Inversion of water depth by satellite green band water bottom trend normalization model

Abstract

The invention discloses a satellite green wave band water bottom trend normalization model inversion water depth aiming at water surface fluctuation, water body suspended matter scattering, non-uniform water bottom substrate reflection and non-linear noise superposition characteristics. According to the characteristics that low-resolution water depth data points are lack of underwater topographic data but can provide the underwater trend and water area satellite images are interfered by noise but have high-resolution remote sensing data, advantages are complementary, noise is suppressed, an underwater trend green band normalized image is constructed, and in combination with priori water depth point data, an underwater trend green band normalized image is obtained. And inverting the parameters of the linear water depth inversion model through a least square method to invert the water depth. Through verification of a Sailey island test area, compared with a single-band model (RMSE = 4.19 m, R = 0.79), a logarithmic ratio model (RMSE = 5.62 m, R = 0.62), a ratio logarithmic model (RMSE = 6.02 m, R = 0.56) and a multi-band model (RMSE = 3.88 m, R = 0.82), the method has the advantages that the RMSE is 1.40 m (the precision improvement rate is 63.9%-76.7%) and the R is 0.97, the water depth inversion precision is improved, and a new technology is provided for navigation safety and engineering construction application water depth acquisition.

Description

Technical Field

[0001] This invention relates to the field of remote sensing water depth inversion, specifically to the inversion of water depth using a satellite green band bottom trend normalization model. Background Technology

[0002] Nearshore water depth data is crucial for maritime safety and engineering construction. Traditional single-beam / multi-beam bathymetry techniques offer high accuracy but are inefficient and struggle to cover large sea areas. Remote sensing satellite imagery-based water depth retrieval technology has become a research hotspot due to its advantages of wide coverage, low cost, and high speed. Currently, it mainly relies on traditional empirical models and semi-analytical models for water depth retrieval.

[0003] Traditional models (single-band / dual-band ratio method / log-linear model) are effective in optically simple water bodies and are showing a trend towards spectral stratification and incorporating bottom sediment characteristics. For example, Cheng Liang's 2024 patent "A Shallow Water Depth Inversion Method Based on Spectral Stratification" (Publication No. CN115235431B) divides the inversion intervals based on the differences in the penetrating power of red, green, and blue light, and constructs Stumpf ratio and Lyzenga polynomial sub-models respectively, alleviating the problem of traditional algorithms "not taking into account the spectral depth measurement limit". Chu Sensen's 2025 patent "A Shallow Water Depth Hyperspectral Inversion Method Considering Bottom Sediment Characteristics" (Publication No. CN119672518A) combines hyperspectral imagery and spaceborne lidar data, weakening heterogeneous interference by using the mean of bottom sediment spectral sampling, but it does not solve the nonlinear modulation effect of suspended sediment on reflectivity. Although the aforementioned patents have optimized the sediment and spectral transmittance processing, in waters with suspended solids concentrations >50 mg / L (such as estuaries), the enhanced backscattering caused by sediment still leads to an absolute shift in reflectance, resulting in an inversion system error of +2.8 to +3.5 m.

[0004] Semi-analytical models for retrieving water depth using the radiative transfer equation have seen rapid development in recent years. For example, Chen Peng's 2024 patent, "A Water Depth Retrieval Method Based on the Fusion of Active and Passive Remote Sensing Using Spaceborne Single-Photon Laser" (Publication No. CN113960625B), utilizes the fusion of ICESat-2 photon point clouds and Sentinel-2 images to achieve uncontrolled water depth retrieval. Cui Lei's patent, "A Shallow Water Depth Retrieval Method and System Based on a Multi-Angle Optical Radiative Transfer Model" (Publication No. CN120008564A), constructs a multi-angle optical model to improve adaptability to complex optical environments, but suffers from low computational efficiency (single scene > 25 minutes) and sensitivity to abrupt changes in water optical parameters. Song Mingzhu's invention patent, "An Uncontrolled Multispectral Remote Sensing Water Depth Retrieval Method and System" (Publication No. CN119198570A), classifies bottom reflectivity types using a depth invariance index, but relies on the assumption of "adjacent optical deep water areas," leading to matching failures in turbid waters due to decreased visibility.

