CN-122017873-A - Underwater topography inversion method based on satellite-borne ICESat-2 photon satellite

CN122017873ACN 122017873 ACN122017873 ACN 122017873ACN-122017873-A

Abstract

The invention discloses an underwater topography inversion method based on a satellite-borne ICESat-2 photon satellite, and relates to the technical field of remote sensing water depth inversion. Inversion is carried out by applying ICESat-2 photon denoising and local robust estimation method, data acquisition and preprocessing are carried out, large-area photon denoising and signal extraction based on PAConv-PointNet ++, PAConv-PointNet ++ is to directly embed PAConv into PointNet ++, water depth physical correction comprises water depth refraction correction and water depth tide correction, and fine denoising and water depth track fitting are carried out based on the local robust estimation method. According to the invention, the clustering parameters are not required to be preset manually, and the method can be used for autonomous learning in a data-driven mode, so that the high-precision segmentation of signal photons and noise photons can be realized in a complex shallow water environment with high turbidity, extremely shallow water and high noise ratio, and the denoising performance and environmental adaptability of sparse photon data are obviously improved.

Inventors

SUN HAI
LIANG BINGCHEN
WANG ZHENGUO
WANG ZHENLU
WANG HUIQIAN
ZHANG JUNKAI

Assignees

中国海洋大学

Dates

Publication Date: 20260512
Application Date: 20260416

Claims (9)

