Method and system for improving satellite remote sensing inversion accuracy by integrating hyperspectral near-field sensing

Abstract

The present invention discloses a method and system for improving the accuracy of satellite remote sensing inversion by fusing hyperspectral near-sensing, and belongs to the technical field of water environment and water ecology monitoring. A water body hyperspectral near-sensing reflectance spectral data set, a satellite remote sensing surface reflectance data set and a water environment measured data set are obtained and preprocessed; a synchronous data set containing water quality and water quality is constructed; based on the synchronous data with water quality, synchronous data with water quality is generated, and synchronous data with water quality is obtained; for normalization, a quantitative conversion relationship is constructed band by band to eliminate errors; a fused satellite-ground data set is generated, and a training water quality parameter inversion model is constructed; the corresponding method and data set are updated in real time by repeating the specified steps to optimize the model. The present invention displays and downloads the inversion results according to the query instruction, improves the satellite spectral accuracy and the scale of the satellite-ground synchronous data set, and is used for high-precision water environment monitoring.

Description

Technical Field

[0001] The present invention belongs to the technical field of water environment and water ecology monitoring, and specifically relates to a method and system for integrating hyperspectral near-field sensing to improve the accuracy of satellite remote sensing inversion. Background Art

[0002] Traditional water quality monitoring mainly relies on manual collection and laboratory measurements, which is time-consuming and labor-intensive, and its timeliness cannot meet the needs of large-scale water quality monitoring, especially for rapidly changing water bodies. Satellite remote sensing is widely used in water environment inversion monitoring due to its large scale, periodicity and cost-effectiveness. However, due to the influence of weather and incomplete atmospheric correction, the accuracy of satellite spectra still needs to be further improved. In addition, since satellite remote sensing has the characteristics of large-scale instantaneous collection, but the water quality of rivers and lakes has highly dynamic changes, it is difficult to obtain quasi-synchronous satellite-ground data that is consistent in time and space. Usually, 3 satellite data before and after the satellite passes are used. h Matching with ground data.

[0003] Furthermore, the accuracy and generalizability of water quality remote sensing inversion models depend on the quantity and diversity of quasi-synchronous satellite-ground datasets. However, geosynchronous data, which contain a large amount of diversity, is limited by the availability of ground sampling personnel and costs, making it difficult to obtain a large number of rigorously diverse synchronized water samples. In summary, using satellite remote sensing to monitor water quality currently still suffers from shortcomings such as low atmospheric correction accuracy, inconsistent satellite-ground synchronization times, and limited data and diversity in strictly synchronized satellite-ground datasets. Summary of the Invention

[0004] Purpose of the invention: To address the above problems, the present invention provides a method and system for integrating hyperspectral near-field sensing to improve the accuracy of satellite remote sensing inversion.

[0005] The present invention adopts the following technical solution: a method for improving the accuracy of satellite remote sensing inversion by integrating hyperspectral near-field sensing, comprising the following steps:

[0006] Step 1: Obtain a water body hyperspectral near-sensing reflectance spectral dataset, a satellite remote sensing surface reflectance dataset, and a water environment measured dataset, and use corresponding data preprocessing methods for preprocessing;

[0007] Step 2: Determine the best matching window using the time matching principle and construct a synchronous data set; the synchronous data set includes: Synchronous data with water quality, Synchronous data with water quality; wherein, is the pre-processed satellite remote sensing surface reflectance data, is the pre-processed water body hyperspectral near-sensing reflectance spectrum data, For the band;

[0008] Step 3: Create a hyperspectral conversion multispectral data simulation method based on the Corresponding to the synchronous data generation of water quality Synchronous data with water quality and get and The synchronization data, is the simulated satellite remote sensing surface reflectance;

[0009] Step 4: Based on and The pre-processed satellite remote sensing surface reflectance data is synchronized with the data Perform normalization processing to obtain new satellite remote sensing surface reflectance data , the new satellite remote sensing surface reflectivity data Compared with the simulated satellite remote sensing surface reflectivity Conduct comparative analysis and build band by band and Quantitative conversion relationship between them to eliminate errors;

[0010] Step 5: Synchronous data with water quality and Combined with the synchronous data of water quality to generate a fused satellite-ground dataset; With synchronized water quality data and fused satellite-ground datasets as input features and water environment measured datasets as output features, a water quality parameter inversion model is constructed and trained;

[0011] Step 6. Repeat steps 3 to 5, update the hyperspectral conversion multispectral data simulation method, satellite spectral accuracy band improvement method, and fusion satellite-ground data set in real time, continuously optimize the water quality parameter inversion model, and display and download the inversion results according to the query instruction until the user's query end instruction is received.