[0005] While existing satellite remote sensing methods for water depth measurement have yielded results, they suffer from poor adaptability to complex environments such as surface ripples, scattering by suspended matter in the water, and uneven reflection from the seabed. In practical applications, combining these methods with high-precision measured data is necessary to achieve high-precision water depth retrieval from satellite imagery. Summary of the Invention

[0006] This invention addresses the characteristics of surface ripples, scattering by suspended matter in water, and uneven reflection from the seabed, as well as the superposition of nonlinear noise. Leveraging the lack of high-resolution underwater topographic data in low-resolution water depth data (electronic nautical charts) and the availability of high-resolution remote sensing data from satellite imagery, this invention combines the advantages of low-resolution water depth trends (electronic nautical charts) with the high resolution of satellite imagery signals. It suppresses the nonlinear noise generated by the exponential decay of water bodies in satellite signals due to the influence of various factors such as the water surface, water volume, and seabed sediment. A "green band seabed trend normalized" image is constructed, and then the model parameters are inverted using known low-resolution water depth reference points to obtain the model. Finally, the water depth is inverted using the "green band seabed trend normalized" image. The specific method is as follows:

[0007] S1: Constructing a "green band underwater trend normalization" image

[0008] S1-1: Principle of Third-Order Inverse-Range Quadratic Convolution Interpolation. The location information and inverse-square distance relationship between the interpolation point of low-resolution water depth data and 16 surrounding points are used. The existing water depth data is resampled using the third-order inverse-range quadratic convolution interpolation method to obtain water depth raster data with the same resolution as satellite imagery. The water depth value D of the interpolation point (iu,j+v) is to be determined. (i-u,j+v) The interpolation point (i,j) is obtained by taking the inverse square weighted average of the distances from the sixteen nearest sampling points p1-p16 in the rectangular grid to (iu,j+v). The weight of each sampling point is determined by the inverse square of the distance from that point to the interpolation point (i,j) in both the horizontal and vertical directions. Specifically, the distances from the interpolation point (i,j) to the interpolation point (iu,j+v) in the y-axis and x-axis directions are u and v, respectively, so the weight is w. (i-u,j+v) Similarly, the weights of the remaining 15 pixel coordinates can be obtained. Then, the water depth D at the interpolation point (iu, j+v) is... (i-u, j+v) The result will be obtained by formula (1):

[0009] (1)

[0010] In the formula, D (i+u,j+v) The water depth values ​​to be interpolated are given by geographic coordinates. A is the weight vector, B is the vector of 16 sampling points, and w (i+0,j+0) The weight of point (i,j).

[0011] S1-2: Underwater topographic data is normalized. A linear normalization formula is used to map the low-resolution water depth data to the [0,1] interval. The normalized underwater trend is denoted as D. nij .

[0012] (2)

[0013] S1-3: Normalization of the green band of remote sensing image. The logarithm of the green band value of the remote sensing image is taken using formula (3) to establish a linear relationship between water depth and green band radiance.

[0014] (3)

[0015] In the formula, G ij This represents the radiance value of the i-th row and j-th column of the green band in the remote sensing image.

[0016] Using a linear normalization formula, the green band of the remote sensing image is mapped to the [0,1] interval. The normalized green band data is denoted as L. nij .

[0017] (4)

[0018] S1-4: Construct the "green band underwater trend normalization" image. Calculate the "underwater topographic trend normalization" image using formula (5).

[0019] (5)

[0020] S2: Constructing a linear water depth inversion model

[0021] (6)

[0022] S3: Combining the prior water depth data of the "reference water depth data" and the "green band bottom trend normalized" image data, the least squares method is used to invert the model parameters.

[0023] S3-1: Least squares method to invert model parameters. The process of constructing the error function E using formula (7) to minimize the loss function involves differentiating α with respect to a and β, and setting the partial derivatives to 0 to solve for the optimal model parameters a and β.

