A method, system, medium, and device for remotely sensing and retrieving inherent optical properties of water bodies

Abstract

The present invention discloses a method, system, medium, and device for remote sensing inversion of inherent optical properties of water bodies, which relates to the technical field of measurement of inherent optical properties of water bodies. The method includes: obtaining the inherent optical property data and sampling point coordinates of water body samples; obtaining multispectral satellite remote sensing data according to the spatio-temporal distribution of water body samples and performing preprocessing, extracting the reflectance of each band of the corresponding pixel at the sampling point coordinates from the preprocessed data, and calculating characteristic indices; combining the inherent optical property data, the reflectance of each band, and the characteristic indices into a data set A, and dividing the data set A into a training set and a test set; constructing a deep learning model, including: an input layer, a feature extraction layer, a feature enhancement fully connected block, and an output layer; using the training set and the test set to train and test the model, and the tested model is used for predicting the inherent optical properties of water bodies. The method for predicting the inherent optical properties of water bodies of the present invention has strong versatility and stable inversion results.

Description

Technical Field

[0001] The present invention relates to the technical field of measuring inherent optical properties of water bodies, and particularly to a method, system, medium, and device for remotely sensing and inverting inherent optical properties of water bodies. Background Art

[0002] Inherent Optical Properties (IOPs) of water bodies refer to water optical parameters (such as absorption coefficient, scattering coefficient, backscattering coefficient, attenuation coefficient) that are directly related to the propagation and absorption characteristics of light. They are determined by the physical and chemical composition of the water body, reflect the optical characteristics of the water body itself, and are not affected by external light field conditions. IOPs are an important basis for understanding the water optical environment, analyzing water quality, monitoring marine ecosystems, and performing water color remote sensing inversion and other applications.

[0003] Traditional measurement and analysis of the inherent optical properties of water bodies usually involve laboratory and field measurements, combined with optical instruments and data analysis methods. Although accurate optical characteristic data can be obtained to a certain extent, since these methods usually require high input of human and material resources and are limited in space and time, they may face some challenges in practical applications.

[0004] With the development of technology, remote sensing technology provides an effective means for measuring and analyzing the inherent optical properties of water bodies, enabling efficient monitoring in the spatial and temporal dimensions, and is gradually supplementing and replacing these traditional methods. It mainly inverses the total absorption coefficient and backscattering coefficient of water bodies by using band reflectance obtained from in-situ measurements or satellites. Common inversion algorithms can be roughly divided into two categories: empirical algorithms and semi-empirical semi-analytical algorithms. Empirical algorithms, such as the well-established linear inversion model between the total absorption coefficient of water bodies, the absorption coefficient of phytoplankton, and the spectral reflectance at 440 nm constructed by Lee et al.; Li et al. established an inversion model between the backscattering coefficient at 778 nm and the band reflectance of the water body's lower surface. Statistical relationships between band reflectance and IOPs are established based on simple or multiple regression methods. This type of algorithm mostly depends on data and has poor generality.

[0005] Semi-empirical semi-analytical algorithms obtain an approximate equation of water body radiative transfer through numerical simulation methods to clarify the relationship between water body IOPs and water body component contents. Among them, Maritorena et al. developed the GSM algorithm on the basis of Garver et al. to obtain parameters such as chlorophyll concentration and backscattering coefficient, which has been applied to the global IOPs inversion of MODIS. However, this algorithm requires prior setting of relevant parameters and setting the range of measured data in a certain area. Therefore, in practical applications, the inversion results in some coastal or inland areas are not very stable, and full-band IOPs inversion cannot be achieved. Summary of the Invention

[0006] The object of the present invention is to propose a remote sensing inversion method for inherent optical properties of water bodies, including the following steps, in order to solve the problem of poor generality of the measurement method for inherent optical properties of water bodies:

[0007] S1. Obtain the data of the inherent optical properties of the water body sample and the coordinates of the sampling points;