[0012] In a further embodiment, the data preprocessing step of the water body hyperspectral near-sensing reflectance spectrum dataset includes:

[0013] The water body hyperspectral near-sensing reflectance spectral dataset was obtained by continuous collection under various weather conditions. , calculate the water body hyperspectral near-sensing reflectance spectrum for each band and 420-830 nm Average value of inner hyperspectral near-sensing reflectance The ratio of the normalized hyperspectral near-sensing reflectance corresponding to the band is obtained. :

[0014] ;

[0015] Refer to the spectral data of synchronous ground object spectrum Perform radiation correction, the calculation formula is as follows:

[0016] ;

[0017] Where, is the radiation gain value of the land-based hyperspectral proximity sensor, is the radiation deviation value of the land-based hyperspectral proximity sensor.

[0018] In a further embodiment, the data preprocessing step of the satellite remote sensing surface reflectance dataset includes:

[0019] Obtain satellite remote sensing surface reflectance dataset , extract the quality band qa 60, according to the bit operation principle, the pixel value of the cloud mask file is obtained :

[0020] Using bitwise operations and quality bands qa 60 Generate opaque cloud mask files and cirrus cloud mask files respectively, use and The () function merges the opaque cloud mask file and the cirrus cloud mask file to generate a cloud mask file, thereby obtaining the pixel value of the mask file ;

[0021] The following formula is used to obtain the preprocessed satellite remote sensing surface reflectance data :

[0022] , is the surface reflectivity from satellite remote sensing.

[0023] In a further embodiment, the process of determining the best matching window is as follows:

[0024] The sum of the coefficients of determination of all bands is defined as S , used to filter out the optimal time matching window, the calculation formula is as follows:

[0025] ;

[0026] in, 、 Bands The coefficient of determination of S The time matching window at the maximum value is the best matching window.

[0027] In a further embodiment, the formula of the hyperspectral conversion multispectral data simulation method is as follows:

[0028] , ;

[0029] Where, is the pre-processed water body hyperspectral near-sensing reflectance spectrum data, For Spectral response function of the satellite sensor at .

[0030] In a further embodiment, the pre-processed satellite remote sensing surface reflectivity data The calculation formula for the normalization process is as follows:

[0031] ;

[0032] in, , Indicates the band is 443 nm Satellite remote sensing surface reflectivity data, and so on.

[0033] In a further embodiment, the and The quantitative conversion relationship between them is expressed as follows:

[0034] ;

[0035] in, is the linear regression slope, indicating and The gain between is a constant.

[0036] In a further embodiment, the step of eliminating the error comprises:

[0037] Build a linear regression based on the slope The error analysis principle is based on the error analysis principle. and The error situation between them is analyzed, and targeted optimization is performed according to the error situation;

[0038] The error analysis principle is:

[0039] ;

[0040] When the above-mentioned overestimation or underestimation occurs, the correction accuracy of the satellite sensor facing the water body is further optimized.

[0041] In a further embodiment, the water quality parameter inversion model adopts a machine learning model, and the machine learning model includes at least: a deep neural network model, a random forest model, a support vector machine model, an extreme gradient tree model, a Gaussian regression model and a backward transmission neural network model.

[0042] The system for improving satellite remote sensing inversion accuracy by fusing hyperspectral near-field sensing is used to implement the above-mentioned method for improving satellite remote sensing inversion accuracy by fusing hyperspectral near-field sensing, including:

[0043] The near-sensing hyperspectral sensor, satellite sensor and water quality detector are configured to obtain a water body hyperspectral near-sensing reflectance spectral dataset, a satellite remote sensing surface reflectance dataset and a water environment measured dataset respectively;

[0044] The first module is configured to select corresponding data processing methods to preprocess the water hyperspectral near-sensing reflectance spectral dataset, satellite remote sensing surface reflectance dataset, and water environment measured dataset respectively;

[0045] The second module is configured to determine the best matching window and construct a synchronous data set using a time matching principle; the synchronous data set includes: Synchronous data with water quality, Synchronous data with water quality; wherein, is the pre-processed satellite remote sensing surface reflectance data, is the pre-processed water body hyperspectral near-sensing reflectance spectrum data, For the band;

[0046] The third module is configured to create a hyperspectral conversion multispectral data simulation method based on the Corresponding to the synchronous data generation of water quality Synchronous data with water quality and get and The synchronization data, is the simulated satellite remote sensing surface reflectance;

