Coastal water quality classification inversion method and system based on multi-source remote sensing data

Abstract

The invention discloses a coastal water quality classification inversion method and system based on multi-source remote sensing data, and the method comprises the steps: a data collection and preprocessing step: carrying out the consistency analysis of remote sensing reflectivity of all wavebands of two satellites, and obtaining similar wavebands and corresponding remote sensing reflectivity; a training sample data set construction step: distributing the matched data points according to water quality classification types to obtain a training set and a test set of each water quality category; a water quality classification model training step: selecting input features of the water quality classification model, and training the water quality classification model; and a water quality classification inversion step: obtaining a to-be-inverted satellite remote sensing image of the target sea area, and carrying out water quality classification inversion on the to-be-inverted satellite remote sensing image by using the trained water quality classification model to obtain a water quality level spatial distribution condition of the target sea area. According to the method, the precision and efficiency of coastal water quality classification remote sensing inversion are improved by balancing the training samples of the water quality classification model and selecting and optimizing the input features.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental monitoring and environmental protection, and specifically to a coastal water quality classification inversion method and system based on multi-source remote sensing data. Background Art

[0002] Satellite remote sensing technology has become an important means of investigating the spatial distribution of marine water quality due to its high spatial coverage and quasi-real-time advantages, and is widely used in coastal water quality assessment. Conventional water quality remote sensing monitoring methods mainly use remote sensing reflectivity and field observation data to invert specific water quality parameters, and then classify water quality levels according to water quality classification standards. This method is complex to operate, and due to the complex atmospheric conditions and optical properties of water bodies in the coastal area, the inversion error of water quality parameters is large.

[0003] Machine learning algorithms have obvious advantages in dealing with multivariate and nonlinear problems, and have become an important direction for building remote sensing inversion models for water quality assessment. However, machine learning models usually require a large amount of training sample data. In practical applications, due to the lack of field observation data and satellite remote sensing images, model training sample data is relatively limited, resulting in the need to improve model accuracy and generalization ability. Summary of the invention

[0004] The purpose of the present invention is to address the defects and shortcomings in the prior art and propose a coastal water quality classification inversion method and system based on multi-source remote sensing data. A training set and a test set are obtained based on on-site water quality monitoring data and multi-source satellite remote sensing data. A high-precision and high-efficiency water quality classification model is obtained by balancing the training set of the water quality classification model and selecting and optimizing the input features, thereby improving the accuracy and efficiency of coastal water quality classification remote sensing inversion.

[0005] The technical solution of the present invention is as follows.

[0006] On the one hand, a coastal water quality classification inversion method based on multi-source remote sensing data includes:

[0007] Data collection and preprocessing steps: atmospheric correction is performed on the remote sensing data collected from the two satellites to obtain the remote sensing reflectance of all bands of the two satellites; consistency analysis is performed on the remote sensing reflectance of all bands of the two satellites to obtain similar bands of the two satellites and the corresponding remote sensing reflectance;

[0008] The training sample data set construction step is to extract spatially corresponding matching data points from the remote sensing data of the two satellites based on the latitude and longitude information of the collected on-site water quality monitoring data; the matching data points are allocated according to the water quality classification type to obtain the training set and test set of each water quality category, and when the number of training sets of each water quality category is unbalanced, oversampling is used to obtain a balanced number of training sets;

[0009] The water quality classification model training step is to select the input features of the water quality classification model based on the pollution factors of coastal water quality and the remote sensing reflectance of similar bands; the water quality classification model is trained using the training sets of each water quality category to obtain a trained water quality classification model;

[0010] The water quality classification inversion step obtains the satellite remote sensing images to be inverted of the target sea area. Based on the test set of each water quality category, the trained water quality classification model is used to perform water quality classification inversion on the satellite remote sensing images to be inverted, and the spatial distribution of water quality levels in the target water area is obtained.

[0011] Preferably, the coastal water quality classification inversion method based on multi-source remote sensing data also includes: a water quality classification model evaluation step of evaluating the water quality classification results using a confusion matrix.

[0012] Preferably, the remote sensing reflectance of all bands of the two satellites is analyzed for consistency to obtain similar bands of the two satellites, specifically including:

[0013] Based on the remote sensing reflectivity of each band of the two satellites, the Pearson correlation coefficient, mean absolute percentage error, mean deviation, unbiased root mean square error and average ratio are used to evaluate the band consistency in turn. If the mean absolute percentage error, mean deviation and unbiased root mean square error of the two bands are lower than the set values, and the average ratio and Pearson correlation coefficient remain at a relatively high value close to 1, it means that the remote sensing reflectivity of the two bands is consistent, thereby obtaining all similar bands of the two satellites.

[0014] Preferably, when the number of training sets for each water quality category is unbalanced, oversampling is used to obtain a training set with a balanced number, specifically including:

[0015] When the number of training sets for each water quality category is unbalanced, the SMOTE algorithm is used to obtain a balanced number of training sets, so that the number of training sets for each water quality category is the same.

[0016] Preferably, based on the pollution factors of coastal water quality and the remote sensing reflectance of similar bands, the input features of the water quality classification model are selected, specifically including:

[0017] The collected on-site water quality monitoring data and satellite remote sensing data were analyzed using correlation analysis to obtain the original feature subset;

[0018] The Shapley additive interpretation method SHAP is used to rank the importance of each feature in the original feature subset, and the preset features ranked first are selected as the initial optimized feature subset according to the importance score;

[0019] The particle swarm algorithm is used to calculate and optimize the initial optimized feature subset to obtain the optimal feature subset.

[0020] Preferably, the on-site water quality monitoring data and satellite remote sensing data are analyzed using a correlation analysis method to obtain an original feature subset, specifically including: performing a correlation analysis on the remote sensing reflectance of different bands in the satellite remote sensing data and the pollution factor, and finding a sensitive feature band with a correlation coefficient higher than a preset threshold; obtaining an original feature set based on the remote sensing reflectance of the sensitive feature band; the original feature subset includes single-band features, band ratio features, three-band combination features and spectral index features.

[0021] Preferably, the Shapley additive interpretation method SHAP is used to rank the importance of each feature in the original feature subset, and the preset features ranked first are selected as the initial optimized feature subset according to the importance score, specifically including:

[0022] The Shapley additive interpretation method SHAP is used to calculate the overall importance score of each feature in the original feature subset;

[0023] Sort by the overall importance score of each feature to obtain the overall importance ranking;

[0024] The Shapley additive interpretation method SHAP was used to calculate the classification importance score of each feature in the original feature subset for each type of water quality;

[0025] Sort by the classification importance score of each feature to obtain the classification importance ranking;

[0026] Combine the overall importance ranking and the classification importance ranking to obtain the top several features in the overall importance ranking and the top several features in the classification importance ranking as the initial optimized feature subset.