[0008] S2. Obtain multi-spectral satellite remote sensing data and perform preprocessing. Extract the reflectance of each band of the corresponding pixel at the sampling point coordinates from the preprocessed data, and calculate the characteristic index;

[0009] S3. Combine the data of the inherent optical properties, the reflectance of each band, and the characteristic index into a data set A, and divide the data set A into a training set and a test set;

[0010] S4. Construct a deep learning model, including: an input layer, a feature extraction layer, a feature enhancement fully connected block, and an output layer;

[0011] Using the data set A as the input data, the input layer receives the input data, and the wavelength information in the input data is extracted by the feature extraction layer;

[0012] The feature enhancement fully connected block receives the input data and the wavelength information. After being processed by N feature enhancement fully connected blocks, the obtained feature vector passes through the output layer, and the prediction of the inherent optical properties of the water body in a given band is output;

[0013] Among them, the feature enhancement fully connected block includes: a fully connected layer and a splicing layer;

[0014] S5. Use the training set and the test set to train and test the model, and the tested model is used for the prediction of the inherent optical properties of the water body.

[0015] Furthermore, perform radiometric calibration, atmospheric correction, orthorectification, and water body masking processing on the multi-spectral satellite remote sensing data to obtain the preprocessed data.

[0016] Furthermore, the reflectance of each band includes: the reflectance of the red, green, blue, near-infrared, and short-wave infrared bands.

[0017] Furthermore, the characteristic index includes: the normalized difference water index, the green difference water index, the environmental water index, the water body relative index, the phytoplankton index, the normalized difference humidity index, the land surface moisture index, and the turbidity index;

[0018] The calculation formulas are as follows:

[0019] ;

[0020] ;

[0021] ;

[0022] ;

[0023] ;

[0024] ;

[0025] ;

[0026] ;

[0027] Among them, NDWI is the Normalized Difference Water Index, NDGI is the Normalized Difference Greenness Index, EWI is the Environmental Water Index, WRI is the Water Relative Index, ChI is the Phytoplankton Index, NDMI is the Normalized Difference Moisture Index, LSWI is the Land Surface Water Index, and TI is the Turbidity Index; represents the reflectance of the green light band, represents the reflectance of the near-infrared light band, represents the reflectance of the red light band, represents the reflectance of the blue light band, represents the reflectance of the short-wave infrared light band.

[0028] Furthermore,

[0029] The first fully connected layer of the feature enhancement fully connected block processes the input data. The first concatenation layer of the feature enhancement fully connected block concatenates the output of the first fully connected layer and the wavelength information output by the feature extraction layer to obtain the output of the first feature enhancement fully connected block. The input of the second to the Nth fully connected layers is the output of the previous feature enhancement fully connected block. The second to the Nth concatenation layers concatenate the wavelength information output by the feature extraction layer and the output of the current fully connected layer to obtain the output of the second to the Nth feature enhancement fully connected blocks.

[0030] The present invention also proposes a remote sensing inversion system for inherent optical properties of water bodies, including:

[0031] A first data acquisition module, configured to acquire the inherent optical property data and sampling point coordinates of a water body sample;

[0032] A second data acquisition module, configured to acquire multi-spectral satellite remote sensing data and perform preprocessing, extract the reflectances of each band of the pixels corresponding to the sampling point coordinates from the preprocessed data, and calculate the characteristic indices;

[0033] A data partitioning module, configured to combine the inherent optical property data, the reflectances of each band, and the characteristic indices into a data set A, and partition the data set A into a training set and a test set;

[0034] A model construction module for constructing a deep learning model, including: an input layer, a feature extraction layer, a feature enhancement fully connected block, and an output layer;

[0035] Taking dataset A as input data, the input layer receives the input data, and the wavelength information in the input data is extracted by the feature extraction layer;

[0036] The feature enhancement fully connected block receives the input data and the wavelength information. After being processed by N feature enhancement fully connected blocks, the obtained feature vector passes through the output layer to output the prediction of the inherent optical properties of water bodies in a given band;

[0037] Among them, the feature enhancement fully connected block includes: a fully connected layer and a splicing layer;

[0038] A model training and testing module for training and testing the model using a training set and a testing set. The tested model is used for the prediction of the inherent optical properties of water bodies.