1. An underwater topography inversion method based on a satellite-borne ICESat-2 photon satellite is characterized by applying a ICESat-2 photon denoising and local robust estimation method to perform inversion, and comprises the following steps: s1, data acquisition and preprocessing; S2, large-area photon denoising and signal extraction based on PAConv-PointNet ++ model, wherein the PAConv-PointNet ++ model is formed by directly embedding PAConv into PointNet ++ model; s3, water depth physical correction, including water depth refraction correction and water depth tide correction; And S4, carrying out fine denoising and water depth track fitting based on a local robust estimation method.
2. The underwater topography inversion method as claimed in claim 1, wherein the step s1 comprises: S11, collecting ICESat-2 ATL03 global geographic positioning photon data of a research area, screening satellite orbit data extending along a coastal zone and containing submarine photon reflection, and completing batch downloading of original data; S12, primary screening is carried out on original ATL03 data, ocean type photons are extracted, and extreme abnormal values are removed according to the effective depth measurement range of an optical shallow water area; s13, carrying out coordinate transformation on the screened photon data, and projecting the original longitude and latitude geographic coordinates to a UTM coordinate system; s14, carrying out segmentation processing on the projected photon data; s15, collecting tidal model data corresponding to a research area; s16, collecting actual measurement water depth verification data of the research area, and unifying coordinates and elevation references.
3. The underwater topography inversion method as claimed in claim 1, wherein the step S2 comprises: s21, based on the fact that the along-orbit photons show double-layer structure photon characteristics formed by sea surface photons and seabed photons in the elevation direction, primarily classifying the double-layer structure photons; S22, constructing PAConv-PointNet ++ deep learning model, extracting self-adaptive characteristics of photon distribution, and realizing large-area denoising of underwater photons.
4. A method of inverting an underwater topography as claimed in claim 3 wherein step S21 comprises: s211, quantitatively identifying the photon characteristics of the sea surface-seabed double-layer structure by carrying out nuclear density estimation on the elevation distribution of the along-orbit photons, marking the sea surface photon signals, and completing the primary extraction of sea surface photons; S212, calculating the elevation difference of the adjacent submarine photons based on the characteristic that the elevation difference of the adjacent submarine photons is small due to continuous change of submarine topography Screening and meeting of elevation difference frequency histogram and exponential decay model Taking the maximum elevation difference of the condition as a threshold value, executing three rounds to realize gradual elimination of noise photon points, and marking the effective signal photons on the seabed, wherein c is a constant term, and y is a model fitting frequency corresponding to the elevation difference delta elev of the adjacent photons; s213, marking photons which do not accord with an elevation difference threshold value as noise photons based on the characteristic of elevation difference distribution dispersion of randomly distributed noise photons, and finishing preliminary division of sea surface, submarine signal photons and noise photons; S214, manually checking and correcting a complex scene sample by adopting a light-weight strategy of few batches, multiple rounds and modularization, defining the physical boundary of sea surface and submarine photons aiming at the signal aliasing problem of an extremely shallow water scene, correcting the wrong division of an aliasing area, and distinguishing weak signal photons from high-density noise clusters by combining the optical characteristics of a water body and the spatial distribution rule of photons aiming at the signal noise boundary blurring problem of a high-noise scene so as to avoid the wrong rejection of effective submarine signals.
5. A method of inverting an underwater topography as claimed in claim 3 wherein step S22 comprises: S221, sampling original input point cloud data in PAConv-PointNet ++ model by utilizing a sampling layer, selecting a center point of a local neighborhood, and adopting a furthest point sampling method to ensure that sampling centers are uniformly distributed in space; The grouping layer in the PAConv-PointNet ++ model is responsible for fusing local areas of sampling points, each sampling point is taken as a circle center for an input sampling point matrix, a local area is constructed by combining the circle center and the area radius, and local feature points are sampled in each area to obtain a point set matrix divided into the local areas; In the PAConv-PointNet ++ model, PAConv is responsible for encoding the local neighborhood characteristics after grouping and outputting local characteristic vectors with discriminant power; s224, generating a dynamic convolution kernel adapted to the current point position by using a weight library, a scoring network and a dynamic kernel of PAConv in PAConv-PointNet ++ model; S225, a characteristic splicing stage adopts a layered progressive implementation mode, a distance weighted interpolation and cross-layer jump connection mechanism is fused, the spliced characteristics are input into a shared MLP unit to carry out nonlinear transformation and characteristic extraction, the characteristics are propagated to the original point cloud resolution layer by layer, and finally the classification probability of each photon point is output through a full connection layer, so that the point-by-point segmentation task of signal photons and noise photons is completed; and S226, training and optimizing the model by using the labeled training data set, inputting the underwater photon data set into the model after training, outputting the classification probability of each photon point, and completing the point-by-point segmentation of the underwater signal photons and the noise photons.
6. The underwater topography inversion method as claimed in claim 1, wherein the water depth refraction correction comprises: Based on the optical propagation characteristics and Snell's law, a refraction correction model of laser at an air-water interface is established, a quantitative relation between a refraction angle and an incident angle of the laser is defined, and a correction amount calculation formula of the underwater photon and the vertical direction is deduced by combining a trigonometric function relation, wherein the correction amount calculation formula is as follows: , Wherein Q is the geometric deviation vector modular length, , , Wherein S is uncorrected skew distance, R is refraction-corrected skew distance, And The incident angle and refraction angle are shown, respectively, n 1 is the air refractive index, n 2 is the sea water refractive index, and D is the uncorrected water depth value.
7. The underwater topography inversion method as claimed in claim 6, wherein the correction amount calculation formula is obtained by calculating an offset direction by angle decomposition, the angle decomposition calculation offset direction calculation formula is as follows: wherein gamma is the total deviation angle, alpha is the vertical deviation angle, and beta is the horizontal deviation angle.
8. The underwater topography inversion method as claimed in claim 1, wherein the water depth tide correction comprises: Acquiring a tide level value corresponding to the observation time from tide model or tide level observation data, and obtaining a water depth value under a unified reference plane through correction, wherein a specific correction formula is as follows; , Wherein, the To achieve a tidal corrected water depth value, For ICESat-2 of the underwater photon elevation acquired at time t, As the value of the instantaneous tide level, , wherein, 、 And The amplitude, angular frequency and initial phase of the G-th moisture component are respectively represented, G takes values of 1, 2 and 3.
9. The underwater topography inversion method as claimed in claim 1, wherein the step S4 comprises: S41, constructing an iteration weighting model of local robust estimation, adopting equal-weight independent observation assumption, constructing an error equation of a Gaussian-Markov model, adopting a formula 19, combining the local robust estimation to construct an objective function, adopting a formula 20, introducing an IGG3 scheme based on post-test unit weight error iteration to construct an robust equivalent weight function, adopting a formula 21, and continuously reducing the weight of an abnormal value in the estimation process through an iteration weighting strategy; 19, wherein V is a residual vector, A is a coefficient matrix, L is an observation vector, P is an observation weight matrix, D (L) is the variance-covariance matrix of the observation vector; 20, Wherein, the For the i-th component of the matrix a, For the i-th component of the observation vector L, For the weight corresponding to the i-th observation, V i is the i-th residual, n is the number of total observations, i.e. the total number of photon points in the strip, Is a parameter Is used for the estimation of the (c), As a weight function in the robust estimate, The operation of the minimum value is pointed out; 21, Wherein, the Is based on residual error Is used for the normalization of the residual component, Is equivalent weight, is to original observation weight The new weight value obtained after the correction is carried out, And Is an empirical constant, take , The specific value is adjusted according to the data scale and the noise characteristics; s42, setting sliding window parameters by combining ICESat-2 system parameters, and taking the elevation median value of the sounding photons in each window as an initial parameter estimation value; s43, performing local robust estimation iterative solution, and in each sliding window, updating an equivalent weight matrix and a model parameter estimation value through iteration until iteration converges, and gradually reducing the equivalent weight of a corresponding outlier observation value to approach zero so as to effectively inhibit an outlier; s44, based on the iteration solving result, converting the discrete submarine signal photon points into a spatially continuous and smooth along-orbit water depth track, and outputting a final water depth inversion result.