[0047] The fourth module is set up based on and The pre-processed satellite remote sensing surface reflectance data is synchronized with the data Perform normalization processing to obtain new satellite remote sensing surface reflectance data , the new satellite remote sensing surface reflectivity data Compared with the simulated satellite remote sensing surface reflectivity Conduct comparative analysis and build band by band and Quantitative conversion relationship between them to eliminate errors;

[0048] The fifth module is configured to Synchronous data with water quality and Combined with the synchronous data of water quality to generate a fused satellite-ground dataset; With synchronized water quality data and fused satellite-ground datasets as input features and water environment measured datasets as output features, a water quality parameter inversion model is constructed and trained;

[0049] The sixth module is configured to repeat steps three to five, updating the hyperspectral conversion multispectral data simulation method, the satellite spectral accuracy band improvement method, and the fusion of satellite and ground data sets in real time to continuously optimize the water quality parameter inversion model;

[0050] The terminal display module is configured to display and download the inversion results according to the query instruction until receiving the user's query end instruction.

[0051] Beneficial effects of the present invention: The present invention demonstrates to users a method for improving satellite water quality remote sensing inversion based on near-ground hyperspectral data, provides an optimal time window for matching satellite and measured water quality data, and ensures that the spectrum obtained by users is completely one-to-one corresponding to the measured water quality.

[0052] The present invention expands the scale of satellite and ground synchronous measured data through simple and easy-to-operate hyperspectral near-sensing data simulation, improves the diversity and complexity of modeling data, ensures the stability and robustness of the model, reduces the difficulty for users to obtain satellite-ground synchronous data sets, and improves model accuracy.

[0053] This method enables the updating, querying, statistics, uploading, and downloading of water quality information, ensuring users have access to water quality information at multiple temporal and spatial scales, assisting them in decision-making. Compared to traditional manual monitoring, which suffers from information lag, temporal and spatial discreteness, and a lack of long-term and multi-spatial variation characteristics, this method offers a more convenient and efficient method and system.

[0054] The present invention supports interactive query of a web page interface, and supports and assists users in obtaining water quality information of their interest without being disturbed by other information. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a flow chart of the method for fusing hyperspectral near-field sensing to improve satellite remote sensing inversion accuracy in Example 1.

[0056] Figure 2 3 is a response function diagram of the hyperspectral conversion multispectral data simulation of Example 1.

[0057] Figure 3 The synchronization of Example 1 and Analytical diagram of consistency between .

[0058] Figure 4 Example 1 and Quantitative conversion relationship diagram between .

[0059] Figure 5 The chlorophyll of Example 1 a Schematic diagram of the inversion model. DETAILED DESCRIPTION

[0060] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0061] Example 1

[0062] Inland water bodies are complex and diverse. In order to improve the accuracy of satellite spectra, enhance the consistency of satellite-ground spectra time matching, and expand the number and diversity of synchronized data, this embodiment provides a method for integrating hyperspectral near-field sensing to improve the accuracy of satellite remote sensing inversion (hereinafter referred to as this method). Figure 1 The following steps are shown:

[0063] Step 1: Obtain a water body hyperspectral near-sensing reflectance spectral dataset, a satellite remote sensing surface reflectance dataset, and a water environment measured dataset, and use the corresponding data preprocessing method for preprocessing.

[0064] Step 2: Determine the best matching window using the time matching principle and construct a synchronous data set; the synchronous data set includes: Synchronous data with water quality, Synchronous data with water quality; wherein, is the pre-processed satellite remote sensing surface reflectance data, is the pre-processed water body hyperspectral near-sensing reflectance spectrum data, For the band;

[0065] Step 3: Create a hyperspectral conversion multispectral data simulation method based on the Corresponding to the synchronous data generation of water quality Synchronous data with water quality and get and The synchronization data, is the simulated satellite remote sensing surface reflectance;

[0066] Step 4: Based on and The pre-processed satellite remote sensing surface reflectance data is synchronized with the data Perform normalization processing to obtain new satellite remote sensing surface reflectance data , the new satellite remote sensing surface reflectivity data Compared with the simulated satellite remote sensing surface reflectivity Conduct comparative analysis and build band by band and Quantitative conversion relationship between them to eliminate errors;

[0067] Step 5: Synchronous data with water quality and Combined with the synchronous data of water quality to generate a fused satellite-ground dataset; With synchronized water quality data and fused satellite-ground datasets as input features and water environment measured datasets as output features, a water quality parameter inversion model is constructed and trained;

[0068] Step 6. Repeat steps 3 to 5, update the hyperspectral conversion multispectral data simulation method, satellite spectral accuracy band improvement method, and fusion satellite-ground data set in real time, continuously optimize the water quality parameter inversion model, and display and download the inversion results according to the query instruction until the user's query end instruction is received.