[0039] The present invention also proposes a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the above-mentioned remote sensing inversion method for the inherent optical properties of water bodies is realized.

[0040] The present invention also proposes an electronic device, including a processor and a memory, the processor is interconnected with the memory, wherein the memory is used for storing a computer program, the computer program includes computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the above-mentioned remote sensing inversion method for the inherent optical properties of water bodies.

[0041] The beneficial effects brought by the technical solution provided by the present invention are:

[0042] The present invention introduces multiple exponential features, which to a certain extent reduces the influence of spatio-temporal factors on data analysis, suppresses the noise commonly existing in each band, effectively improves the model's data analysis ability for data at different times and different locations, enhances the model's prediction performance, enhances the robustness of the model, and the inversion result is stable. And through the feature enhancement fully connected block, the band information is enhanced during model training, effectively realizing the prediction of the inherent optical properties of water bodies in the full band, and reducing the model's dependence on data and parameters. Description of the Drawings

[0043] Figure 1 is a flowchart of the remote sensing inversion method for the inherent optical properties of water bodies in an embodiment of the present invention;

[0044] Figure 2 is a structural diagram of the deep learning model constructed in an embodiment of the present invention;

[0045] Figure 3It is a block diagram of an electronic device in an exemplary embodiment of Embodiment 1 of the present invention;

[0046] Figure 4 They are the predicted value and the true value of the total particulate backscattering coefficient obtained by using the validation set of the model in the embodiment of the present invention. Detailed implementation manners

[0047] To make the objectives, technical solutions and advantages of the present invention clearer, the embodiments of the present invention will be further described below in conjunction with the accompanying drawings.

[0048] The flowchart of the method for remotely sensing and inverting the inherent optical properties of water bodies in the embodiment of the present invention is as Figure 1 , and specifically includes the following steps:

[0049] S1. Obtain the inherent optical property data and sampling point coordinates of the water body samples. Select the experimental data of n water body samples, and the sample collection locations are , and the corresponding inherent optical properties of the water body are , and the collection date is . Among them , represents the wavelength value.

[0050] The inherent optical property refers to the quantity that is only related to the water body components and does not change with the illumination conditions, including:

[0051] (1) The absorption coefficient, scattering coefficient, and scattering phase function of water molecules;

[0052] (2) The absorption coefficient, unit absorption coefficient, scattering coefficient, unit scattering coefficient, backscattering coefficient, forward scattering coefficient, and scattering phase function of Chl-a;

[0053] (3) The unit absorption coefficient of yellow substances;

[0054] (4) Other components, including the absorption and scattering characteristics of inorganic substances, debris, etc.

[0055] S2. Obtain the multi-spectral satellite remote sensing data and perform preprocessing. Extract the reflectance of each band of the corresponding pixel at the sampling point coordinates from the preprocessed data, and calculate the characteristic index.

[0056] According to the selected spatio-temporal distribution of the water body samples, obtain the multi-spectral images taken by the same multi-spectral remote sensing satellite at the corresponding date and location. Use ENVI5.3 or other remote sensing professional software to complete radiometric calibration, atmospheric correction, orthorectification, and water body masking processing on the multi-spectral images; or add water body masking processing through band analysis on the existing image preprocessing products to obtain the preprocessed data.

[0057] According to the geographical location of the water sample, find the corresponding pixel position in the image, and read the pixel values of the red, green, blue, near-infrared, and short-wave infrared bands of this pixel. Obtain the reflectance of the red, green, blue, near-infrared, and short-wave infrared bands.