[0027] Preferably, the particle swarm algorithm is used to perform computational optimization on the initial optimized feature subset to obtain the optimal feature subset, which specifically includes:

[0028] The initial optimized feature subset is substituted into the particle swarm algorithm for training, the optimal parameters of the model are determined by the grid search method, and the adjustable parameter penalty parameter C and kernel function kernel of the support vector machine are set; the particle swarm algorithm randomly initializes the speed and position of the particles, calculates the fitness function value of the particles, and continuously updates the individual extreme value and global extreme value of the particles according to the update formula of the particle swarm algorithm to find the optimal solution, and the features corresponding to the optimal solution are used as the optimal feature subset.

[0029] Preferably, the water quality classification model adopts a random forest machine learning algorithm.

[0030] On the other hand, a coastal water quality classification inversion system based on multi-source remote sensing data includes:

[0031] The data collection and preprocessing module is used to perform atmospheric correction on the remote sensing data collected from the two satellites to obtain the remote sensing reflectivity of all bands of the two satellites; perform consistency analysis on the remote sensing reflectivity of all bands of the two satellites to obtain similar bands of the two satellites and the corresponding remote sensing reflectivity;

[0032] The training sample data set construction module is used to extract spatially corresponding matching data points in the remote sensing data of two satellites based on the latitude and longitude information of the collected on-site water quality monitoring data; the matching data points are allocated according to the water quality classification type to obtain the training set and test set of each water quality category, and when the number of training sets of each water quality category is unbalanced, oversampling is used to obtain a balanced number of training sets;

[0033] The water quality classification model training module is used to select the input features of the water quality classification model based on the pollution factors of coastal water quality and the remote sensing reflectance of similar bands; the water quality classification model is trained using the training sets of each water quality category to obtain a trained water quality classification model;

[0034] The water quality classification inversion module is used to obtain the satellite remote sensing images to be inverted of the target sea area. Based on the test set of each water quality category, the trained water quality classification model is used to perform water quality classification inversion on the satellite remote sensing images to be inverted, and the spatial distribution of water quality levels in the target water area is obtained.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] (1) The present invention increases the number of training samples through consistency analysis of multi-source satellite remote sensing data, and uses optimization to improve the accuracy of the water quality classification model constructed by the small sample training data set, thereby constructing a remote sensing inversion model suitable for offshore water quality classification, further expanding the application of satellite remote sensing in marine water quality monitoring and assessment, and providing technical support for real-time monitoring of offshore water environment quality and marine ecological pollution control;

[0037] (2) The present invention constructs an online processing method of the ACOLITE atmospheric correction algorithm on the GEE big data platform, thereby realizing the online acquisition and preprocessing of two satellite remote sensing images; at the same time, the remote sensing reflectance of similar bands after atmospheric correction of the two satellites is analyzed for consistency, and bands with good consistency on the daily and annual scales are found; these processing technologies provide a reference for realizing water quality monitoring with higher sampling frequency and longer time series and developing a multi-source satellite remote sensing water quality inversion model (water quality classification model);

[0038] (3) The present invention adopts the SMOTE algorithm (Synthetic Minority Oversampling Technique) to solve the problem of unbalanced classification of model training samples, and adopts the SHAP-PSO (Particle Swarm) optimization algorithm to screen the optimal feature subset required by the machine learning algorithm. These optimization methods significantly improve the accuracy of the coastal water quality classification remote sensing model;

[0039] (4) The present invention constructs a coastal water quality classification model based on the random forest algorithm through quantitative evaluation and comparative analysis of the temporal and spatial distribution characteristics of typical bay water body classification levels. The model has high water quality classification accuracy and can provide a new technical means for the application of coastal water quality remote sensing evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 This is a flow chart of a coastal water quality classification inversion method based on multi-source remote sensing data according to an embodiment of the present invention;

[0042] Figure 2 It is a structural flow chart of a coastal water quality classification inversion method based on multi-source remote sensing data according to an embodiment of the present invention;

[0043] Figure 3 are statistical indicators of the remote sensing reflectance of Landsat-8 and Sentinel-2 in similar bands at various annual scales in the embodiment of the present invention; wherein (a) represents the unbiased root mean square error; (b) represents the Pearson correlation coefficient; (c) represents the mean absolute percentage error; (d) represents the mean deviation; and (e) represents the mean ratio;

[0044] Figure 4 Schematic diagram of the SMOTE algorithm principle of an embodiment of the present invention;

[0045] Figure 5 A flow chart of model input feature selection and optimization according to an embodiment of the present invention;

[0046] Figure 6 It is a heat map of the correlation between water quality parameters and single bands of the embodiment of the present invention; wherein (a) represents the correlation analysis between the remote sensing reflectance of the five bands B1-B5 / 8A and active phosphate and inorganic nitrogen; (b) represents the correlation analysis between the remote sensing reflectance of the five bands B1-B5 / 8A and dissolved oxygen;

[0047] Figure 7 : This is a heat map of the correlation between the band ratio feature and the pollution factor of the embodiment of the present invention; wherein (a) indicates that the pollution factor is active phosphate; (b) indicates that the pollution factor is inorganic nitrogen; (c) indicates that the pollution factor is dissolved oxygen;

[0048] Figure 8 The importance score results of each feature calculated by SHAP in the embodiment of the present invention;

[0049] Fig. 9 The top 20 characteristic variables contributing to the prediction scores of different categories of water quality according to the embodiment of the present invention; (a) indicates that the category is Class I water; (b) indicates that the category is Class II water; (c) indicates that the category is Class III water; (d) indicates that the category is Class IV water; (e) indicates that the category is Class IV water;

[0050] Fig.10 Schematic diagram of the optimal feature subset of an embodiment of the present invention;

[0051] Fig.11 The water quality grade distribution map based on the random forest model inversion of the embodiment of the present invention; wherein (a) represents the quality classification map of a coastal area of ​​City A; (b) represents the quality classification map of another coastal area of ​​City A; (c) represents the quality classification map of a coastal area of ​​City B;

[0052] Fig.12 This is a structural block diagram of a coastal water quality classification inversion system based on multi-source remote sensing data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] See also Figure 1 and Figure 2 As shown, this embodiment discloses a coastal water quality classification inversion method based on multi-source remote sensing data, including:

[0055] In the data collection and preprocessing step S101, atmospheric correction is performed on the remote sensing data collected from the two satellites to obtain the remote sensing reflectance of all bands of the two satellites; consistency analysis is performed on the remote sensing reflectance of all bands of the two satellites to obtain similar bands of the two satellites and corresponding remote sensing reflectance.