[0069] The following describes steps 1 through 6 in detail. In this embodiment, the water hyperspectral near-field reflectance spectral dataset, satellite remote sensing surface reflectance dataset, and water environment measurement dataset all include basic data information, such as the latitude and longitude of the collection point, the collection time (accurate to the minute), the type of water body collected, and weather conditions. Furthermore, weather conditions include sunny, overcast, partly cloudy, and rainy days, as well as wind speed and other information.

[0070] Based on the above basic data information, the water body hyperspectral near-sensing reflectance spectral dataset is obtained by near-sensing hyperspectral sensors at 2 to 10 meters above the water surface. m The ratio of the irradiance on the water surface up to that on the water surface down is obtained from the height of the data. Among them, the near-sensing hyperspectral sensor includes at least: a ground object spectrometer and a land-based hyperspectral near-sensing monitoring instrument ( GBPSs ) two monitoring devices. The satellite remote sensing surface reflectance dataset is a multi-band dataset collected by a satellite-borne platform covering visible light to near-infrared bands, which has been corrected by radiation, geometry and atmosphere. L Level 2 surface reflectance. Finally, the water environment measured data set includes at least chlorophyll a , total suspended matter concentration, transparency, diffuse attenuation coefficient and absorption coefficient of colored soluble organic matter and other optical activity parameters.

[0071] For example, the satellite sensor of this embodiment selects Sentinel-2 The near-sensing hyperspectral sensor uses a land-based hyperspectral near-sensing monitor, and the water quality parameter is chlorophyll. a .

[0072] Based on the above description, considering the differences in the acquisition environments and acquisition parameters of the water hyperspectral near-sensing reflectance spectral dataset, satellite remote sensing surface reflectance dataset, and water environment measured dataset, this embodiment uses different data preprocessing methods for the water hyperspectral near-sensing reflectance spectral dataset, satellite remote sensing surface reflectance dataset, and water environment measured dataset. Specifically, the data preprocessing steps of the water hyperspectral near-sensing reflectance spectral dataset include:

[0073] The water body hyperspectral near-sensing reflectance spectral dataset was obtained by continuous collection under various weather conditions. In order to avoid the inconsistency of hyperspectral near-sensing reflectance in complex light environments caused by the sun’s altitude, light intensity and water surface waves, it is necessary to analyze the hyperspectral near-sensing reflectance spectrum of water bodies. Normalization is performed. Therefore, the water body hyperspectral near-sensing reflectance spectrum of each band is calculated and 420-830 nm Average value of inner hyperspectral near-sensing reflectance The ratio of the normalized hyperspectral near-sensing reflectance corresponding to the band is obtained. :

[0074] ;

[0075] In order to further improve the accuracy and reliability of the hyperspectral near-sensing band, the spectral data of the synchronous ground object spectrum are used to compare the Perform radiation correction, the calculation formula is as follows:

[0076] ;

[0077] Where, is the radiation gain value of the land-based hyperspectral proximity sensor, is the radiation deviation value of the land-based hyperspectral proximity sensor. In this embodiment, The value of is 1.086, The value of is 0.0054.

[0078] It is worth mentioning that the average value of the hyperspectral near-sensing reflectance mentioned above is The calculation formula is as follows:

[0079] Where, For band 420 nm The water body hyperspectral near-sensing reflectance, and so on .

[0080] Correspondingly, the data preprocessing steps for satellite remote sensing surface reflectance datasets include:

[0081] Obtain satellite remote sensing surface reflectance dataset , extract the quality band qa 60, according to the bit operation principle, the pixel value of the cloud mask file is obtained :

[0082] Using bitwise operations and quality bands qa 60 Generate opaque cloud mask files and cirrus cloud mask files respectively, use and The () function merges the opaque cloud mask file and the cirrus cloud mask file to generate a cloud mask file, thereby obtaining the pixel value of the mask file .

[0083] In a further embodiment, the opaque cloud mask file is generated by shifting the binary bitwise number 1 to the left by 10 bits, and then Bit 10th digit (the 11th digit from right to left) and qa The band value of 60 (after conversion to binary bit operation) Bit The 10th bit (the 11th number from the right to the left) is compared: if the two are the same, it means that there is cloud in the pixel and it is thick cloud, and the pixel value of the pixel is assigned to 1; if the two are different, the pixel value of the pixel is assigned to 0, and finally an opaque cloud mask file is generated.