[0058] The characteristic indices include: Normalized Difference Water Index, Green Difference Water Index, Environmental Water Index, Water Relative Index, Phytoplankton Index, Normalized Difference Moisture Index, Land Surface Water Index, Turbidity Index;

[0059] The calculation formulas are as follows:

[0060] ;

[0061] ;

[0062] ;

[0063] ;

[0064] ;

[0065] ;

[0066] ;

[0067] ;

[0068] Among them, NDWI is the Normalized Difference Water Index, NDGI is the Green Difference Water Index, EWI is the Environmental Water Index, WRI is the Water Relative Index, ChI is the Phytoplankton Index, NDMI is the Normalized Difference Moisture Index, LSWI is the Land Surface Water Index, and TI is the Turbidity Index; represents the reflectance of the green band, represents the reflectance of the near-infrared band, represents the reflectance of the red band, represents the reflectance of the blue band, represents the reflectance of the short-wave infrared band.

[0069] S3. Combine the inherent optical quantity data, the reflectance of each band, and the characteristic indices into dataset A, and divide dataset A into a training set and a test set.

[0070] Dataset , where:

[0071] .

[0072] S4. Construct a deep learning model, including: an input layer, a feature extraction layer, a feature enhancement fully connected block, and an output layer.

[0073] Using dataset A as input data, the input layer receives the input data, and the feature extraction layer extracts the wavelength information in the input data.

[0074] The feature enhancement fully connected block receives the input data and the wavelength information. After being processed by N feature enhancement fully connected blocks, the obtained feature vector passes through the output layer to output the prediction of the inherent optical properties of water in a given band.

[0075] Among them, the feature enhancement fully connected block includes: a fully connected layer and a splicing layer.

[0076] The first fully connected layer of the feature enhancement fully connected block processes the input data. The first splicing layer of the feature enhancement fully connected block splices the output of the first fully connected layer and the wavelength information output by the feature extraction layer to obtain the output of the first feature enhancement fully connected block. The input of the second to the Nth fully connected layers is the output of the previous feature enhancement fully connected block. The second to the Nth splicing layers splice the wavelength information output by the feature extraction layer and the output of the current fully connected layer to obtain the output of the second to the Nth feature enhancement fully connected blocks.

[0077] The constructed deep learning model of the present invention refers to Figure 2 .

[0078] S5. Use the training set and the test set to train and test the model. The tested model is used for the prediction of the inherent optical properties of water.

[0079] The K-fold cross-validation method is used to train the model, where K = 4, that is, the training set is randomly divided into 4 subsets for multiple training and validation cycles. In each cycle, 25% of the subsets are used as the validation set, and a set of model results with the best evaluation index in the cycle is selected. During training, the model is optimized for parameters through the Adam optimizer. The loss function is the mean squared error function (Mean Squared Error, MSE), and the evaluation index uses the mean absolute error function (Mean Absolute Error, MAE). The formulas are as follows:

[0080] ;

[0081] ;

[0082] Among them, k is the number of test samples, is the actual value of the test sample, is the predicted value of the test sample.

[0083] In an exemplary embodiment, it includes a remote sensing inversion system for inherent optical properties of water bodies, comprising:

[0084] A first data acquisition module, configured to acquire the inherent optical property data of the water body sample and the sampling point coordinates;

[0085] A second data acquisition module, configured to acquire multi-spectral satellite remote sensing data and perform preprocessing, extract the reflectance of each band of the pixel corresponding to the sampling point coordinates from the preprocessed data, and calculate the characteristic index;

[0086] A data division module, configured to combine the inherent optical property data, the reflectance of each band, and the characteristic index into a data set A, and divide the data set A into a training set and a test set;

[0087] A model construction module, configured to construct a deep learning model, including: an input layer, a feature extraction layer, a feature enhancement fully connected block, and an output layer;

[0088] Taking the data set A as input data, the input layer receives the input data, and the wavelength information in the input data is extracted by the feature extraction layer;

[0089] The feature enhancement fully connected block receives the input data and the wavelength information. After being processed by N feature enhancement fully connected blocks, the obtained feature vector passes through the output layer, and the prediction of the inherent optical property of the water body in a given band is output;

[0090] Among them, the feature enhancement fully connected block includes: a fully connected layer and a splicing layer;

[0091] A model training and testing module, configured to train and test the model using the training set and the test set, and the tested model is used for the prediction of the inherent optical property of the water body.