[0056] Before atmospheric correction is performed on the remote sensing data collected from the two satellites, the field monitoring data are also collected and preprocessed, as follows.

[0057] Obtain on-site monitoring data of coastal waters. Monitoring parameters include water temperature, salinity, suspended solids, dissolved oxygen, pH, active phosphate and inorganic nitrogen. The water quality is classified according to the "Seawater Quality Standard" (GB3097-1997) and the single factor evaluation method. The single factor evaluation method refers to comparing the monitoring concentrations of sensory, toxic and biological evaluation factors in the water body with their respective standard concentrations, and then determining the category of each parameter based on the ratio. Finally, the worst category is selected from the water quality categories of all items as the comprehensive water quality evaluation result. In China, the pollution factors in most sea areas are active phosphate and inorganic nitrogen, and some are dissolved oxygen. The classification standards for these parameters are shown in Table 1.

[0058] Table 1 Classification and partial standards of seawater quality

[0059]

[0060]

[0061] Compared with traditional water quality monitoring methods, remote sensing technology has the advantages of low cost, quasi-synchronization, and wider monitoring range. However, due to the long revisit period of a single satellite and the interference of weather and clouds, fewer remote sensing images can be obtained and used. In order to obtain more training sample data sets, the present invention uses Landsat-8 and Sentinel-2 satellite remote sensing images as data sources. The settings of the main bands of these two satellites are shown in Tables 2 and 3 respectively.

[0062] Table 2 Landsat-8 band parameters

[0063]

[0064] Table 3 Sentinel-2 band parameters

[0065]

[0066]

[0067] The Landsat-8 and Sentinel-2 remote sensing image data used in the present invention are obtained and processed on the Google Earth Engine (Google Earth Engine, GEE) platform. GEE is a cloud-based online remote sensing image computing platform jointly developed by Google, the United States Geological Survey (USGS) and Carnegie Mellon University. It is currently widely used in processing satellite remote sensing image data and other observation data. It includes Sentinel-2 and Landsat-8 and other geospatial data sets. Visitors can access it very well through the Python Script application programming interface, or they can directly and quickly search and download the corresponding satellite image reflectivity product data on its homepage. Compared to traditional image processing tools such as ENVI and ArcGIS, GEE can process huge amounts of remote sensing images in batches more quickly and efficiently. The present invention uses Landsat-8 and Sentinel-2 original images (Landsat-8 satellite Level 1 products and Sentinel-2 satellite 1C products), and these images need to be atmospherically corrected before use.

[0068] The atmospheric correction uses Acolite software. ACOLITE is an atmospheric correction algorithm developed by Dr. Vanhellemont and Dr. Ruddick of the Royal Belgian Institute of Natural Sciences (RBINS) specifically for high turbidity nearshore waters, and is particularly suitable for inland and marine waters. This software mainly performs atmospheric correction processing on remote sensing data obtained by high spatial resolution satellites such as Landsat-5 / 7 / 8 / 9 and Sentienl-2. Here, the ACOLITE dark spectrum fitting method (DSF method for short) is used for atmospheric correction.

[0069] However, the application of ACOLITE algorithm can generally only call local image data. Therefore, the present invention uses Google Colab to build a "bridge" between ACOLITE algorithm and GEE platform through autonomous programming, so as to realize the online processing of remote sensing images by ACOLITE method. Google Colab is a Jupyter Notebook-like environment, which does not require any third-party library to be configured locally, and can be used directly online and run completely in the cloud. First, use Colab to configure the cloud development environment of GEEpython, and according to the input time series and station longitude and latitude, filter out all the original remote sensing image sets containing the station and store them in the folder of the cloud disk, and convert the original image to nc format, and then call the ACOLITE python source code with set parameters for atmospheric correction, and store all the images in nc format after atmospheric correction in Google Cloud Disk; use the processed nc file, extract all the bands Rrs matching the spatial coordinates according to the longitude and latitude coordinates of the station, and finally store all the extracted Rrs and the corresponding station name, coordinate longitude and latitude, imaging time and other information as csv files and save them to the cloud disk folder.

[0070] This method not only realizes the application of ACOLITE atmospheric correction on the GEE platform, greatly simplifies the operational process of remote sensing image preprocessing, but also improves the application efficiency and scope of the ACOLITE algorithm.

[0071] The spatial resolution of Landsat-8 and Sentinel-2 satellite remote sensing data is relatively high (≤30m), which is very suitable for water quality inversion research in small sea areas. In addition, the two satellites use the same geographic coordinate system and their transit times are also very close. Therefore, the present invention compares the consistency of remote sensing reflectance of similar bands (Table 4) of Landsat-8 and Sentinel-2 satellites through correlation analysis, hoping to combine the remote sensing data of the two satellites, so as to construct a remote sensing reflectance data set with higher sampling frequency, larger data volume and more complete time series between them, thereby improving the time resolution of the water quality classification model.

[0072] In order to analyze the consistency of the two satellite remote sensing reflectances, the present invention uses a set of statistical indicators for evaluation, including the Pearson correlation coefficient (r), mean absolute percent difference (MAPD), mean deviation (δ), unbiased root mean square (RMS) error (Δ μ,RMS ) and the average ratio The calculation formulas are as follows. Considering that the variation range of remote sensing reflectance in each band in the study area is relatively large, in order to more accurately describe the relative variation of remote sensing reflectance between the two satellites, this paper uses the logarithm of remote sensing reflectance for calculation, which can better characterize the difference between the remote sensing reflectance of each band of the two satellites.

[0073] Table 4 Comparison of similar bands between Landsat-8 and Sentinel-2

[0074]

[0075] The Pearson correlation coefficient (r) calculation formula is:

[0076]

[0077] The mean absolute percentage error (MAPD) ​​is calculated as:

[0078]

[0079] The mean deviation (δ) is calculated as:

[0080]

[0081] Unbiased RMS Error (Δ μ,RMs ) is calculated as:

[0082]

[0083] Average ratio Calculation formula:

[0084]

[0085] Where N is the total number of matching points of Landsat-8 sensor OLI and Sentinel-2 sensor MSI in the study area during a certain study period, and x oLI Represents the remote sensing reflectance dataset of the satellite Landsat-8 sensor OLI in the specified band, x MsI Represents the remote sensing reflectance dataset of the satellite Sentinel sensor MSI in the specified band.

[0086] If the mean absolute percentage error MAPD, mean deviation δ and unbiased root mean square error Δ μ,RMS Keep it at a relatively low value, average The Pearson correlation coefficient r fluctuates around 1, and remains at a relatively high value close to 1, which indicates that the remote sensing reflectance of the satellites Landsat-8 and Sentinel-2 after AOLITE atmospheric correction is consistent.