[0084] Correspondingly, the steps for generating the cirrus cloud mask file are as follows: shift the binary bit operation number 1 to the left by 11 bits, and then Bit 11th digit (the 12th digit from right to left) and qa The band value of 60 (after conversion to binary bit operation) Bit The 11th bit (the 12th number from the right to the left) is compared: if the two are the same, it means that there is cloud in the pixel and it is thin cloud, so the pixel value of the pixel is assigned to 1; if the two are different, the pixel value of the pixel is assigned to 0, and finally the cirrus cloud mask file is generated.

[0085] The opaque cloud mask file and cirrus cloud mask file obtained by the above method are used and () function is used to merge and obtain the cloud mask file, and the corresponding assignment is used to obtain the pixel value of the mask file The specific code is expressed as:

[0086] qa 60. bitwiseAnd (1<<10). eq (0). and ( qa 60. bitwiseAnd (1<<11). eq (0)).

[0087] The following formula is used to obtain the preprocessed satellite remote sensing surface reflectance data :

[0088] , is the surface reflectivity from satellite remote sensing.

[0089] Furthermore, the data preprocessing method for the water environment measured data set includes: cleaning abnormal data below zero. Specifically, the chlorophyll measured in the laboratory is a Negative and zero values ​​in the concentration were deleted as outliers. In addition, in order to further eliminate the data anomalies caused by non-standard sampling, the chlorophyll content was compared with the field sampling photos and laboratory measurements. a Concentration comparison, when the bloom intensity increases chlorophyll a If the height is obviously too low, it should be removed.

[0090] Traditionally, satellite sensors are usually used to measure ±3 h ~±7 d However, because water quality is subject to dynamic changes due to climate and river runoff, the inconsistency between satellite data and ground-based data acquisition time may cause a mismatch between the water body spectrum and the satellite spectrum, directly affecting the accuracy and stability of the water quality inversion model.

[0091] Furthermore, the time windows are extracted based on the satellite sensor transit time as a reference. 、 、 、 、 、 、 The water body hyperspectral near-sensing reflectance, where h Indicates hours, d Represents day. Matching is performed according to spatial and spectral criteria, and calculations are performed under different time matching windows. The coefficient of determination of the band , the calculation results are shown as follows:

[0092] ;

[0093] in, .

[0094] Therefore, it is necessary to further optimize the best matching window for satellite-ground matching. The process of determining the best matching window using the time matching principle in step 2 is as follows: The sum of the determination coefficients of all bands is defined as S , used to filter out the optimal time matching window, the calculation formula is as follows:

[0095] ;

[0096] in, 、 Bands The coefficient of determination of S The time matching window at the maximum value is the best matching window.

[0097] Chlorophyll a As an example, the best matching window obtained based on the above method is obtained With chlorophyll a 77 pairs of synchronous data, With chlorophyll a 1736 pairs of synchronous data.

[0098] Furthermore, given that satellite multispectral data and hyperspectral near-sensing reflectance are different in band position, band width and wavelength range. Figure 2 , in step three, a method for simulating hyperspectral conversion to multispectral data is created, specifically,

[0099] , ;

[0100] Where, is the pre-processed water body hyperspectral near-sensing reflectance spectrum data, For Based on the above example and combined with the simulation method, 46 pairs of satellite sensor spectral response functions can be obtained. and Sync data.

[0101] For the 46 pairs and The synchronous data of the same band were analyzed and found and There are significant differences, such as Figure 3 As shown. Among them, Figure 3 ( a )and( b ) are and In 443~783 nm The reflectance spectrum between .

[0102] Therefore, it is necessary to pre-process the satellite remote sensing surface reflectivity data For further processing, the processing method of this embodiment is as follows: ;

[0103] in, , Indicates the band is 443 nm Satellite remote sensing surface reflectivity data, and so on.

[0104] The new satellite remote sensing surface reflectivity data obtained after processing Compared with the simulated satellite remote sensing surface reflectivity Comparison and analysis revealed that the spectral shapes of the two are basically the same.

[0105] Furthermore, band by band and The quantitative conversion relationship between and The quantitative conversion relationship between them is expressed as follows:

[0106] ;

[0107] in, is the linear regression slope, indicating and The gain between is a constant. Combined Figure 4 , Figure 4 ( a )to( h ) in 443 nm , 490 nm , 560 nm 、665 nm 、705 nm , 743 nm and 783 nm Department and Scatter plot of the quantitative conversion relationship between Figure 4 middle( i ) as an example.