[0092] In an exemplary embodiment, it includes a computer-readable storage medium, and the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned remote sensing inversion method for the inherent optical property of the water body is implemented.

[0093] Please refer to Figure 3 , in an exemplary embodiment, it further includes an electronic device, including at least one processor, at least one memory, and at least one communication bus.

[0094] Among them, a computer program is stored on the memory, the computer program includes computer-readable instructions, and the processor calls the computer-readable instructions stored in the memory through the communication bus to execute the above-mentioned remote sensing inversion method for the inherent optical property of the water body.

[0095] In the embodiment of the present invention, 600 groups of inherent optical property sample data of water bodies are selected, and the multi-spectral band images of the Sentinel-2 satellite are collected according to the spatio-temporal information of the samples. (including blue band, green band, red band, near-infrared band, short-wave infrared band).

[0096] Use the GEE (Google Earth Engine) platform to extract the reflectance of red, green, blue, near-infrared, and short-wave infrared bands at the spatio-temporally matched pixels of the sampling points .

[0097] And calculate the index features according to the formula, and finally obtain 200 data sets .

[0098] Among them:

[0099] .

[0100] Randomly divide the data set A into a training set and a validation set according to a ratio of 8:2. The training set has a total of 480 samples, and the test set has a total of 120 samples.

[0101] Define an input layer during model construction to accept the data set ; Build a custom feature extraction layer to separately extract band information; build three feature enhancement fully connected blocks to extract deep spectral features, and build an output layer to predict the inherent optical quantity of water bodies in the specified band, taking the total particle backscattering coefficient as an example.

[0102] Input 480 training samples into the model for training, and use the K-fold cross-validation method, where K = 4. The batch size during the training process is 16, the number of iterations is 500, the initial learning rate is set to 0.001, and the learning rate self-adjustment strategy is enabled. The model optimizes the parameters through the Adam optimizer, the loss function is the mean squared error, and the early stopping method is used to stop training in advance when the performance on the validation set no longer improves, to avoid overfitting caused by excessive number of iterations.

[0103] Input 120 validation samples into the trained model for prediction, and use the root mean square error to evaluate the prediction performance. The predicted value and the true value of the total particle backscattering coefficient obtained by the model in the embodiment of the present invention using the validation set are as Figure 4 shown. Figure 4 Among them, taking the true value of the total particle backscattering coefficient as the abscissa and the corresponding predicted value as the ordinate, the ideal predicted value should be equal to the true value, so the ideal predicted value and the true value are a straight line with a slope of 1 in this coordinate system, as Figure 4 shown by the dashed line in. In fact, when taking the true value as the abscissa, the straight line fitted by the predicted value has a certain deviation from the ideal one, as Figure 4 shown by the solid line in.

[0104] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A remote sensing inversion method for intrinsic optical quantities of water bodies, characterized in that: The following steps are involved: S1. Obtain the intrinsic optical quantity data and sampling point coordinates of water samples; S2. Obtain multispectral satellite remote sensing data and preprocess them according to the temporal and spatial distribution of water samples, extract the reflectance of each band of the corresponding pixel at the sampling point coordinates from the preprocessed data, and calculate the characteristic index; The characteristic indexes include: normalized difference water index, green difference water index, environmental water index, water relative index, phytoplankton index, normalized difference humidity index, land surface moisture index, and turbidity index; The calculation formula is as follows: ； ； ； ； ； ； ； ； Among them, NDWI is the normalized difference water index, NDGI is the green difference water index, EWI is the environmental water index, WRI is the water relative index, ChI is the phytoplankton index, NDMI is the normalized difference moisture index, LSWI is the land surface moisture index, and TI is the turbidity index; represents the reflectivity of the green light band, represents the reflectivity in the near-infrared band, represents the reflectivity of the red light band, Represents the reflectivity of the blue light band, Indicates the reflectivity of the short-wave infrared light band; S3, combining the intrinsic optical quantity data, the reflectivity of each band, and the characteristic index into a data set A, and dividing the data set A into a training set and a test set; S4. Build a deep learning model, including: input layer, feature extraction layer, feature enhancement fully connected block, and output layer; Dataset A is used as input data. The input layer accepts the input data, and the feature extraction layer extracts the wavelength information in the input data. The feature enhancement fully connected block accepts input data and wavelength information, and after being processed by N feature enhancement fully connected blocks, the feature vector is passed through the output layer to output the prediction of the inherent optical quantity of the water body in a given band; Among them, the feature enhancement fully connected block includes: a fully connected layer and a splicing layer; S5. Use the training set and the test set to train and test the model. The tested model is used to predict the inherent optical quantity of the water body.