[0087] In order to verify the consistency of remote sensing reflectivity, the present invention counted the statistical indicators of remote sensing reflectivity of two satellites in similar bands on each annual scale, such as Figure 3 As shown. Figure 3 a and Figure 3 c It can be seen that the unbiased RMS errors of bands B1 to B4 are always kept within 0.15, and the average percentage errors are basically lower than 20%, while the errors in the near-infrared and far-infrared bands are relatively large, among which the average percentage errors of bands B6 / B11 and B7 / B12 exceed 60%. Figure 3 b shows that the correlation coefficient of band B1-B5 / B8A is always greater than 0.75 during the five years, which indicates that the remote sensing reflectivity of the two satellite bands after atmospheric correction has a strong correlation. Figure 3 In d, the average deviation of bands B1-B5 / B8A remained within the interval of ±0.2 during the five years, while bands B6 / B11 and B7 / B12 had larger deviations compared with the other five bands. Figure 3 e represents the average ratio of the remote sensing reflectance of each band of the two satellites from 2017 to 2021. It can be seen that the visible light band, blue, green, red band and near-infrared band are all stable near the Y=1 function, but the average ratio of the short-wave infrared band is always high.

[0088] A comprehensive comparison of all the above statistical indicators shows that the remote sensing reflectance of Landsat-8 and Sentinel-2 bands B1 to B5 / B8A has good consistency, while the consistency of short-wave infrared bands B6 / B11 and B7 / B12 is poor. Therefore, the remote sensing reflectance of these five bands can be used to further develop water quality classification models.

[0089] In step S102, the training sample data set is constructed, based on the longitude and latitude information of the collected on-site water quality monitoring data, spatially corresponding matching data points are extracted from the remote sensing data of the two satellites; the matching data points are allocated according to the water quality classification type to obtain training sets and test sets for each water quality category, and when the number of training sets for each water quality category is unbalanced, oversampling is used to obtain a balanced number of training sets, as follows.

[0090] S1021, acquisition of training sample set

[0091] Using the latitude and longitude information of the field observation data, the spatially corresponding matching data points of Landsat and Sentinel remote sensing images were extracted. According to the time series sampled from 2015 to 2021 and the latitude and longitude information of 234 national and provincial control stations, a total of 3477 Landsat-8 satellite data points and 5659 Sentinel-2 satellite data points were obtained. The generation time of the satellite data points and the sampling time of the field monitoring were matched according to the time window of ±1 day, and a total of 296 matching samples were obtained.

[0092] S1022, Processing of Class Imbalanced Samples

[0093] The 296 matched data points are distributed according to each type of water quality, and the number of each type of water quality is shown in Table 5. It can be seen from Table 5 that there are only 15 Class III water samples, which is quite different from the number of samples of other types of water quality, that is, there is an imbalance in the training sample categories. Class imbalance refers to the situation where the number of training samples of different categories in the classification task is very different. This imbalance will cause great trouble to the learning process, and it is easy for the model to accurately predict the category with a large number of samples, but cannot make any predictions for the category with a small number.

[0094] The present invention mainly uses the SMOTE (Synthetic Minority Oversampling Technique) algorithm of the "oversampling" method to preprocess the class imbalanced training set. The SMOTE algorithm generates additional sample examples that are not in the original sample by interpolating the minority class samples in the training set. First, a sample x is randomly selected from the minority class, and then the k nearest neighboring values ​​of this sample are found (usually k = 5), one of the neighboring values ​​x1 is randomly selected, and a synthetic new sample is created on the line between the two randomly selected points, such as Figure 4 shown.

[0095] The matching data set is divided into a training set and a test set in a ratio of 5:1, where the test set data samples are randomly selected from the 2021 sample set, and the number of each category is shown in Table 5. Due to the small number of samples in the training set, this paper uses the 5-fold cross-validation method to train and adjust the model until the best model and parameter configuration are determined (compared based on the classification accuracy of the model in the 5-fold cross-validation validation set), and then substitutes it into the test set to evaluate the best model.

[0096] Table 5 Number of water quality types in the data set

[0097]

[0098] The training set that has not been "oversampled" was substituted into the random forest model for 5-fold cross validation without parameter adjustment, and the model accuracy was 48.32%. The training set was then "oversampled" and the above steps were repeated, and the classification accuracy reached 73.24%. Similarly, when substituted into the support vector machine and decision tree models, the accuracy was 52.16% and 43.27%, respectively. The accuracy of the model after "oversampling" was increased to 72.86% and 55.54%. It can be seen that the classification accuracy of the model constructed using the training set preprocessed by the SMOTE algorithm is significantly higher than that of the model trained with the initial training set. Therefore, when the number of classification categories is unbalanced, the SMOTE algorithm can greatly improve the classification accuracy of the machine learning model.

[0099] In the water quality classification model training step S103, based on the pollution factors of coastal water quality and the remote sensing reflectance of similar bands, the input features of the water quality classification model are selected; the water quality classification model is trained using the training sets of each water quality category to obtain a trained water quality classification model.

[0100] For details, see Figure 5 As shown in the figure, the sensitive bands of inorganic nitrogen, active phosphate and dissolved oxygen, the main pollutants affecting coastal water quality, were analyzed by relevant statistical methods, and band feature variables including spectral index, single band, band ratio and three-band combination were selected to construct the original feature set. Then, the Shapley Additive Explanation (SHAP) method was used to sort the feature importance of the original feature set, and then the initial feature subset was extracted by combining the manual screening method. Finally, the particle swarm optimization (PSO) algorithm was used to optimize the initial feature subset. The number of feature variables contained in the optimized optimal feature subset was much smaller than the number of original feature sets. Through such a feature optimization process, irrelevant or redundant features in the original feature data set can be removed, the dimension of the data set can be reduced, and the efficiency and accuracy of the model can be improved.

[0101] Based on the pollution factors of coastal water quality and the remote sensing reflectance of similar bands, the input features of the water quality classification model are selected, including:

[0102] S1031, using correlation analysis method to analyze the collected on-site water quality monitoring data and satellite remote sensing data to obtain the original feature subset;

[0103] S1032, using Shapley additive interpretation method SHAP to rank the importance of each feature in the original feature subset, and selecting the preset features ranked first as the initial optimized feature subset according to the importance score;

[0104] S1033, using a particle swarm algorithm to perform computational optimization on the initial optimized feature subset to obtain an optimal feature subset.