[0108] Correspondingly, the steps to eliminate the error include: establishing a linear regression slope The error analysis principle is based on the error analysis principle. and The error situation between them is analyzed, and targeted optimization is performed according to the error situation;

[0109] The error analysis principle is:

[0110] ;

[0111] When the above-mentioned overestimation or underestimation occurs, the correction accuracy of the satellite sensor facing the water body is further optimized; further examples are given to illustrate that For example, the quantitative conversion relationship of each band is expressed as follows:

[0112] ;

[0113] The results show that the different bands in this embodiment slop are all lower than 1 and are not equal, indicating Sentinel -2 has different degrees of overestimation in different bands, which may be related to Sentinel -2 The official atmospheric correction algorithm was developed for land use and is insufficiently corrected for water bodies.

[0114] In order to improve as much as possible Sentinel -2 The accuracy of the data band after conversion is selected The bands above 0.5 are converted to generate Sentinel -2 Image standardization process and method, which includes 443 bands nm , 490 nm 、665 nm 、705 nm And the ratio of 705 / 665.

[0115] In a further embodiment, the water quality parameter inversion model described in step 5 adopts a machine learning model, and the machine learning model includes at least: a deep neural network model, a random forest model, a support vector machine model, an extreme gradient tree model, a Gaussian regression model and a backward propagation neural network model. Further description, the input features are divided into a training set and a test set in a ratio of 7:3, and are sequentially input into the machine learning model to compare and select chlorophyll. a Inversion model. The spectral band contains at least 443 nm , 490 nm 、665 nm 、705 nm And the ratio of 705 / 665.

[0116] At the same time, the following methods can be used to verify the machine learning model, such as calculating the coefficient of determination of the model, the mean absolute percentage error ( MAPE ) and the root mean square error ( RMSE Then, the remaining 30% The synchronized data of water quality and the fused satellite-ground data set were used as the model validation set and input into the above-built model to calculate the coefficient of determination, mean absolute percentage error ( MAPE ) and the root mean square error ( RMSE ). According to the two mean absolute percentage errors in the training dataset and the validation dataset ( MAPE ) and the root mean square error ( RMSE ) The smallest model was selected for the water quality parameter chlorophyll a The inversion model of the two synchronous data sets of chlorophyll is compared. a The accuracy of the model is verified to see if there is an improvement effect. In this embodiment, the chlorophyll content of the fused satellite-ground dataset is aModel accuracy ratio Chlorophyll data synchronized with water quality a The model accuracy is about 22.5% higher, proving that integrating hyperspectral near-field sensing data can significantly improve Sentinel -2 Remote Sensing Chlorophyll Inversion a The accuracy, such as Figure 5 As shown. Among them, Figure 5 ( a )and( b ) are based on Synchronize with water quality data to build and verify chlorophyll a Inversion model scatter plot, Figure 5 ( c )and( d ) are respectively constructed and verified based on the fusion of satellite and ground datasets. a Scatter plot of the inversion model.

[0117] Example 2

[0118] In order to implement the method of improving satellite remote sensing inversion accuracy by fusing hyperspectral near-field sensing as described in Example 1, this embodiment discloses a system for improving satellite remote sensing inversion accuracy by fusing hyperspectral near-field sensing, including:

[0119] The near-sensing hyperspectral sensor, satellite sensor and water quality detector are configured to obtain a water body hyperspectral near-sensing reflectance spectral dataset, a satellite remote sensing surface reflectance dataset and a water environment measured dataset, respectively.

[0120] The first module is configured to select corresponding data processing methods to preprocess the water body hyperspectral near-sensing reflectance spectral dataset, satellite remote sensing surface reflectance dataset and water environment measured dataset respectively.

[0121] The second module is configured to determine the best matching window and construct a synchronous data set using a time matching principle; the synchronous data set includes: Synchronous data with water quality, Synchronous data with water quality; wherein, is the pre-processed satellite remote sensing surface reflectance data, is the pre-processed water body hyperspectral near-sensing reflectance spectrum data, For the band.

[0122] The third module is configured to create a hyperspectral conversion multispectral data simulation method based on the Corresponding to the synchronous data generation of water quality Synchronous data with water quality and get and The synchronization data, is the simulated satellite remote sensing surface reflectivity.

