Three-dimensional Stereo Reconstruction Method for Water Environment Parameters Based on Active and Passive Optical Remote Sensing

Abstract

The present invention discloses a three-dimensional stereoscopic reconstruction method for water environment parameters based on active and passive optical remote sensing, which relates to the technical field of sea water environment parameter monitoring. The method includes obtaining water surface environment parameter data and point cloud data at the vertical depth D_z of water environment parameters; matching the two sets of data to obtain the pixel points in the water surface environment parameter data that are closest to each point cloud data in the horizontal distance, and assigning values; through a reconstruction model, obtaining a horizontal spatial distribution data set of water environment parameters at the underwater depth of D_z; using the horizontal spatial distribution data of water environment parameters at the underwater depth of D_z as surface data, and through the reconstruction model, obtaining the horizontal spatial distribution data of water environment parameters at D_z + x. The present invention gets rid of the limitation of obtaining three-dimensional stereoscopic observation data of water bodies by using on-site experimental data in the past, and the obtained results can provide a large amount of three-dimensional stereoscopic all-round data support for scientific researchers studying marine geophysical characteristics.

Description

Technical Field

[0001] The present invention relates to the technical field of monitoring water environment parameters in sea areas, and in particular to a three-dimensional reconstruction method for water environment parameters based on active and passive optical remote sensing. Background Art

[0002] Marine water environment parameters, such as chlorophyll a concentration, suspended sediment concentration (SPM), particulate organic carbon concentration (POC), etc., are important indicators reflecting the status of the marine ecological environment. The traditional method for obtaining the vertical distribution data of these water environment parameters mainly relies on on-site sampling and measurement by shipborne instruments, but this method has high costs, low efficiency, and a small observation range. Existing observation platforms and technologies are still unable to conduct a large-scale and continuous three-dimensional evaluation of these marine water environment parameters.

[0003] The development of optical remote sensing satellites and lidar satellite technologies has provided new means for marine environment monitoring. Optical remote sensing satellites can obtain sea surface reflectance information and can be used to retrieve surface water environment parameters; lidar satellites can obtain water body vertical profile information and can be used to detect the internal structure of the water body. However, the optical remote sensing satellite data obtains the spatial distribution results of the surface water ecological environment parameters, while the lidar satellite obtains the vertical profile point cloud data results. Currently, there is a lack of an inversion algorithm model that effectively fuses the two satellite data to achieve the inversion of the vertical three-dimensional spatial distribution of water environment ecological parameters in large sea areas. Summary of the Invention

[0004] In order to overcome the above problems existing in the prior art, the present invention proposes a three-dimensional reconstruction method for water environment parameters based on active and passive optical remote sensing.

[0005] The technical solution adopted by the present invention to solve its technical problems is: a three-dimensional reconstruction method for water environment parameters based on active and passive optical remote sensing, including the following steps:

[0006] Step 1, determine the study area and time period, obtain the surface water environment parameter data C_level1 and the point cloud data Chlapro at the vertical depth D_z of the water environment parameter, and record the total number of the point cloud data Chlapro as num1;

[0007] Step 2, perform data matching on the two groups of data obtained in Step 1 to obtain the pixel points with the closest horizontal distance to each point cloud data in the surface water environment parameter data, and assign values;

[0008] Step 3, through the water environment parameter horizontal spatial distribution data reconstruction model at the vertical depth D_z of the water body, obtain the water environment parameter horizontal spatial distribution data set C_level2 at the underwater depth D_z;

[0009] Step 4: Using the horizontal spatial distribution data of water environment parameters at the underwater depth D_z as surface data, and through the reconstruction model described in Step 3, obtain the horizontal spatial distribution data of water environment parameters at D_z + x.

[0010] For the above three-dimensional vertical reconstruction method of water environment parameters based on active and passive optical remote sensing, Step 1 specifically includes: the water environment parameter data C_level1 at the water surface is obtained from the optical remote sensing data through the water remote sensing reflectance spectrum data, and the point cloud data Chlapro at the vertical depth D_z of the water environment parameters is obtained from lidar remote sensing data or on-site measurements.

[0011] The lidar remote sensing data is used to calculate the point cloud data Chlapro at the vertical depth D_z of the water environment parameters through the laser attenuation coefficient of the ocean water body.