[0105] Specifically, in order to better utilize machine learning methods to classify water quality, it is necessary to first find out various sensitive characteristic bands that may affect pollution factors as the original feature set. The present invention uses the correlation analysis method to correlate the remote sensing reflectance of different bands of Landsat / Sentinel with major pollution factors such as active phosphate, inorganic nitrogen, dissolved oxygen, etc., to find sensitive characteristic bands with high correlation coefficients, and obtain the original feature set based on the remote sensing reflectance of the sensitive characteristic bands. These original feature sets include:

[0106] 1. Spectral index characteristics

[0107] Since active phosphate, inorganic nitrogen, dissolved oxygen and other pollution factors are usually related to the changes of optical active parameters such as chlorophyll a and turbidity, it can be seen from the correlation analysis that they have a strong correlation with each other. Therefore, the spectral index commonly used to invert the chlorophyll a and turbidity concentration of water bodies is selected and added to the original feature set. These spectral indices mainly include the normalized band difference turbidity index (NDTI), the normalized fluorescence height index (NFHI) and the band ratio index, as shown in Table 6.

[0108] Table 6 Spectral index information

[0109]

[0110] Among them, B3, B4 and B5 refer to the remote sensing reflectance of the 3rd to 5th bands of Landsat8 after atmospheric correction. For the remote sensing data of Sentinel-2 satellite, the settings of its B3, B4 and B8A bands correspond to B3, B4 and B5 of Landsat 8, respectively. Therefore, B8A in Table 6 refers to the remote sensing reflectance of the 8A band of Sentinel-2.

[0111] 2. Single-band remote sensing reflectivity

[0112] Correlation analysis was performed between remote sensing reflectance of the five bands B1-B5 / 8A and active phosphate, inorganic nitrogen and dissolved oxygen. Figure 6 As can be seen from the figure, active phosphate shows a certain correlation with B2, B3, and B4; while inorganic nitrogen shows a relatively strong positive correlation with the five bands from B1 to B5 / 8A; on the contrary, dissolved oxygen has a strong negative correlation with B2, B3, and B4. Therefore, B1 to B5 are selected as sensitive spectral single bands to be added to the original feature set.

[0113] 3. Band ratio

[0114] The band ratio method is a very common band combination feature in the field of water environment remote sensing monitoring. The present invention selects the band ratio method to construct the band combination and calculates the correlation coefficient between the band ratio and three non-optical parameters, namely inorganic nitrogen, phosphate and dissolved oxygen. Figure 7 As shown, the horizontal axis represents the reflectance of the band as the numerator in the band ratio combination, and the vertical axis represents the reflectance of the band as the denominator in the band ratio combination. According to the correlation heat map, the correlation heat map is diagonally symmetrical, and the areas with high correlation between the three parameters and the band ratio are basically concentrated in the second to fourth rows. Therefore, the band ratio combinations B1 / B2, B1 / B3, B1 / B4, B2 / B3, B2 / B4 and B3 / B4 are selected to be added to the original feature set.

[0115] 4. Choice of three-band combination

[0116] The machine learning model is a "black box". Users cannot know the internal calculation formulas and principles, so there are many factors that affect its performance. Sometimes the most influential input features may not have special physical meanings. In order to include more band combinations, the present invention permutes and combines the sensitive bands B1 to B5 / 8A of the inactive optical water quality parameters, and obtains a total of 50 three-band combination formulas that are added to the original feature set, as shown in Table 7.

[0117] Table 7 Three-band combination feature set

[0118]

[0119]

[0120] So far, the embodiment of the present invention has obtained a total of 64 elements of the original feature set, including three spectral index features of NDTI, NFHI, and BRI, five single bands from B1 to B5, 6 band ratios and 50 three-band combinations.

[0121] Feature set optimization is a crucial step in building a classification model. A suitable feature subset can achieve higher classification accuracy with fewer feature variables, thereby improving model efficiency. The 64 original feature set elements that have been obtained may contain a lot of redundancy between each other. Using them directly for training will reduce model efficiency, and the classification accuracy may not be optimal. Therefore, feature selection and optimization are very necessary.

[0122] Specifically, S1032 first uses the Shapley Additive Explanation algorithm (SHAP) to rank the importance of the 64 features in the original feature set. SHAP is a method to solve the model interpretability. It is based on the concept of Shapley value proposed in cooperative game theory and constructs an additive interpretation model, that is, all features in the model are regarded as "contributors" and the internal operations of the model are interpreted by calculating the marginal contribution of each feature to the model output.

[0123] The importance scores of each feature calculated using SHAP are as follows: Figure 8 As shown. Obviously, the single band B3 has the highest feature importance score, followed by B2. From the global "contribution", it can be seen that the single band has the highest contribution score, and the spectral index NDTI and NFHI are ranked later. This may be because although the non-optical activity parameters show a certain correlation with chlorophyll a and turbidity, it is not obvious from the overall performance of the model when extended to qualitative issues such as water quality grade. The contribution values ​​of the band ratio and the three-band combination basically fluctuate at the median level.

[0124] Classification problems not only focus on the overall accuracy of the model, but also need to pay attention to the impact of each feature variable on different types of water quality. The SHAP algorithm can not only rank the contribution of features from a global perspective, but also explain the impact of features locally. Fig. 9 As shown in the figure, the top 20 feature variables ranked by importance for the five types of water quality prediction are displayed. Each point on the figure is the SHAP value of a feature for a data. The vertical axis is the feature variable, and the horizontal axis represents the SHAP value of each feature, which can also be considered as the probability of the model judging the water quality as this type. The closer the color of the point is to red, the higher the value of the feature variable is, and the closer it is to blue, the lower the value is. Fig. 9It can be seen that the most influential features for Class I water are the three-band combination (B1-B4) / B1 and (B2-B4) / B2. When the values ​​of these two features are larger, the SHAP value is larger, and the probability of being judged as Class I water is higher. Similarly, the most influential features for Class II water are single bands B2, B3, and B4. When their values ​​are larger, the probability of being judged as Class II water is higher; for Class III water, B3 also has the greatest influence, but in contrast to Class II water, when the value of B3 is larger, the probability of being judged as Class III water is lower; and the larger the values ​​of B3, the three-band combination (B1-B4) / B1, (B2-B4) / B2, and the band ratio B2 / B4, the lower the probability of being judged as Class IV water. The feature importance of the three-band ratio is also important for Class IV water. For the spectral index, NFHI and BRI were not among the top 20 in importance of water quality characteristics of each grade, while NDTI had a certain influence on the judgment of Class I and Class II water. The larger the NDTI, the smaller the probability of judging it as Class I water, and the greater the probability of judging it as Class II water.