[0024] (7)

[0025] The advantages of this invention are as follows: Addressing the characteristics of surface fluctuations, scattering by suspended matter in water, and uneven reflection from the seabed, as well as the superposition of nonlinear noise, this invention combines the advantages of low-resolution water depth data (electronic charts) which lacks underwater topographic data but can provide underwater topographic trends, and high-resolution remote sensing data from water satellite imagery which is subject to noise interference. This suppresses nonlinear noise generated by various factors such as the water surface, water body, and seabed sediment affecting satellite signals, constructing an "underwater trend green band normalized remote sensing image." Then, using low-resolution known water depth reference points and least squares inversion model parameters, the "trend green band normalized remote sensing image" is substituted into the model to invert water depth. Attached Figure Description

[0026] Figure 1 The principle of normalized inversion of water depth based on satellite green band underwater trend

[0027] Figure 2 The principle of third-order inverse-distance squared bicubic convolution interpolation

[0028] Figure 3 Experimental Procedure

[0029] Figure 4 Overview of the Saipan Experimental Zone

[0030] Figure 5 For the Saipan Island experimental area, the data points and checkpoints for water depth, preliminary water depth, and line layout are set up.

[0031] Figure 6 This includes single-band models, logarithmic ratio models, logarithmic ratio models, multi-band models, and the inverted depth map of this invention.

[0032] Figure 7 The actual water depth of Saipan is compared with the single-band bathymetry model, the logarithmic ratio bathymetry model, the logarithmic ratio bathymetry model, the multi-band model, and the scatter plot of the water depth retrieved by this invention.

[0033] Figure 8 Error diagrams for single-band model, logarithmic ratio model, logarithmic ratio model, multi-band model, and the inversion depth of this invention.

[0034] Figure 9 The diagram shows the water depth error of the single-band model, logarithmic ratio model, logarithmic ratio model, multi-band model, and the three profile inspection lines for water depth inversion according to this invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of the present invention clearer, preferred embodiments are described below, and the specific implementation of the present invention is further described in detail with reference to the accompanying drawings.

[0036] Example:

[0037] Combination Figure 1 The left image illustrates the principle of satellite-based water depth retrieval. First, sunlight is absorbed and scattered by gas molecules and cloud particles in the atmosphere. Part of the transmitted light reaches the water surface, while the rest is reflected back to the satellite sensor. A portion of the light reaching the water depth penetrates the surface and is absorbed and scattered by water molecules, chlorophyll, yellow substances, and non-pigmented particles, reaching the bottom and being reflected back into the water, where it is again absorbed and scattered by these particles. Part of the light passes through the water surface into the atmosphere, is absorbed and scattered by gas molecules and cloud particles, and is then received by the satellite sensor. The remaining light that does not pass through the water surface either returns to the water to be absorbed and scattered, or it passes through the water surface along with other light and is then received by the satellite sensor. Therefore, it can be inferred that the reflection effect caused by water surface ripples, the absorption and scattering by suspended matter in the water, and the heterogeneous reflectivity of the bottom are the main noise sources.

[0038] like Figure 1 As shown in the figure above, in this embodiment, a profile is acquired at the same location in the green band image and the measured water depth map, and the corresponding profiles GP and RP are drawn. The comparison shows that the distribution of areas A and B in GP is very different from that of areas A' and B' in RP. Therefore, low-resolution bottom data normalization by interpolation is used to correct the inconsistency between the distribution of satellite image and bottom data.

[0039] like Figure 1 As shown in the figure below, in this embodiment, the average value of the normalized bottom trend and the normalized green band image is taken, and combined with the known reference point, the least squares method is used to invert the model parameters to obtain a linear inversion model. The model is then substituted into the "normalized green band bottom trend" image to invert the water depth.

[0040] Combination Figure 2 This section explains the principle of third-order inverse distance-squared bicubic convolution interpolation for water depth. In this embodiment, the water depth of the low-resolution point to be interpolated is compared with the positional information and inverse square distance relationship of 16 surrounding points. Third-order inverse distance-squared bicubic convolution interpolation is used to resample the existing water depth data into raster water depth data with the same resolution as the satellite imagery. The water depth value D of the interpolation point (iu,j+v) is to be determined. (i-u,j+v) The interpolation point is obtained by taking the inverse square weighted average of the distances from the sixteen nearest sampling points p1-p16 in the rectangular grid to (iu,j+v). The weight of each sampling point is determined by the inverse square of the distance from that point to the interpolation point (including the distances in both the horizontal and vertical directions). See formula (1) for details.