[0012] For the above three-dimensional vertical reconstruction method of water environment parameters based on active and passive optical remote sensing, the assignment process in Step 2 is specifically as follows: the water environment parameter data matrix S_level2 at the underwater depth D_z corresponding one-to-one to the water environment parameter data at the water surface has an initial value of NaN. The pixel point in S_level2 that is closest to each point cloud data in terms of horizontal distance among the water environment parameter data at the water surface is assigned a value at the corresponding position, making it equal to the point cloud data at the vertical depth D_z of the water environment parameters.

[0013] For the above three-dimensional vertical reconstruction method of water environment parameters based on active and passive optical remote sensing, Step 3 is specifically as follows: calculate the proportion of the data volume matched in Step 2 in the total number of pixels. If the proportion is greater than k, then assign an initial value to C_level2 such that C_level2 is equal to C_level1. At the same time, inject the known data points in S_level2 into C_level2. Based on the correlation between C_level1 and C_level2 and the smoothness of C_level2 itself, construct the objective function T, set the constraint conditions and the gradient function, and find the minimum value of the objective function through the optimization algorithm.

[0014] If the proportion is not greater than k, expand from each obtained pixel point to a×a pixel points around it, intercept the dataset of C_level1 in the a×a area, label it as S_level1. At the same time, take the S_level2 data at this pixel point and label it as T_level2. Normalize the S_level1 data to obtain a weight matrix of size a×a, and apply this weight matrix to reconstruct the water environment parameters of the lower-layer S_level2 matrix at the corresponding a×a position to obtain S_level2'; Reconstruct S_level1 for all matching points according to the above method, and calculate whether the proportion of the effective data of the reconstructed S_level2 in the total number of pixels exceeds k. If the proportion is still not greater than k, then expand from the edge points of the effective data of S_level2' to a×a pixel points around it and continue to reconstruct until the proportion of the effective data of S_level2 in the total number of pixels exceeds k. At this time, reassign the effective data volume of S_level2' to num1 and perform the relevant operations when the proportion is greater than k.

[0015] For the above three-dimensional stereo reconstruction method of water environment parameters based on active and passive optical remote sensing, the objective function is expressed as:

[0016] ;

[0017] ;

[0018] ;

[0019] ;

[0020] Where a1 is the weight consistent with the surface layer, a2 is the row smoothing weight, and a3 is the column smoothing weight; T1 is the sum of the squared differences between the lower-layer data and the surface-layer data multiplied by the weight, T2 is the sum of the squared first-order differences in the row direction of the lower-layer data multiplied by the weight, and T3 is the sum of the squared first-order differences in the column direction of the lower-layer data multiplied by the weight; m and n are the number of rows and columns of the arrays C_level1 and C_level2; i and j are the array subscripts, and i takes values between [1 m], and j takes values between [1 n].

[0021] For the above three-dimensional stereo reconstruction method of water environment parameters based on active and passive optical remote sensing, the gradient function is expressed as:

[0022] ;

[0023] ;

[0024] ;

[0025] ;

[0026] Among them, grad is the gradient function.

[0027] The beneficial effects of the present invention are as follows. The present invention can obtain the horizontal spatial distribution of water environment parameters (such as chlorophyll a concentration, suspended sediment concentration, particulate organic carbon concentration, etc.) at any depth (within the effective detection depth of lidar) in a large range of sea areas by applying active and passive remote sensing data, getting rid of the limitation of obtaining three-dimensional observation data of water bodies by using on-site experimental data in the past, and proposing a reconstruction model that can invert the three-dimensional numerical values of water environment parameters based on optical remote sensing data and lidar detection data. The obtained results can provide a large amount of three-dimensional all-round data support for researchers studying marine geophysical characteristics, and solve the problem that researchers are limited by the lack of vertical data and cannot thoroughly study marine geophysical characteristics. Description of the Drawings

[0028] Figure 1 is a schematic flowchart of the present invention;

[0029] Figure 2 is a schematic diagram of the processing process when the proportion of the matched data volume in the total pixel volume in the embodiment of the present invention is not greater than 30%;

[0030] Figure 3 is a schematic diagram of the processing process when the proportion of the matched data volume in the total pixel volume in the embodiment of the present invention is greater than 30%;

[0031] Figure 4 is a schematic diagram of the reconstruction result in the embodiment of the present invention, where (a) is the surface distribution result of chlorophyll a concentration in the target sea area in November 2013; (b) is the distribution result obtained from the in-situ chlorophyll a concentration point cloud data of the target sea area in November 2013 through the Figure 2 steps shown; (c) is the horizontal distribution result of chlorophyll a concentration at a depth of 5 m underwater in the target sea area in November 2013 reconstructed by this algorithm. Detailed Embodiments

[0032] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be described in detail below in conjunction with the drawings and specific embodiments.

