .

[0036] The specific process of step (5) is as follows:

[0037] The relationship between water depth and spectral value is fitted using the reference pixel for the pixel to be measured, and the logarithmic equation of water depth and red, green and blue three bands is expressed as:

[0038] ；

[0039] Wherein, , , respectively correspond to the parameters required for fitting in the red, green and blue three bands, are respectively the spectral values of red, green and blue three bands, , , are respectively the spectral values of red, green and blue three bands, , , are respectively the spectral values of red, green and blue three bands at the deep water;

[0040] The water depth and the logarithmic equation of red, green and blue three bands are expressed as a matrix form:

[0041] ;

[0042] wherein, is a parameter matrix to be solved, is a spectral value matrix of the reference pixel, is the water depth of the reference pixel; and the parameter matrix solved by the least square method is represented as:

[0043] ;

[0044] The weight matrix obtained by introducing step (4) is , and the parameter matrix is represented as:

[0045] ;

[0046] The parameter matrix is fitted for each water depth pixel to be measured, and the spectral value and the parameter matrix are substituted into the three-band equation to realize water depth inversion.

[0047] Compared with the prior art, the present application has the following beneficial effects:

[0048] 1. The present application avoids the excessive influence of the high-density sample area on the overall fitting result by introducing the spectral weight, effectively solving the water depth fitting deviation problem caused by uneven sample distribution.

[0049] 2. The present application gives higher weight to the pixels with similar spectral characteristics, so that the reference pixels with similar water body characteristics, water depth and bottom material have greater influence on the pixel to be solved, thereby reducing the interference of the water body environment difference and significantly improving the overall water depth inversion accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0051] ​Figure 1 A flow chart of an adaptive spectral weighted active-passive fusion water depth inversion method is provided for an embodiment of the present application.

[0052] Figure 2 A comparison of adaptive spectral weighted active-passive fusion inversion water depth and in-situ data on Santa Cruz Island is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0054] It should be noted that the features in the following embodiments and implementation manners can be combined with each other without conflict.

[0055] An embodiment of the present application selects the Santa Cruz Island in the United States as the water depth inversion area, with a latitude and longitude range of 17.60°-17.86°N, -65°-64.4°E. The spaceborne photon counting lidar inversion data used is the track of ICESat-2 in this area from October 2018 to December 2021, and the multispectral satellite data is the image of the domestic satellite HY-1D on April 4, 2023.

[0056] As shown in Figure 1 An adaptive spectral weighted active-passive fusion water depth inversion method includes the following steps:

[0057] Step S1: Perform atmospheric correction on the multispectral satellite image, extract the target area, and remove the land and deep water area.

[0058] ACOLITE software is used to perform atmospheric correction on HY-1D, and the Santa Cruz Island area (17.60°-17.86°N, -65°-64.4°E) is cropped out through latitude and longitude. The grid coordinates in the range of 0 to -100 m of the GEBCO (General Bathymetric Chart of the Oceans) global DEM data are used as a mask to coarsely remove the land and deep water area data in the multispectral remote sensing image. However, the resolution of GEBCO is lower than that of the actual remote sensing image, and there may still be part of the land that is not removed, which will affect the subsequent water depth inversion, so it is necessary to further finely remove the land. Segmentation is performed according to the normalized difference index, which is defined as:

[0059] ;

[0060] wherein is the green band reflectance value, is the near infrared band reflectance, by setting the threshold value 0 can accurately remove the land and low quality water depth area.

[0061] Step S2: Extracting sea surface and sea bottom information from the spaceborne photon counting lidar sea water point cloud, and performing refractive index correction and tidal correction as the reference water depth.

[0062] For each ICESat-2 track, the ICESat-2 point cloud in the region of 17.60°- 17.86°N, -65°- 64.4°E is screened out, and the original photon point cloud is divided into multiple segments with an interval of 1km along the track direction. For each segment of photons, a vertical interval of 0.15m is set to count the photon height distribution to generate a histogram; a Gaussian function is used to fit the histogram, and the Gaussian parameters of the sea surface photon distribution are calculated and the standard , and the sea surface photon range is determined as ; the elevation average value of the sea surface photon in the determined range is calculated every 10 pulses as the instantaneous sea surface elevation, and the photons less than are water and sea bottom photons. Then, the DBSCAN algorithm (Density Based Spatial Clustering of Applications with Noise) is used to extract the sea bottom, and the uncorrected water depth is obtained by subtracting the instantaneous sea surface elevation from the sea bottom elevation . Then, refractive index correction is performed, and the corrected water depth is:

[0063] ;

[0064] wherein, is the laser incidence angle, and are the refractive indices of air and water, respectively.