[0041] Combination Figure 3 The experimental procedure for this embodiment is as follows:

[0042] S1: Combination Figure 4This indicates that the experimental area in this embodiment is Saipan Island. Saipan Island is located in the western shelf area of ​​the Northern Mariana Islands in the western Pacific Ocean. Administratively, it belongs to the Commonwealth of the Northern Mariana Islands of the United States. It presents a rectangular research area with geographical coordinates of 15°12'1.95"N to 15°13'38.12"N and 145°41'8.57"E to 145°42'56.88"E. The basement is composed of Cenozoic volcanic island arc rock layers, and the surface layer features coral reef ecosystems and calcareous bioclastic deposits.

[0043] S1-1: Combination Figure 4 This section describes the satellite imagery data used in this embodiment. The experimental area utilized Landsat 8 multispectral remote sensing satellite imagery across nine bands. The spatial resolution of the multispectral band image was 30 meters, and the spatial resolution of the panchromatic band image was 15 meters, with a width of 185 kilometers. The imaging time was 2018-08-13 T00:37:33 (GMT). Preprocessing was then performed, including data radiometric calibration, atmospheric correction using a second simulation of satellite signals from the solar spectral (6S) model, solar flare removal, water body extraction, and cropping of the study area imagery data.

[0044] S2: Combination Figure 5 This section describes the water depth data for the experimental area, as well as the prior water depth data, checkpoints, and lines. The underwater topographic trend and prior water depth data used for water depth inversion from satellite images of Saipan Island are both from NOAA electronic chart data, with a maximum depth of 30.74m. The verification water depth data is from lidar water depth data, with a maximum depth of 31.40m, which differs by 0.66m due to the influence of tide height. The electronic chart data and lidar water depth data are corrected based on the water depth and tide height at the time of satellite transit. The electronic chart data of the study area are interpolated using the third-order inverse distance squared bicubic convolution interpolation method (formula (1)) to obtain the DEM data of the study area, calculate the water depth trend data, and extract the reference point water depth data. In addition, 7 prior water depth data points 1-7, 7 checkpoints 1-7, and 3 check lines 1-3 are set up in the verification water depth data for subsequent water depth inversion and verification.

[0045] S3: Combination Figure 6 This demonstrates that the traditional model and the depth sounding obtained by inversion in this invention yield a different depth map.

[0046] S3-1: Implementation of the water depth inversion model of this invention. Using formula (1), bicubic convolution interpolation is performed on the water depth data to convert the discrete water depth point set into a regular grid, generating a continuous water depth surface D. ij The interpolated water depth data is normalized using formula (2), denoted as D. nij Using formulas (3) and (4), the green band values ​​of the remote sensing image are logarithmically normalized and denoted as L. lijUsing formula (5) “green band bottom trend normalization” imagery, construct linear water depth inversion model formula (6).

[0047] S3-2: The model parameters of the "green band bottom trend normalized" images corresponding to the 7 prior water depth data points and their geographical locations were inverted using the least squares method (formula (7)) to construct a normalized model and invert the water depth at the time of satellite transit in the study area. At the same time, the water depth at the time of satellite transit in the study area was inverted using the single-band model, logarithmic ratio model, ratio logarithmic model and multi-band model.

[0048] S4: Perform scatter plot analysis, regional error analysis, check line analysis, and check point error analysis on the water depth of the study area inverted by the model of this invention and four traditional models and the corrected satellite border time real water depth.

[0049] S4-1: Combination Figure 7 This paper presents a scatter plot analysis of the actual water depth of Saipan compared to traditional models and the inverted water depth of this invention. Compared to the single-band model (RMSE=4.19 m, R²=0.79), the logarithmic ratio model (RMSE=5.62 m, R²=0.62), the logarithmic ratio model (RMSE=6.02 m, R²=0.56), and the multi-band model (RMSE=3.88 m, R²=0.82), the model of this invention (RMSE=1.40 m, R²=0.97) shows improvements in RMSE of 66%, 75%, 77%, and 64%, respectively, and improvements in R² of 23%, 56%, 73%, and 18%, respectively. By comparing the accuracy of water depth inversion, it was found that traditional models have defects such as underestimation of the overall water area (single-band model / multi-band model), deviation of deep-water clusters, and underestimation (logarithmic ratio / logarithmic ratio model). In contrast, the inversion range of this invention is normal, the fluctuation is weak, and it has stronger compactness than traditional models, resulting in good inversion performance.