[0033] The present invention discloses a three-dimensional reconstruction method for water environment parameters based on active and passive optical remote sensing, including the following steps:

[0034] Step 1, determine the research area and time period, obtain the water surface water environment parameter data C_level1 and the point cloud data Chlapro at the vertical depth D_z of the water environment parameter, and record the total number of the point cloud data Chlapro as num1.

[0035] Step 1 specifically includes: The water surface environmental parameter data C_level1 is obtained from the optical remote sensing data through the water remote sensing reflectance spectral data, and the point cloud data Chlapro at the vertical depth D_z of the environmental parameters is obtained from lidar remote sensing data or on-site measurements;

[0036] The lidar remote sensing data is used to calculate the point cloud data Chlapro at the vertical depth D_z of the water environment parameters through the laser attenuation coefficient of the ocean water body.

[0037] Step 2: Match the two groups of data obtained in Step 1 to obtain the pixel points in the water surface environmental parameter data that are horizontally closest to each point cloud data, and perform assignment.

[0038] The specific process of assignment in Step 2 is as follows: The water environment parameter data matrix S_level2 at the underwater depth D_z corresponding one-to-one to the water surface environmental parameter data has an initial value of NaN. The pixel points in the water surface environmental parameter data that are horizontally closest to each point cloud data are assigned at the corresponding positions in S_level2, making it equal to the point cloud data at the vertical depth D_z of the water environment parameters.

[0039] Step 3: Reconstruct the model through the horizontal spatial distribution data of the water environment parameters at the vertical depth D_z of the water body to obtain the horizontal spatial distribution data set C_level2 of the water environment parameters at the underwater depth D_z.

[0040] Step 3 is specifically as follows: Calculate the proportion of the data volume matched in Step 2 in the total number of pixels. If the proportion is greater than k, then assign an initial value to C_level2, making C_level2 equal to C_level1. At the same time, inject the known data points in S_level2 into C_level2. Based on the correlation between C_level1 and C_level2 and the smoothness of C_level2 itself, construct the objective function T, set the constraint conditions and the gradient function, and find the minimum value of the objective function through the optimization algorithm;

[0041] If the proportion is not greater than k, taking each obtained pixel point as the center, expand it to a×a pixel points around, intercept the dataset of C_level1 within the a×a area, label it as S_level1. At the same time, take the S_level2 data at this pixel point and label it as T_level2. Normalize the S_level1 data to obtain a weight matrix of size a×a, and apply this weight matrix to reconstruct the water environment parameters of the lower-layer S_level2 matrix at the corresponding a×a position to obtain S_level2'; Reconstruct S_level1 for all matching points according to the above method, calculate whether the proportion of the effective data of the reconstructed S_level2 exceeds k in the total number of pixels. If the proportion is still not greater than k, then take the edge points of the effective data of S_level2' as the center and expand it to a×a pixel points around, and continue to reconstruct until the proportion of the effective data of S_level2 exceeds k in the total number of pixels. At this time, reassign the effective data volume of S_level2' to num1, and perform the relevant operations with a proportion greater than k.

[0042] The objective function is expressed as:

[0043] ;

[0044] ;

[0045] ;

[0046] ;

[0047] Where a1 is the weight consistent with the surface layer, a2 is the row smoothing weight, and a3 is the column smoothing weight; T1 is the sum of the squared differences between the lower-layer data and the surface-layer data multiplied by the weight, T2 is the sum of the squared first-order differences in the row direction of the lower-layer data multiplied by the weight, and T3 is the sum of the squared first-order differences in the column direction of the lower-layer data multiplied by the weight; m and n are the number of rows and columns of the arrays C_level1 and C_level2, and i, j are the array subscripts, where i ranges from [1 m] and j ranges from [1 n].