Description

Underwater topography inversion method based on satellite-borne ICESat-2 photon satellite Technical Field The invention relates to the technical field of remote sensing water depth inversion, in particular to an underwater topography inversion method based on a satellite-borne ICESat-2 photon satellite. Background The offshore shallow water area is a core area of interaction between sea and land, is influenced by multiple factors such as waves, tides, runoffs and the like, has severe dynamic change of topography and frequent evolution of a tidal channel system. The method for accurately measuring the water depth data and the topography dynamic of the shallow water area has important theoretical and practical significance for coastal zone evolution mechanism research, ocean resource management, channel safety guarantee, ecological environment protection and the like. However, existing water depth measurement techniques face many challenges in shallow water complex environments. Although the traditional shipborne acoustic sounding can reach the centimeter level in precision, the traditional shipborne acoustic sounding has the limitations of high operation cost and limited coverage range, is easy to be stranded in a shoal area, and is difficult to touch remote or dangerous sea areas. The airborne laser radar sounding makes up the limitation of ships to a certain extent, but the technical threshold and the unit area cost are higher, the airborne laser radar sounding depends on proprietary software and professionals, and large-scale continuous sounding is difficult to realize. The Advanced Topography Laser Altimetry System (ATLAS) carried by ICESat-2 satellites can emit green laser with the wavelength of 532nm and has the capacity of penetrating through water, and ATL03 global geographic positioning photon data provides a totally new high-precision data source for water depth inversion of a near-shore optical shallow water area. However, the existing water depth denoising method based on ICESat-2 photon data has the obvious technical defects that 1) strong absorption and scattering effects can occur in the water body transmission process of laser, so that the signal photon density of water bottom echo is extremely sparse along with the increase of water depth, daytime observation data is seriously affected by solar background noise, the noise photon occupancy rate is far higher than that of effective signal photons, and the water depth precision is reduced. 2) DBSCAN, OPTICS and other traditional density clustering denoising methods need to manually adjust parameters such as neighborhood radius, minimum point number and the like, have poor parameter adaptability, are good in performance only in clear water bodies, and suddenly drop in denoising performance in complex environments such as turbid and extremely shallow water. 3) A large amount of noise and outliers still remain in the primarily denoised underwater signal photons, the existing least square fitting method is highly sensitive to the outliers, is easily interfered by outliers to cause distortion of water depth fitting results, and lacks means of robust outlier inhibition and continuous water depth track fitting. In view of the problems, the existing method is difficult to meet the requirements of the water depth measurement precision, the robustness and the environmental adaptability of ICESat-2 photon data under the complex environment of a shallow water area, and the invention is particularly provided. Disclosure of Invention Aiming at the problems of low denoising precision, high manual intervention degree and poor scene adaptability of ICESat-2 photon data, the invention provides a water depth inversion method based on a satellite-borne ICESat-2 photon satellite, which is based on a two-step method of optical shallow water region ICESat-2 photon denoising and local robust estimation water depth inversion method. The denoising performance and the environmental adaptability of the sparse photon data are obviously improved. The invention aims at realizing the following technical scheme: An underwater topography inversion method based on a satellite-borne ICESat-2 photon satellite applies ICESat-2 photon denoising and local robust estimation method to carry out inversion, comprising the following steps: s1, data acquisition and preprocessing; S2, large-area photon denoising and signal extraction based on PAConv-PointNet ++ model, wherein the PAConv-PointNet ++ model is formed by directly embedding PAConv into PointNet ++ model; s3, water depth physical correction, including water depth refraction correction and water depth tide correction; And S4, carrying out fine denoising and water depth track fitting based on a local robust estimation method. Preferably, step s1 comprises: S11, collecting ICESat-2 ATL03 global geographic positioning photon data of a research area, screening satellite orbit data extending along a coastal zone and containing submarin