[0050] S4-2: Combination Figure 8 This paper explains the error analysis between the actual water depth of Saipan and the traditional model and the water depth inversion method of this invention. The water depth inversion methods of five different models are subtracted from the measured water depth data to obtain a depth measurement error map. The accuracy is evaluated by statistically analyzing the percentage of errors within the range of -2m to 2m.

[0051] (1) The absolute error of the single-band model is 46.47% within 0-2m and 53.53% beyond 2m (concentrated in the deep sea, which is related to reflectivity saturation caused by optical signal attenuation);

[0052] (2) The absolute errors of the logarithmic ratio and the logarithmic ratio model accounted for 43.96% and 46.32% respectively within the range of 0-2m, and both exceeded 53% beyond 2m. Furthermore, the errors diverged significantly in the mid-deep to deep-sea regions;

[0053] (3) The absolute error of the multi-band model is 48.77% within 0-2m and 51.23% beyond 2m (the percentage varies from shallow to deep sea).

[0054] (4) The absolute error of the model in this paper accounts for 91.36% within 0-2m and 8.64% beyond 2m (the spatial distribution is in the form of discrete circular spots, which is highly consistent with the abrupt changes in the bottom sediment around the island in the remote sensing image (such as reef edge and bedrock), verifying the strong robustness of the model to complex underwater topography.

[0055] S4-3: Combination Figure 9 This study explains the error analysis of the actual water depth of Saipan compared to traditional models and the inverted water depth inspection line profile of this invention.

[0056] Inspection Line 1:

[0057] For single-band model inversion of water depth from the left endpoint 0~360m: the error fluctuates around -1.5m; from 360~1710m: the error fluctuates around 3m; after 1710m: the error fluctuates significantly within the range of -6m~5m (significantly exceeding the reference range).

[0058] For the logarithmic ratio model, the absolute error of the water depth inversion distance from the left endpoint 0~1980m is greater than 1.5m; for 1980~2070m, the error is within the range of -1.5m~1.5m; for 2070~2580m, the absolute error is greater than 1.5m; and beyond 2580m, the error is within the range of -1.5m~1.5m.

[0059] For the ratio logarithmic model, the absolute error of the water depth inversion distance from the left endpoint 0~1950m is greater than 1.5m; for 1950~2100m, the error is within the range of -1.5m to 1.5m; after 2100m, the error fluctuates around -1.5m.

[0060] The multi-band model inverted the water depth distance from the left endpoint 0~360m, which fluctuated around -1.5m; from 360~1710m, which fluctuated around 3m; and after 1710m, it fluctuated significantly within the range of -6m~5m.

[0061] The inversion depth distances of this invention from the left endpoint 0~720m: error within the range of -1.5m~1.5m; 720~1350m: error approximately -1.5m; 1350~1500m: absolute error greater than 1.5m; 1500~2580m: error within the range of -1.5m~1.5m; 2580~2730m: error greater than 1.5m; after 2730m: error within the range of -1.5m~1.5m.

[0062] Inspection line 2:

[0063] For single-band model inversion of water depth, the error is mostly within -2m to 2m from the left endpoint of 0 to 3270m.

[0064] For the logarithmic ratio model, the absolute error of the water depth inversion distance from the left endpoint 0~1050m is greater than 2m; for 1050~1260m, the error fluctuates around -2m; after 1260m, the error is within the range of -2m~2m.

[0065] The ratio logarithmic model inverts the water depth distance from the left endpoint of 0~1050m, which exceeds -2m~2m; from 1050~1260m, it fluctuates around -2m; and after 1260m, it falls within the range of -2m~2m.

[0066] The multi-band model inverted water depth within the range of 0~3270m from the left endpoint, with most errors exceeding -2m~2m.