[0048] The gradient function is expressed as:

[0049] ;

[0050] ;

[0051] ;

[0052] ;

[0053] Where grad is the gradient function.

[0054] Step 4: Using the horizontal spatial distribution data of water environment parameters at the underwater depth D_z as the surface layer data, and through the reconstruction model described in Step 3, obtain the horizontal spatial distribution data of water environment parameters at D_z + x.

[0055] In this embodiment, taking the reconstruction of the horizontal distribution dataset of chlorophyll a concentration at the depth D_z in water as an example, the specific implementation of the method of the present invention is described. The chlorophyll a concentration here can be replaced by any water environment parameter data such as suspended sediment concentration SPM, particulate organic carbon concentration POC, etc. The specific steps are as Figure 1 shown.

[0056] Determine the study area and time period, and search for optical remote sensing data and lidar remote sensing data within the study area. From the optical remote sensing data, the chlorophyll a concentration data C_level1 (assumed to be m rows and n columns) of the water surface layer can be obtained through the water remote sensing reflectance spectrum Rrs(λ) data obtained therefrom. The chlorophyll a concentration of each pixel point is denoted as C_level1(log_i, lat_j, 0). Known inversion algorithms include YOC, OC2, OC3, and neural network methods, etc. An appropriate inversion method can be selected according to the differences in the study sea area. From the lidar remote sensing data, the lidar attenuation coefficient of the ocean water K lidar (λ, z) can be used to calculate the point cloud data of the chlorophyll a concentration profile of the water body, where the chlorophyll a concentration at the depth D_z is denoted as Chla pro (log_0, lat_0, D_z). Denote the total number of chlorophyll a concentration data of the water body at the depth D_z as num1 (num1 ≥ 1). It should be noted that Chla pro (log_0, lat_0, D_z) can not only be obtained through lidar remote sensing data, but also be measured data obtained through on-site experiments.

[0057] Combine the surface layer chlorophyll a concentration data C_level1(log_i, lat_j, 0) with the num1 chlorophyll a concentrations Chla at the depth D_z proMatch the data of (log_0, lat_0, D_z). The principle of spatio-temporal matching can be determined according to requirements. A total of num1 pixel points (log_x, lat_y) that are horizontally closest to each point cloud data position (log_0, lat_0) in the surface chlorophyll a concentration data are obtained. Assume that the chlorophyll a concentration data matrix at the D_z depth corresponding to the surface data one by one is S_level2 (m rows, n columns), and the initial values are all NaN. Assign the num1 pixel points (log_x, lat_y) to the corresponding positions in S_level2 so that they are equal to the chlorophyll a concentration point cloud data at the D_z depth, as shown in the following formula:

[0058] S_level2(log_x, lat_y, D_z)=Chla pro (log_0, lat_0, D_z).

[0059] Accordingly, the known terms obtained are C_level1 and S_level2. Among them, S_level2 contains the values of num1 matched data points S_level2 known , and the unknown data (i.e., the data that needs to be filled in, S_level2 missing ) is represented by NaN. In order to fill in the unknown data in the S_level2 data to obtain the complete chlorophyll a concentration spatial distribution dataset C_level2 at the D_z depth (i.e., the filled S_level2), a model for solving the optimization problem is constructed in this embodiment.

[0060] Reconstruction model

[0061] First, determine whether the amount of matched data num1 accounts for 30% of the total number of pixels m×n (this index is obtained from multiple experiments).

[0062] If num1 / (m×n) ≤ 30%, as Figure 2 shown, then expand from each obtained pixel point (log_x, lat_y) to 3×3 pixel points around it, intercept the dataset of the surface chlorophyll a concentration C_leve1 in this 3×3 area, label it as S_level1, and at the same time take the S_level2 data at this pixel point and label it as T_level2. Normalize the S_level1 data to obtain a weight matrix of size 3×3:

[0063] Weight = S_level1(i, j) / sum(S_level1);

[0064] Apply this weight matrix to reconstruct the chlorophyll a concentration at the corresponding 3×3 positions in the lower-layer S_level2 matrix to obtain a new S_level2:

[0065] Tol = 3×3×T_level2;

[0066] S_level2(log_x - 1:log_x + 1, lat_y - 1:lat_y + 1, D_z) = Weight×Tol;

[0067] Reconstruct the S_level2 data for all num1 matching points according to the above method, and calculate whether the effective data of the reconstructed S_level2 accounts for 30% of the total amount of m×n. If it is still less than <30%, then expand from the edge points of the effective data of S_level2 to 3×3 pixel points around the center, and continue the above algorithm for data reconstruction until the effective data of S_level2 accounts for more than 30% of the total amount of m×n. At this time, reassign the effective data volume of S_level2 to num1, and then proceed to the following steps.