[0123] The fourth module is set up based on and The pre-processed satellite remote sensing surface reflectance data is synchronized with the data Perform normalization processing to obtain new satellite remote sensing surface reflectance data , the new satellite remote sensing surface reflectivity data Compared with the simulated satellite remote sensing surface reflectivity Conduct comparative analysis and build band by band and The quantitative conversion relationship between them is used to eliminate errors.

[0124] The fifth module is configured to Synchronous data with water quality and Combined with the synchronous data of water quality to generate a fused satellite-ground dataset; The water quality parameter inversion model is constructed and trained with the synchronized data of water quality and the fused satellite-ground dataset as input features and the measured water environment dataset as output features.

[0125] The sixth module is configured to repeat steps three to five, and to update the hyperspectral conversion multispectral data simulation method, the satellite spectral accuracy band improvement method, and the fusion of satellite and ground data sets in real time to continuously optimize the water quality parameter inversion model.

[0126] The terminal display module is configured to display and download the inversion results according to the query instruction until receiving the user's query end instruction.

[0127] Specifically, it is used to display and update the linear method of satellite data and hyperspectral data band conversion, as well as the chlorophyll after fusing hyperspectral near-sensing data. a Inversion model accuracy and image results. Receive user query requirements at different time scales, count and display water quality spatiotemporal change images and time series change trend characteristics; the different time scales include at least four time scales: daily, monthly, seasonal and inter-annual. Receive user query requirements for different regional scales, count and display water quality images and time series changes at different time scales; the different regional scales include at least lake bays and open water areas. Receive user download and upload instructions, construct links according to the instructions to download water quality change graphs and tables at different time scales and regional scales, until the query signal ends.

Claims

1. A method for improving satellite remote sensing inversion accuracy by integrating hyperspectral near-field sensing, characterized in that: The following steps are involved: Step 1: Obtain a water body hyperspectral near-sensing reflectance spectral dataset, a satellite remote sensing surface reflectance dataset, and a water environment measured dataset, and use corresponding data preprocessing methods for preprocessing; Step 2: Determine the best matching window and construct a synchronous data set using the time matching principle; the synchronous data set includes: Synchronous data with water quality, Synchronous data with water quality; wherein, is the preprocessed satellite remote sensing surface reflectance data, is the preprocessed water body hyperspectral near-sensing reflectance spectral data, is a band; the process of determining the best matching window is as follows: The sum of the coefficients of determination of all bands is defined as S , used to filter out the optimal time matching window, the calculation formula is as follows: ； Band The coefficient of determination of S The time matching window at the maximum value is the best matching window; Step 3: Create a hyperspectral conversion multispectral data simulation method based on the Corresponding to the synchronous data generation of water quality Synchronize data with water quality and get and The synchronization data, is the simulated satellite remote sensing surface reflectivity; Step 4: Based on and The preprocessed satellite remote sensing surface reflectance data Normalization processing is performed to obtain new satellite remote sensing surface reflectivity data , the new satellite remote sensing surface reflectivity data Compared with the simulated satellite remote sensing surface reflectivity Conduct comparative analysis and build band by band and Quantitative conversion relationship between them to eliminate errors; Step 5: Synchronous data with water quality and Combined with the synchronous data of water quality to generate a fused satellite-ground data set; With the synchronized data of water quality and the fused satellite-ground dataset as input features and the measured dataset of water environment as output features, the water quality parameter inversion model is constructed and trained; Step 6. Repeat steps 3 to 5, update the hyperspectral conversion multispectral data simulation method, the satellite spectral accuracy band improvement method and the fusion satellite-ground data set in real time, continuously optimize the water quality parameter inversion model, and display and download the inversion results according to the query instruction until the user's end query instruction is received.

2. The method for improving satellite remote sensing inversion accuracy by fusing hyperspectral near-field sensing according to claim 1 is characterized in that: The data preprocessing steps of the water body hyperspectral near-sensing reflectance spectral dataset include: The water body hyperspectral near-sensing reflectance spectral dataset is obtained by continuously collecting data in various weather conditions. , calculate the water body hyperspectral near-sensing reflectance spectrum for each band With 420-830 nm Average value of near-sensing reflectance of inner hyperspectral The ratio of the normalized hyperspectral near-sensing reflectance corresponding to the band is obtained. : ； Refer to the spectral data of synchronous ground object spectrum Perform radiation correction, the calculation formula is as follows: ； In the formula, is the radiation gain value of the land-based hyperspectral proximity sensor, is the radiation deviation value of the land-based hyperspectral proximity sensor.