[0125] After analysis, the embodiment of the present invention selects the top 20 features in terms of overall importance score and the top five features in terms of importance to each type of water quality characteristics, as well as the spectral variable NDTI to form an initial optimized feature subset, including B1-B5, B1 / B4, B2 / B4, (B1-B4) / B4, etc., a total of 26 feature variables as the initial optimized feature subset.

[0126] The initial optimized feature subset basically covers the features with the largest contribution and tries to ensure the integrity of the feature types. The main purpose is to delete most of the features with low scores and remove redundant variables, so as to prepare for the next step of particle swarm optimization (PSO) optimization.

[0127] Specifically, the S1033 uses a particle swarm optimization algorithm to screen the initial optimized feature subset to obtain the optimal feature subset. In the feature selection process, the particle swarm optimization algorithm repeatedly iterates the sample set, and each sample data finds and tracks the "optimal value" through iterations, and continuously updates itself, and finally finds the optimal solution. Due to its simple principle and fewer parameters, the particle swarm optimization algorithm is widely used in machine learning and neural network training. However, for feature sets with high feature dimensions, the training speed of the particle swarm optimization algorithm will be very slow. Therefore, the SHAP algorithm is first used to obtain an initial optimized feature subset containing 26 feature variables by manual selection, and then the particle swarm optimization algorithm is substituted to screen the feature subset to obtain the final optimized subset. The SHAP algorithm can score the importance of features involved in the classification and inversion of water quality grades, but the feature variables with high importance scores are not necessarily the optimal features, because there may be strong correlations between features, and the SHAP algorithm can only sort the importance of features, and cannot reflect the correlation problem between features. The particle swarm optimization algorithm takes into account the influencing factors such as the cooperation mechanism, mutual relationship and environmental factors between feature variables, so that the most suitable feature subset can be screened more comprehensively.

[0128] Specifically, the embodiment of the present invention combines the particle swarm optimization algorithm with the support vector machine (SVM) algorithm, takes the classification accuracy of the SVM algorithm as the fitness function, and searches for the classification feature set with the highest classification accuracy. Support vector machine is a classification model algorithm, which is very suitable for small data samples, can solve high-dimensional and nonlinear classification problems, and has strong learning ability and high generalization rate. It is one of the most popular machine learning models today. Support vector machines can generally be divided into three categories: linearly separable support vector machines, linear support vector machines, and nonlinear support vector machines. These three models are used to solve three types of training data, namely, linearly separable training data, approximately linearly separable training data, and linearly inseparable training data. This article trains the third type of nonlinear support vector machine. When the data is linearly inseparable, the nonlinear support vector machine is learned by kernel techniques and soft interval maximization. This article mainly optimizes the adjustable parameters of the support vector machine, such as the penalty parameter C, the kernel function kernel, and the kernel function coefficient gamma. The SVM algorithm has the characteristics of small sample learning and high learning efficiency. It is a research hotspot in the field of machine learning in recent years, and it can be effectively applied to feature selection.

[0129] The initial optimized feature subset of the training set is substituted into the PSO algorithm for training, and the optimal parameters of the model are determined by the grid search method. The penalty term C of SVM is set to 239, and the kernel function type is selected as RBF function; 1c of the PSO algorithm is set to 0.5, 2c is set to 0.5, w is set to 0.8, the population size is set to 30, the number of iterations is set to 200, the speed and position of the particle are randomly initialized, and the fitness function value of the particle is calculated. According to the update formula of the PSO algorithm, the individual extreme value and global extreme value of the particle are continuously updated to find the optimal solution.

[0130] After PSO algorithm optimization, it was found that the model can obtain the optimal solution after selecting 11 features, see Fig.10 As shown, 11 features are selected here to form the optimal feature subset, as shown in Table 8:

[0131] Table 8 Optimal feature subsets

[0132]

[0133] The present invention reduces the number of original features by 82.8% through the SHAP-PSO optimization algorithm. The results show that the more eigenvalues ​​there are, the higher the model accuracy is not necessarily the case. Feature screening can help integrate useful information in high-dimensional data and improve the accuracy of model classification.

[0134] Specifically, the effectiveness of the SHAP-PSO algorithm is evaluated. To verify the effectiveness of the SHAP-PSO algorithm for feature selection, the embodiment of the present invention uses parameters such as precision and recall to compare the classification accuracy of the initial optimized feature subset and the optimized best subset.

[0135] The present invention uses a confusion matrix to compare the classification accuracy of the initial optimized feature subset and the optimized best subset. The results show that the use of the optimized feature subset can greatly improve the performance of the classification model, and the SHAP-PSO optimization algorithm is effective.

[0136] According to different classification categories, the classification problem can be divided into binary classification and multi-classification. In the present invention, the water quality level is divided into Class I, Class II, Class III, Class IV and Class IV, a total of 5 levels, which is a multi-classification problem.

[0137] A classifier is a classification model or decision function learned based on an existing training data set. The process of a classifier predicting the output of a new input variable is called classification. The operation of a classification algorithm mainly goes through two processes: learning and classification. In the learning process, a known training data set is input, and the classifier is continuously trained to determine the appropriate parameter combination and finally determine the best model. In the classification process, the learned classifier is used to classify the new data x. n+1 Perform classification and obtain the classification label y of the predicted output n+1Therefore, the key to solving classification problems in different scenarios is how to select the most suitable one among many classifiers through the learning process, and how to achieve the best model performance in the classification process.

[0138] In the scenario of water quality grade classification, machine learning classification is to establish the relationship between the remote sensing reflectivity of the remote sensing image and the measured water quality grade. Many studies have given a number of effective machine learning algorithms for water quality prediction, including multivariate linear regression, support vector machine, decision tree, naive Bayes and random forest, and all point out that these machine learning models are very suitable for solving water quality prediction and classification problems. Although the neural network algorithm has achieved good results in this type of water quality prediction research, it requires a large number of samples for training, otherwise it is easy to overfit. Considering the lack of on-site monitoring data and the mismatch between the sampling time and the satellite imaging time, the training set data obtained is relatively small. Therefore, the present invention finally selects the traditional classification model random forest algorithm suitable for small samples to perform qualitative remote sensing inversion on the coastal water quality classification level.

[0139] Random forest is a (parallel) ensemble algorithm composed of multiple decision trees. By combining the classification rules and voting results of multiple weak classifiers, the final model has high generalization ability and stability, which is also the origin of the word "forest". The core of the random forest model lies in "random" and "forest", which is the main reason for its good performance. "Random" is mainly reflected in the fact that the random forest first randomly selects n features from the feature set of each node of each decision tree, and then selects an optimal subset from these n features for decision making. This random selection method will form the differences between various decision tree models and obtain different classification rules. There are many adjustable parameters in the random forest. This article mainly tunes the most important parameters, namely the number of decision trees n_estimators (the more decision trees, the better the model stability and generalization ability, but the model will also run slower), tree depth max_depth (if the tree depth is set too large, it will cause over-learning and overfitting problems), the minimum number of samples required for internal node re-division min_samples_split and the minimum number of samples for leaf nodes min_samples_leaf.