[0068] If num1 / (m×n) > 30%, as Figure 3 shown, then first assign an initial value to C_level2 so that it is equal to the surface dataset C_level1, and at the same time inject the num1 known data point values S_level2 in S_level2 known into C_level2:

[0069] ;

[0070] 。

[0071] Then, construct the objective function T based on the correlation between the surface data C_level1 and the lower-layer data C_level2 and the smoothness of the lower-layer data C_level2 itself. T consists of three parts, namely the sum of squared differences between the lower-layer data and the surface data multiplied by the weight (T1), the sum of squared first-order differences in the row direction of the lower-layer data multiplied by the weight (T2), and the sum of squared first-order differences in the column direction of the lower-layer data multiplied by the weight (T3), as shown in the following formula:

[0072] ;

[0073] ;

[0074] ;

[0075] ;

[0076] where a1 is the weight for surface consistency, a2 is the row smoothness weight, and a3 is the column smoothness weight, with values ranging from 0 to 1.

[0077] The gradients of these three parts are respectively:

[0078] ;

[0079] ;

[0080] ;

[0081] The total gradient grad is the sum of three partial gradients:

[0082] .

[0083] Secondly, set the constraint conditions:

[0084] In the known lower-layer dataset is a fixed value, and during the optimization process, the values of these points remain unchanged. That is:

[0085] ;

[0086] Set the maximum number of iterations max_iter and the convergence tolerance Ctol.

[0087] fmincon is a function for solving constrained nonlinear optimization problems. It supports users to provide the objective function and the gradient function, as well as various constraint conditions, and finds the minimum value of the objective function through an optimization algorithm. The algorithm model of the present invention finally uses the fmincon function of MATLAB, based on the above objective function T and gradient function grad, and the set constraint conditions, and solves the optimization problem using the interior-point method to obtain the reconstructed lower-layer chlorophyll a concentration data C_level2.

[0088] It should be particularly noted that when reconstructing the horizontal spatial distribution dataset of chlorophyll a concentration at D_z + x which is deeper than the D_z depth, using the chlorophyll a concentration data at the D_z depth obtained as the surface data and the chlorophyll a concentration data at D_z + x as the lower-layer data to repeat the above reconstruction model, such processing can obtain the horizontal spatial distribution data result of chlorophyll a concentration at D_z + x with higher reconstruction accuracy.

[0089] Applying the above algorithm model, a numerical reconstruction was carried out on the monthly average chlorophyll a concentration level spatial distribution dataset C_level2 at a depth of 5 m underwater in the sea area of 34 - 36°N, 121 - 123°E in November 2013. The known items include the monthly average chlorophyll a concentration C_level1 on the water surface of this sea area in November 2013 obtained from MODIS satellite data, the chlorophyll a concentration values S_level2 at some points at a depth of 5 m underwater obtained from on-site experimental data in November 2013. Through the reconstruction algorithm, the monthly average chlorophyll a concentration level distribution result at a depth of 5 m underwater in November 2013 was obtained, and the result is as Figure 4 shown. Figure 4 (a) is the surface distribution result of chlorophyll a concentration in the target sea area in November 2013; Figure 4 (b) is the distribution result obtained from the on-site chlorophyll a concentration point cloud data in the target sea area in November 2013 through the Figure 2 steps shown; Figure 4 (c) is the horizontal distribution result of chlorophyll a concentration at a depth of 5 m underwater in the target sea area reconstructed by this algorithm in November 2013. It can be found that the horizontal distribution result of the underwater chlorophyll a concentration reconstructed by this algorithm ( Figure 4 (c)) has a wide coverage range and the data is continuous.