3. The method for improving satellite remote sensing inversion accuracy by fusing hyperspectral near-field sensing according to claim 1 is characterized in that: The data preprocessing steps of the satellite remote sensing surface reflectivity data set include: Obtain satellite remote sensing surface reflectance dataset , extract the quality band qa 60, get the pixel value of the cloud mask file based on the bit operation principle : Using bitwise operations and quality bands qa 60 Generate opaque cloud mask files and cirrus cloud mask files respectively, using and The () function combines the opaque cloud mask file and the cirrus cloud mask file to generate a cloud mask file, thereby obtaining the pixel value of the mask file ; The following formula is used to obtain the preprocessed satellite remote sensing surface reflectance data: : , is the surface reflectivity from satellite remote sensing.

4. The method for improving satellite remote sensing inversion accuracy by fusing hyperspectral near-field sensing according to claim 1 is characterized in that: The formula of the hyperspectral conversion multispectral data simulation method is as follows: ， ； In the formula, is the preprocessed water body hyperspectral near-sensing reflectance spectral data, For Spectral response function of the satellite sensor at .

5. The method for improving satellite remote sensing inversion accuracy by fusing hyperspectral near-field sensing according to claim 1 is characterized in that: The preprocessed satellite remote sensing surface reflectivity data The calculation formula for normalization is as follows: ； in, , Indicates that the band is 443 nm Satellite remote sensing surface reflectivity data, and so on.

6. The method for improving satellite remote sensing inversion accuracy by fusing hyperspectral near-field sensing according to claim 1 is characterized in that: Said and The quantitative conversion relationship between them is expressed as follows: ； in, is the linear regression slope, indicating and The gain between is a constant.

7. The method for improving satellite remote sensing inversion accuracy by fusing hyperspectral near-field sensing according to claim 6 is characterized in that: The error elimination step comprises: Build a linear regression based on the slope The error analysis principle is based on the error analysis principle. and The error situation between them is optimized according to the error situation; the error analysis principle is: ； When the above-mentioned overestimation or underestimation occurs, the correction accuracy of the satellite sensor towards the water body is further optimized.

8. The method for improving satellite remote sensing inversion accuracy by fusing hyperspectral near-field sensing according to claim 1 is characterized in that: The water quality parameter inversion model adopts a machine learning model, which includes at least: a deep neural network model, a random forest model, a support vector machine model, an extreme gradient tree model, a Gaussian regression model and a backward transmission neural network model.

9. A system for improving satellite remote sensing inversion accuracy by fusing hyperspectral near sensing, used to implement the method for improving satellite remote sensing inversion accuracy by fusing hyperspectral near sensing as described in any one of claims 1 to 8, characterized in that: include: The near-sensing hyperspectral sensor, satellite sensor and water quality detector are configured to obtain a water body hyperspectral near-sensing reflectance spectral dataset, a satellite remote sensing surface reflectance dataset and a water environment measured dataset, respectively; The first module is configured to select corresponding data processing methods to perform data preprocessing on the water body hyperspectral near-sensing reflectance spectral dataset, satellite remote sensing surface reflectance dataset and water environment measured dataset respectively; The second module is configured to determine the best matching window and construct a synchronous data set using the time matching principle; The synchronization data set includes: Synchronous data with water quality, Synchronous data with water quality; wherein, is the preprocessed satellite remote sensing surface reflectance data, is the preprocessed water body hyperspectral near-sensing reflectance spectral data, For the band; The third module is configured to create a hyperspectral conversion multispectral data simulation method based on the Corresponding to the synchronous data generation of water quality Synchronize data with water quality and get and The synchronization data, is the simulated satellite remote sensing surface reflectivity; The fourth module is set up based on and The preprocessed satellite remote sensing surface reflectance data Normalization processing is performed to obtain new satellite remote sensing surface reflectivity data , the new satellite remote sensing surface reflectivity data Compared with the simulated satellite remote sensing surface reflectivity Conduct comparative analysis and build band by band and Quantitative conversion relationship between them to eliminate errors; The fifth module is configured to convert the Synchronous data with water quality and Combined with the synchronous data of water quality to generate a fused satellite-ground data set; With the synchronized data of water quality and the fused satellite-ground dataset as input features and the measured dataset of water environment as output features, the water quality parameter inversion model is constructed and trained; The sixth module is configured to repeat steps 3 to 5, update the hyperspectral conversion multispectral data simulation method, the satellite spectral accuracy band improvement method and the fusion satellite-ground data set in real time, and continuously optimize the water quality parameter inversion model; The terminal display module is configured to display and download the inversion results according to the query instruction until receiving the user's query end instruction.

Images