[0067] The present invention inverts the water depth distance from the left endpoint: 0~40m: error within -2m~0m; 40~300m: absolute error greater than 2m; 300~630m: error within -2m~2m; 630~780m: absolute error greater than 2m; 780~1050m: error within -2m~2m; 1050~1200m: absolute error greater than 2m; 1200~1830m: error within -2m~2m; 1830~1980m: absolute error greater than 2m; after 1980m: error within -2m~2m.

[0068] Inspection line #3:

[0069] For single-band model inversion of water depth within the range of 0~3270m from the left endpoint, most errors exceed -1.5m~1.5m.

[0070] For the logarithmic ratio model, the inversion depth distance from the left endpoint 0~1380m: the absolute error is greater than 1.5m; 1380~1620m: the error is within the range of -1.5m~1.5m; 1620~1920m: the error is greater than 1.5m; 1920~2130m: the error is within the range of -1.5m~1.5m; after 2130m: it first fluctuates around -1.5m, and then stabilizes within the range of -1.5m~1.5m.

[0071] For the ratio logarithmic model, the inversion of water depth from the left endpoint 0~1440m: the absolute error is greater than 1.5m; 1440~1590m: the error is within the range of -1.5m~1.5m; 1590~1890m: the error is greater than 1.5m; 1890~2100m: the error is within the range of -1.5m~1.5m; after 2100m: it first fluctuates around -1.5m, and then stabilizes within the range of -1.5m~1.5m.

[0072] The multi-band model inverted water depth within the range of 0~3270m from the left endpoint, with most errors exceeding -1.5m~1.5m.

[0073] The inversion depth distance of this invention from the left endpoint 0~1920m: error within the range of -1.5m~1.5m; 1920~2130m: absolute error greater than 1.5m; after 2130m: error within the range of -1.5m~1.5m.

[0074] During the verification of the accuracy of the inverted water depth in this invention, local anomalies were found along the inspection line sections. Specifically, for survey line 1, the anomalies occurred in the range of 1380-1410 m from the left endpoint (accounting for 1% of the total survey line length); for survey line 2, the anomalies occurred in the ranges of 40-300 m, 630-780 m, 1050-1200 m, and 1830-1980 m from the left endpoint (accounting for 19% of the total survey line length); and for survey line 3, the anomalies occurred in the range of 1920-2100 m from the left endpoint (accounting for 5% of the total survey line length). The LiDAR water depth and the water depth inverted by the model of this invention showed an absolute deviation exceeding 2 m. Multi-source data analysis revealed that this anomaly mainly stemmed from interpolation errors caused by insufficient density of original sampling points on the electronic chart. Overall, the absolute error of the model of this invention was most concentrated in the range of -2 m to 2 m in the studied inspection lines, accounting for 91.67%. Therefore, the model in this paper has the smallest inspection line inversion error and the best performance in all study areas.

[0075] S4-4: Table 1 shows the comparison of the water depth errors of the traditional model and the present invention at 7 checkpoints, and Table 2 shows the reduction rate of the error of the present invention model compared to the traditional model. A comprehensive analysis of Tables 1 and 2 shows that the maximum absolute errors of the traditional model in water depth are 2.61m, 9.86m, 11.53m, and 2.61m, respectively, and the average absolute errors are 1.51m, 4.85m, 5.09m, and 1.32m, respectively. In contrast, the maximum absolute error of the present invention inverted water depth is 1.57m, and the average absolute error is 0.93m. The results indicate that the maximum absolute error of the water depth inverted by the present invention is reduced by 39.8%, 84.1%, 86.3%, and 39.8% compared to the traditional model, respectively, and the average absolute error of the water depth inverted by the present invention is reduced by 38.4%, 80.8%, 81.7%, and 29.5% compared to the traditional model, respectively. The root mean square errors of water depth inversion for traditional models are 1.60m, 5.60m, 6.06m, and 1.45m, respectively, while the root mean square error of water depth inversion for this invention is 0.99m. That is, the root mean square error of water depth inversion for this invention is reduced by 38.1%, 82.3%, 83.6%, and 31.7% respectively compared with traditional models.

[0076] Table 1 Comparison of checkpoint errors for five inversion models

[0077]

[0078] Table 2 shows the rate of accuracy reduction of checkpoint error in this model compared to the traditional inversion model.