[0065] Then, according to the ICESat-2 track time and the multi-spectral image time, the TPXO9 tidal model (Tidal Prediction and eXtrapolation Ocean) is used for tidal correction, and the reference water depth is obtained.

[0066] Step S3: Matching the latitude and longitude of each pixel of the HY-1D satellite image with the reference water depth latitude and longitude, and dividing the multi-spectral pixels into reference pixels and to-be-measured pixels according to the matching result.

[0067] Step S3: For each pixel in S1, find the corresponding reference water depth in S2. If there is no reference water depth in the pixel's range, it is classified as a test pixel. If there is a reference water depth in the pixel's range, it is classified as a reference pixel, and the corresponding reference water depth is its water depth. If there are multiple reference water depths in the pixel's range, the average of the reference water depths is its water depth.

[0068] Step S4: For each test pixel, calculate its distance to all known water depth pixels in the Euclidean space, and use the inverse of the distance as the weight.

[0069] For each test pixel, calculate its distance to all reference pixels in the Euclidean space, which can be represented as:

[0070] ;

[0071] wherein, represents the Euclidean distance between the test pixel and the i-th reference pixel; , , , represent the spectral values of the test pixel in the red, green, and blue bands, respectively, , , represent the spectral values of the reference pixel in the red, green, and blue bands, respectively, , is the number of reference pixels.

[0072] The weight given to each reference pixel is the inverse of the Euclidean distance, which can be represented as:

[0073] ;

[0074] wherein, is a small value to avoid division by zero. For each test sample, a unique weight diagonal matrix is constructed. .

[0075] Step S5: Use the reference pixels to fit the relationship between water depth and spectral values for the test pixels. The water depth and red, green, and blue band logarithmic equation can be represented as:

[0076] ;

[0077] wherein, represents the parameters to be fitted, to correspond to the parameters to be fitted for the red, green, and blue bands, is the constant term, is the spectral value at deep water.

[0078] For the non-weighted model, only the water depth and spectral value of the reference pixels are used to fit a set of common parameters . While for the adaptive spectral weighting method, a unique set of parameters is fitted for each water depth to be measured To speed up the process, a matrix form is used to realize batch processing. The matrix form of the three-band equation can be expressed as:

[0079] ;

[0080] where, is the parameter matrix to be solved, is the spectral value matrix of the reference pixels, is the water depth of the reference pixels. The parameter matrix solved by the least square method can be expressed as:

[0081] ;

[0082] The weight matrix in S4 is introduced , then the weight parameter matrix can be expressed as:

[0083] ;

[0084] A unique set of parameters is fitted for each water depth pixel to be measured, and the spectral value and parameter matrix are substituted into the three-band equation to realize water depth inversion. To facilitate further comparison and verification, the water depth is corrected to the average sea level using the tidal correction.

[0085] To verify the effect of the present application, the adaptive spectral weighting method and the ordinary three-band equation are used to invert the water depth, and the results are compared with the in-situ data, as shown in Figure 2 The method of the present application (left side) has obvious overall accuracy improvement compared with the non-weighted fitting method (right side), and effectively corrects the phenomenon of smaller predicted water depth existing in the sparse reference water depth (i.e. water depth greater than 20 m).

[0086] The above-described embodiments have described the technical solutions and beneficial effects of the present application in detail. It should be understood that the above-described only the specific embodiments of the present application, and is not used to limit the present application, any modification, supplement and equivalent replacement made within the principle range of the present application, should be included in the protection scope of the present application.