[0090] The above embodiments are only exemplary embodiments of the present invention and are not used to limit the present invention. Those skilled in the art can make various modifications or equivalent replacements within the essence and protection scope of the present invention, and such modifications or equivalent replacements should also be regarded as falling within the protection scope of the present invention.

Claims

1. A three-dimensional reconstruction method of water environment parameters based on active and passive optical remote sensing, characterized in that: The steps include: Step 1, determine the research area and time period, obtain the surface water environment parameter data C_level1, the point cloud data Chlapro at the vertical depth D_z of the water environment parameter, and record the total number of point cloud data Chlapro as num1; Step 2, matching the two sets of data obtained in step 1, obtaining the pixel point in the surface water environment parameter data that is closest to each point cloud data horizontally, and assigning a value; Step 3, reconstructing the model of the horizontal spatial distribution data of water environment parameters at the vertical depth D_z of the water body to obtain the horizontal spatial distribution data set C_level2 of the water environment parameters at the underwater depth D_z; Step 4, using the horizontal spatial distribution data of the water environment parameters at the underwater depth D_z as the surface data, and obtaining the horizontal spatial distribution data of the water environment parameters at D_z+x through the reconstruction model described in step 3; The step 3 is specifically as follows: calculating the proportion of the amount of data matched in step 2 in the total amount of pixels, if the proportion is greater than k, assigning an initial value to C_level2 so that C_level2 is equal to C_level1, and injecting the known data points in S_level2 into C_level2, constructing the objective function T based on the correlation between C_level1 and C_level2 and the smoothness of C_level2 itself, setting constraints and gradient functions, and finding the minimum value of the objective function through the optimization algorithm; If the proportion is not greater than k, take each pixel point obtained as the center point and expand to a*a pixels in all directions, intercept the data set of C_level1 in the a*a area, mark it as S_level1, and take the S_level2 data at the pixel point as T_level2, normalize the S_level1 data to obtain a weight matrix of size a*a, and use the weight matrix to reconstruct the water environment parameters of the lower layer S_level2 matrix at the corresponding a*a position to obtain S_level2'; reconstruct S_level1 according to the above method for all matching points, calculate whether the proportion of the reconstructed S_level2 valid data in the total pixel volume exceeds k, if the proportion is still not greater than k, then expand to a*a pixels in all directions with the effective data edge point of S_level2' as the center, and continue to reconstruct until the proportion of S_level2 valid data in the total pixel volume exceeds k, at this time, reassign the amount of S_level2' valid data to num1, and perform related operations with a proportion greater than k; The objective function is expressed as: Where a1 is the consistency weight with the surface layer, a2 is the row smoothing weight, and a3 is the column smoothing weight; T1 is the sum of squares of the difference between the lower layer data and the surface layer data multiplied by the weight, T2 is the sum of squares of the first-order differences in the row direction of the lower layer data multiplied by the weight, and T3 is the sum of squares of the first-order differences in the column direction of the lower layer data multiplied by the weight; m and n are the number of rows and columns of arrays C_level1 and C_level2; i and j are array subscripts, i is between [1 m], and j is between [1 n]; The gradient function is expressed as: Among them, grad is the gradient function.

2. The method for three-dimensional reconstruction of water environment parameters based on active and passive optical remote sensing according to claim 1, characterized in that: The step 1 specifically includes: the water surface water environment parameter data C_level1 is obtained through water body remote sensing reflectance spectrum data according to optical remote sensing data, and the point cloud data Chlapro at the vertical depth D_z of the water environment parameter is obtained by laser radar remote sensing data or on-site measurement; The laser radar remote sensing data is calculated through the laser attenuation coefficient of the ocean water body to obtain the point cloud data Chlapro at the vertical depth D_z of the water environment parameters.

3. The method for three-dimensional reconstruction of water environment parameters based on active and passive optical remote sensing according to claim 1, characterized in that: The assignment process in step 2 is specifically as follows: the water environment parameter data matrix at the underwater depth D_z that corresponds one-to-one to the water body surface water environment parameter data is S_level2, the initial value is NaN, and the pixel point in the water body surface water environment parameter data that is closest to each point cloud data in horizontal distance is assigned at the corresponding position of S_level2 to make it equal to the point cloud data at the vertical depth D_z of the water environment parameter.

Images