[0140] The water quality classification inversion step S104 obtains the satellite remote sensing image to be inverted of the target sea area, and based on the test set of each water quality category, uses the trained water quality classification model to perform water quality classification inversion on the satellite remote sensing image to be inverted, so as to obtain the spatial distribution of water quality grades in the target water area.

[0141] At the same time, by selecting remote sensing data of typical dates and typical sea areas, the water quality classification can be calculated using the random forest model to obtain the spatial distribution of water quality levels in the target waters, and the rationality of the spatiotemporal distribution can be judged based on the water quality distribution characteristics. Fig.11 The inverted water quality classification maps for the coastal areas of City A (July 7 and October 5) and City B (October 4) are shown.

[0142] It can be seen that the inversion map of the random forest can relatively intuitively show the differences in the distribution of coastal water quality levels in time and space. From the perspective of spatial distribution, the coastal water quality is poor, and the closer to the offshore water, the better the water quality. In addition, the overall coastal water quality of City A is better than that of City B. This is consistent with the fact that coastal seawater is greatly affected by factors such as human activities and sewage discharge, and the more economically developed the region, the worse the coastal water quality. From a temporal perspective, the water quality in the dry season (October) is worse than that in the normal water season (July), which is consistent with the results of the on-site water quality classification report. It can be seen that the water quality classification map inverted by the random forest model can reasonably reflect the temporal and spatial variation gradient of coastal water quality classification, indicating that the model is effective and reliable.

[0143] In the water quality classification model evaluation step S105, a confusion matrix is ​​used to evaluate the water quality classification results, as follows.

[0144] S1051, Performance evaluation methods for classification models

[0145] In order to evaluate the accuracy and reliability of the water quality classification model based on machine learning, the confusion matrix is ​​used to evaluate its accuracy. The establishment of the confusion matrix is ​​shown in Table 9. The confusion matrix is ​​an n×n matrix (n represents the number of categories), each row of which represents an actual category and each column represents a predicted category. Each element a in the matrix ij It represents the number of samples that are actually of the i-th class but predicted to be of the j-th class.

[0146] Table 9 Confusion Matrix

[0147] True value / predicted value Prediction Category 1 Prediction Category 2 … Prediction category n Real Category 1 <![CDATA[a 11 ]]> <![CDATA[a 12 ]]> … <![CDATA[a 1n ]]> Real Category 2 <![CDATA[a 21 ]]> <![CDATA[a 22 ]]> … <![CDATA[a 2n <!-- 14 -->]]> … … … … … True category n <![CDATA[a n1 ]]> <![CDATA[a n2 ]]> … <![CDATA[a nn ]]>

[0148] The embodiment of the present invention selects accuracy, recall, F1 value and Kappa coefficient as classification accuracy evaluation indicators. The specific calculation formula is as follows:

[0149] Accuracy refers to the ratio of correctly classified samples to all samples:

[0150]

[0151] Recall refers to the proportion of samples that actually belong to a certain class that are correctly predicted to be of that class:

[0152]

[0153] Precision indicates the proportion of samples predicted to be of a certain class that actually belong to that class:

[0154]

[0155] The F1 value is the harmonic mean of precision and recall, which can be used to evaluate whether a model is suitable. The larger the F1 value, the more suitable the model is:

[0156]

[0157] Among them, P and R represent precision and recall respectively.

[0158] The Kappa coefficient is an indicator of consistency between data in statistics. In classification problems, the Kappa coefficient indicates whether the model prediction results are consistent with the actual classification results. When the Kappa coefficient is within the interval [0.61,1], it means that the model test results are highly consistent with the actual values; when the Kappa coefficient is in the interval [0.41,0.6], it indicates that the consistency is moderate; when the Kappa coefficient is less than 0.4, it means that the prediction results are inconsistent with the actual situation and the model performance is poor. The Kappa coefficient is expressed as:

[0159]

[0160] S1052, Performance evaluation of classification models.

[0161] Table 10 shows the classification results of the random forest on the test data set. The random forest model has the best evaluation indicators, with an accuracy of 78.63% and a kappa value of 0.78, indicating good classification performance.

[0162] Table 10 Test set classification accuracy evaluation indicators

[0163]

[0164] See also Fig.12 As shown, this embodiment also discloses a coastal water quality classification inversion system based on multi-source remote sensing data, including:

[0165] The data collection and preprocessing module 1201 is used to perform atmospheric correction on the remote sensing data collected from the two satellites to obtain the remote sensing reflectance of all bands of the two satellites; perform consistency analysis on the remote sensing reflectance of all bands of the two satellites to obtain similar bands of the two satellites and corresponding remote sensing reflectance;

[0166] The training sample data set construction module 1202 is used to extract spatially corresponding matching data points in the remote sensing data of two satellites based on the latitude and longitude information of the collected on-site water quality monitoring data; the matching data points are allocated according to the water quality classification type to obtain the training set and test set of each water quality category, and when the number of training sets of each water quality category is unbalanced, oversampling is used to obtain a balanced number of training sets;

[0167] The water quality classification model training module 1203 is used to select input features of the water quality classification model based on the pollution factors of coastal water quality and the remote sensing reflectance of similar bands; train the water quality classification model using the training sets of each water quality category to obtain a trained water quality classification model;

[0168] The water quality classification inversion module 1204 is used to obtain the satellite remote sensing images to be inverted of the target sea area, and based on the test set of each water quality category, use the trained water quality classification model to perform water quality classification inversion on the satellite remote sensing images to be inverted, so as to obtain the spatial distribution of water quality grades in the target water area;

[0169] The water quality classification model evaluation module 1205 is used to evaluate the water quality classification results using a confusion matrix.

[0170] Specific implementation of a coastal water quality classification inversion system based on multi-source remote sensing data The same coastal water quality classification inversion method based on multi-source remote sensing data is not repeated in this embodiment.

[0171] The principle and operation mode of the present invention are described through the above-mentioned specific implementation cases. These cases are intended to provide readers with a clear understanding framework to grasp the core ideas and operation points of the present invention. However, it should be clear that these cases are not limitations on the scope of application of the present invention. For professionals in this field, various forms of improvements and innovations based on the core ideas of the present invention should be included in the protection scope of the present invention.