Claims

1. An adaptive spectral weighted active-passive fusion method for water depth inversion, characterized in that, Includes the following steps: (1) Perform atmospheric correction on multispectral satellite images, extract the target area, and then remove land and deep water areas; (2) Obtain the seawater point cloud of the spaceborne photon counting lidar in the target area, extract the sea surface and seabed information from the seawater point cloud, and perform refractive index correction and tidal correction to obtain the reference water depth. (3) Match each pixel in the multispectral satellite image with the reference water depth based on latitude and longitude information, and divide them into reference pixels and pixels to be measured respectively. Each reference pixel will use the reference water depth it matches as the water depth. (4) Based on the spectral values ​​of the multispectral satellite image, calculate the Euclidean distance between each pixel to be measured and all reference pixels, and use the reciprocal of the distance as the weight of each reference pixel; (5) Introduce weights into the weighted least squares method to fit the relationship between the spectral value and the water depth for each pixel to be measured, thereby realizing the water depth inversion of the pixel to be measured; the specific process is as follows: Using a reference pixel to fit the relationship between water depth and spectral values ​​for the pixel under test, the logarithmic equation for water depth and the red, green, and blue bands is expressed as: ； in, , , The parameters to be fitted correspond to the red, green, and blue bands, respectively. For constant terms, , , These represent the spectral values ​​for the red, green, and blue bands, respectively. , , These represent the spectral values ​​of the red, green, and blue bands in deep water, respectively. The logarithmic equations for water depth and the red, green, and blue bands are expressed in matrix form: ； in, The parameter matrix to be obtained is... The spectral value matrix of the reference pixel, The water depth is used as a reference pixel; the parameter matrix obtained by the least squares method is... Represented as: ； Introducing the weight matrix obtained in step (4) The parameter matrix is ​​then represented as: ； A parameter matrix is ​​fitted for each pixel of the water depth to be measured. The water depth is then inverted by substituting the spectral values ​​and the parameter matrix into the three-band equation.

2. The adaptive spectral weighted active-passive fusion water depth inversion method according to claim 1, characterized in that, In step (1), the specific process of eliminating land areas and deep water areas is as follows: Using grid coordinates in the range of 0 to -100 m from the GEBCO global DEM data as a mask, coarse removal of land and deep water areas in multispectral satellite images was performed. The multispectral satellite images after coarse removal are segmented according to the normalized water body difference index, as shown in the following formula: ； in, For green band reflectivity, For near-infrared reflectance, by setting the NDWI threshold to 0, fine-grained removal of land and deep-water areas is achieved.

3. The adaptive spectral weighted active-passive fusion water depth inversion method according to claim 1, characterized in that, In step (2), sea surface and seabed information is extracted from the seawater point cloud. The specific process is as follows: Based on latitude and longitude, the seawater point cloud of the spaceborne photon counting lidar in the target area is selected, and the seawater point cloud is divided into multiple segments along the orbital direction. For each segment of the seawater point cloud, the vertical interval is set to statistically analyze the photon height distribution and generate a histogram. The average Gaussian parameter of the sea surface photon distribution was calculated by fitting the histogram with a Gaussian function. and standards And determine the range of photons on the sea surface as For each 10 pulses of photons over a defined area, the average elevation is calculated as the instantaneous sea surface elevation. (The value is less than...) The photons are water and seabed photons; The DBSCAN algorithm is used to extract the seabed elevation. Subtracting the instantaneous sea surface elevation from the seabed elevation yields the uncorrected water depth. .

4. The adaptive spectral weighted active-passive fusion water depth inversion method according to claim 3, characterized in that, In step (2), refractive index correction and tidal correction are performed. The specific process is as follows: Water depth after refractive index correction for: ； in, The laser incident angle, and These are the refractive indices of air and water, respectively. Then, based on the trajectory time of the spaceborne photon counting lidar and the time of the multispectral satellite images, the TPXO9 tidal model was used to... Perform tidal correction to obtain reference water depth .

5. The adaptive spectral weighted active-passive fusion water depth inversion method according to claim 1, characterized in that, The specific process of step (3) is as follows: For each pixel of the multispectral satellite image processed in step (1), perform latitude and longitude spatial matching with the reference water depth. If there is no matching reference water depth within a certain pixel range, then the pixel is classified as a pixel to be tested. If there is a reference water depth within a certain pixel range, then the pixel is classified as the reference pixel, and this reference water depth is used as its water depth; If there are multiple reference water depths within a certain pixel range, then that pixel is classified as a reference pixel, and the average of the multiple reference water depths is taken as its water depth.

6. The adaptive spectral weighted active-passive fusion water depth inversion method according to claim 1, characterized in that, In step (4), the Euclidean distance between each pixel to be tested and all reference pixels is calculated, as follows: ； in, Indicates the difference between the pixel to be measured and the first pixel. Euclidean distance of one reference pixel; , , These represent the spectral values ​​of the pixel under test in the red, green, and blue bands, respectively. , , These represent the spectral values ​​of the reference pixel in the red, green, and blue bands, respectively. , This is the reference pixel count.

7. The adaptive spectral weighted active-passive fusion water depth inversion method according to claim 6, characterized in that, In step (4), the reciprocal of the distance is used as the weight of each reference pixel, as shown in the following formula: ； in, It is a minimum value used to avoid division by zero. For each pixel to be measured, a matrix is ​​constructed using... The weight matrix formed .

Images