2. A method for remote sensing inversion of intrinsic optical quantities of water bodies according to claim 1, characterized in that: The multispectral satellite remote sensing data is subjected to radiation calibration, atmospheric correction, orthorectification and water body mask processing to obtain preprocessed data.

3. A method for remote sensing inversion of intrinsic optical quantities of water bodies according to claim 1, characterized in that: The reflectivity of each band includes: reflectivity of red light, green light, blue light, near infrared light, and short-wave infrared light bands.

4. The method for remote sensing inversion of intrinsic optical quantities of water bodies according to claim 1, characterized in that: The first fully connected layer of the feature enhanced fully connected block processes the input data, and the first splicing layer of the feature enhanced fully connected block splices the output of the first fully connected layer and the wavelength information output by the feature extraction layer to obtain the output of the first feature enhanced fully connected block. The input of the second to Nth fully connected layers is the output of the previous feature enhanced fully connected block. The second to Nth splicing layers splice the wavelength information output by the feature extraction layer and the output of the current fully connected layer to obtain the output of the second to Nth feature enhanced fully connected blocks.

5. A remote sensing inversion system for inherent optical quantities of water bodies, characterized in that: include: The first data acquisition module is used to obtain the intrinsic optical quantity data and sampling point coordinates of the water sample; The second data acquisition module is used to acquire and preprocess multispectral satellite remote sensing data, extract the reflectance of each band of the corresponding pixel at the sampling point coordinates from the preprocessed data, and calculate the characteristic index; The characteristic indexes include: normalized difference water index, green difference water index, environmental water index, water relative index, phytoplankton index, normalized difference humidity index, land surface moisture index, and turbidity index; The calculation formula is as follows: ； ； ； ； ； ； ； ； Among them, NDWI is the normalized difference water index, NDGI is the green difference water index, EWI is the environmental water index, WRI is the water relative index, ChI is the phytoplankton index, NDMI is the normalized difference moisture index, LSWI is the land surface moisture index, and TI is the turbidity index; represents the reflectivity of the green light band, represents the reflectivity in the near-infrared band, represents the reflectivity of the red light band, Represents the reflectivity of the blue light band, Indicates the reflectivity of the short-wave infrared light band; A data partitioning module is used to combine the intrinsic optical quantity data, the reflectivity of each band, and the characteristic index into a data set A, and divide the data set A into a training set and a test set; Model building module, used to build deep learning models, including: input layer, feature extraction layer, feature enhancement fully connected block, output layer; Dataset A is used as input data. The input layer accepts the input data, and the feature extraction layer extracts the wavelength information in the input data. The feature enhancement fully connected block accepts input data and wavelength information, and after being processed by N feature enhancement fully connected blocks, the feature vector is passed through the output layer to output the prediction of the inherent optical quantity of the water body in a given band; Among them, the feature enhancement fully connected block includes: a fully connected layer and a splicing layer; The model training and testing module is used to train and test the model using the training set and the test set. The tested model is used to predict the inherent optical quantity of the water body.

6. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

7. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the method according to any one of claims 1 to 4.

Images