Claims

1. A coastal water quality classification inversion method based on multi-source remote sensing data, characterized in that: include: Data collection and preprocessing steps: atmospheric correction is performed on the remote sensing data collected from the two satellites to obtain the remote sensing reflectivity of all bands of the two satellites; The consistency analysis of the remote sensing reflectance of all bands of the two satellites is carried out to obtain the similar bands of the two satellites and the corresponding remote sensing reflectance; The training sample data set construction step is to extract spatially corresponding matching data points from the remote sensing data of the two satellites based on the latitude and longitude information of the collected on-site water quality monitoring data; the matching data points are allocated according to the water quality classification type to obtain the training set and test set of each water quality category, and when the number of training sets of each water quality category is unbalanced, oversampling is used to obtain a balanced number of training sets; The water quality classification model training step selects the input features of the water quality classification model based on the pollution factors of coastal water quality and the remote sensing reflectance of similar bands; Use the training sets of each water quality category to train the water quality classification model to obtain a trained water quality classification model; The water quality classification inversion step obtains the satellite remote sensing images to be inverted of the target sea area. Based on the test set of each water quality category, the trained water quality classification model is used to perform water quality classification inversion on the satellite remote sensing images to be inverted, and the spatial distribution of water quality levels in the target water area is obtained.

2. The coastal water quality classification inversion method based on multi-source remote sensing data according to claim 1 is characterized in that: Also includes: Water quality classification model evaluation steps using confusion matrix to evaluate water quality classification results.

3. The coastal water quality classification inversion method based on multi-source remote sensing data according to claim 1 is characterized in that: The consistency analysis of the remote sensing reflectivity of all bands of the two satellites was performed to obtain the similar bands of the two satellites, including: Based on the remote sensing reflectivity of each band of the two satellites, the Pearson correlation coefficient, mean absolute percentage error, mean deviation, unbiased root mean square error and average ratio are used to evaluate the band consistency in turn. If the mean absolute percentage error, mean deviation and unbiased root mean square error of the two bands are lower than the set values, and the average ratio and Pearson correlation coefficient remain at a relatively high value close to 1, it means that the remote sensing reflectivity of the two bands is consistent, thereby obtaining all similar bands of the two satellites.

4. The coastal water quality classification inversion method based on multi-source remote sensing data according to claim 1 is characterized in that: When the number of training sets for each water quality category is unbalanced, oversampling is used to obtain a balanced number of training sets, including: When the number of training sets for each water quality category is unbalanced, the SMOTE algorithm is used to obtain a balanced number of training sets, so that the number of training sets for each water quality category is the same.

5. The coastal water quality classification inversion method based on multi-source remote sensing data according to claim 1 is characterized in that: Based on the pollution factors of coastal water quality and the remote sensing reflectance of similar bands, the input features of the water quality classification model are selected, including: The collected on-site water quality monitoring data and satellite remote sensing data were analyzed using correlation analysis to obtain the original feature subset; The Shapley additive interpretation method SHAP is used to rank the importance of each feature in the original feature subset, and the preset features ranked first are selected as the initial optimized feature subset according to the importance score; The particle swarm algorithm is used to calculate and optimize the initial optimized feature subset to obtain the optimal feature subset.

6. The coastal water quality classification inversion method based on multi-source remote sensing data according to claim 5 is characterized in that: The field water quality monitoring data and satellite remote sensing data are analyzed by correlation analysis to obtain the original feature subset, specifically including: performing correlation analysis on the remote sensing reflectance of different bands in the satellite remote sensing data and the pollution factor, and finding the sensitive feature bands with correlation coefficients higher than the preset threshold; obtaining the original feature set based on the remote sensing reflectance of the sensitive feature bands; the original feature subset includes single-band features, band ratio features, three-band combination features and spectral index features.

7. The coastal water quality classification inversion method based on multi-source remote sensing data according to claim 5 is characterized in that: The Shapley additive interpretation method SHAP is used to rank the importance of each feature in the original feature subset, and the preset features ranked first are selected as the initial optimized feature subset according to the importance score, including: The Shapley additive interpretation method SHAP is used to calculate the overall importance score of each feature in the original feature subset; Sort by the overall importance score of each feature to obtain the overall importance ranking; The Shapley additive interpretation method SHAP was used to calculate the classification importance score of each feature in the original feature subset for each type of water quality; Sort by the classification importance score of each feature to obtain the classification importance ranking; Combine the overall importance ranking and the classification importance ranking to obtain the top several features in the overall importance ranking and the top several features in the classification importance ranking as the initial optimized feature subset.

8. The coastal water quality classification inversion method based on multi-source remote sensing data according to claim 5 is characterized in that: The particle swarm algorithm is used to calculate and optimize the initial optimized feature subset to obtain the optimal feature subset, which includes: The initial optimized feature subset is substituted into the particle swarm algorithm for training, the optimal parameters of the model are determined by the grid search method, and the adjustable parameter penalty parameter C and kernel function kernel of the support vector machine are set; the particle swarm algorithm randomly initializes the speed and position of the particles, calculates the fitness function value of the particles, and continuously updates the individual extreme value and global extreme value of the particles according to the update formula of the particle swarm algorithm to find the optimal solution, and the features corresponding to the optimal solution are used as the optimal feature subset.

9. The coastal water quality classification inversion method based on multi-source remote sensing data according to claim 1 is characterized in that: The water quality classification model adopts a random forest machine learning algorithm.

10. A coastal water quality classification inversion system based on multi-source remote sensing data, characterized in that: include: The data collection and preprocessing module is used to perform atmospheric correction on the remote sensing data collected from the two satellites and obtain the remote sensing reflectivity of all bands of the two satellites; The consistency analysis of the remote sensing reflectance of all bands of the two satellites is carried out to obtain the similar bands of the two satellites and the corresponding remote sensing reflectance; The training sample data set construction module is used to extract spatially corresponding matching data points in the remote sensing data of two satellites based on the latitude and longitude information of the collected on-site water quality monitoring data; the matching data points are allocated according to the water quality classification type to obtain the training set and test set of each water quality category, and when the number of training sets of each water quality category is unbalanced, oversampling is used to obtain a balanced number of training sets; The water quality classification model training module is used to select the input features of the water quality classification model based on the pollution factors of coastal water quality and the remote sensing reflectance of similar bands; Use the training sets of each water quality category to train the water quality classification model to obtain a trained water quality classification model; The water quality classification inversion module is used to obtain the satellite remote sensing images to be inverted of the target sea area. Based on the test set of each water quality category, the trained water quality classification model is used to perform water quality classification inversion on the satellite remote sensing images to be inverted, and the spatial distribution of water quality levels in the target water area is obtained.

Images