[0079]

[0080] In conclusion, the satellite green band bottom trend normalization model can improve the accuracy of water depth inversion to a certain extent, making the inverted water depth closer to the true water depth, and thus having more practical significance and value.

[0081] The above embodiments are for illustrative purposes only and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all equivalent technical solutions also fall within the scope of this invention, and the patent protection scope of this invention should be defined by the claims.

[0082] All technical contents not described in detail in this invention are publicly known technologies.

Claims

1. A satellite green band underwater trend normalization model for inverting water depth, characterized in that... The principle of using the satellite green band bottom trend normalization model to retrieve water depth is as follows: Addressing the characteristics of surface fluctuations, scattering by suspended matter in the water, and uneven reflection from the seabed sediment, as well as the superposition of nonlinear noise, this model leverages the advantages of high-resolution satellite imagery—which provides high-resolution water remote sensing data—in addition to low-resolution water depth data (electronic charts) which lack high-resolution underwater topographic data. This approach combines the water depth trend of low-resolution water depth data (electronic charts) with the high-resolution nature of the satellite imagery signal. It suppresses the nonlinear noise generated by the exponential decay of the water body in the satellite signal due to various factors such as the water surface, water body, and seabed sediment. A "green band bottom trend normalization" image is then constructed. The model parameters are then retrieved using low-resolution known water depth reference points to obtain the model, which is finally substituted into the "green band bottom trend normalization" image to retrieve the water depth.

2. The inversion of water depth using the satellite green band bottom trend normalization model as described in claim 1, characterized in that... The specific method is as follows: S1: Constructing "Green Band Underwater Trend Normalization" Image S1-1: Principle of third-order inverse distance-squared bicubic convolution interpolation; The positional information and inverse squared distance relationship between the interpolation point of low-resolution water depth data and 16 surrounding points are used. The existing water depth data is resampled using the third-order inverse distance-squared bicubic convolution interpolation method to obtain water depth raster data with the same resolution as satellite imagery; the water depth value D of the interpolation point (iu,j+v) to be determined is... (i-u,j+v) The interpolation point (i,j) is obtained by taking the inverse square weighted average of the distances from the sixteen nearest sampling points p1-p16 in the rectangular grid to (iu,j+v). The weight of each sampling point is determined by the inverse square of the distance from that point to the interpolation point (i,j) in both the horizontal and vertical directions. Specifically, the distances from the interpolation point (i,j) to the interpolation point (iu,j+v) in the y-axis and x-axis directions are u and v, respectively, so the weight is w. (i-u,j+v) Similarly, the weights of the remaining 15 pixel coordinates can be obtained. Therefore, the water depth D at the interpolation point (iu,j+v) can be calculated. (i-u, j+v) The result will be obtained by formula (1): (1) In the formula, D (i+u,j+v) The water depth values ​​to be interpolated are given by geographic coordinates. A is the weight vector, B is the vector of 16 sampling points, and w (i+0,j+0) The weight of point (i,j); S1-2: Normalize the underwater topographic data; A linear normalization formula is used to map low-resolution water depth data to the [0,1] interval. The normalized bottom trend is denoted as D. nij ; (2) S1-3: Normalization of the green band of remote sensing image; take the logarithm of the green band value of remote sensing image using formula (3) to establish a linear relationship between water depth and green band radiance; (3) In the formula, G ij The radiance value of the green band in the remote sensing image is in the i-th row and j-th column. Using a linear normalization formula, the green band of the remote sensing image is mapped to the [0,1] interval. The normalized green band data is denoted as L. nij ; (4) S1-4: Construct "green band underwater trend normalized" image; calculate "underwater topographic trend normalized" image using formula (5); (5) S2: Constructing a linear water depth inversion model (6) S3: Combining prior water depth point data from the "reference water depth data" and "green band bottom trend normalized" image data, the least squares method is used to invert the model parameters. S3-1: Least squares method to invert model parameters; using formula (7) to construct the error function E, minimizing the loss function, taking the derivatives of α with respect to a and β respectively, and setting the partial derivatives to 0, to solve for the optimal model parameters a and β. (7)。

Images