Chlorophyll a remote sensing inversion method, device, system and storage medium

Abstract

The present invention provides a chlorophyll a remote sensing inversion method, device, system, and storage medium, belonging to the field of pigment processing. The method comprises: S1: obtaining multiple chlorophyll a concentrations from the SeaBASS verification system and multiple satellite sensor measurements from a medium-resolution imaging spectrometer; S2: matching each chlorophyll a concentration with each satellite sensor measurement to obtain a set of satellite in-situ matching pairs; S3: analyzing and screening the set of satellite in-situ matching pairs to obtain a screened matching pair sample set; S4: dividing the screened matching pair sample set into a matching pair training subset and a matching pair test subset according to a preset ratio. The present invention addresses the problems of existing chlorophyll a inversion algorithms, such as insufficient input features and training samples, and the vulnerability of inversion accuracy to the quality of extended features. It also addresses the shortcomings of artificially constructed extensions, noise intolerance, and limited inversion results.

Description

Technical Field

[0001] The present invention mainly relates to the field of pigment processing technology, and specifically relates to a chlorophyll a remote sensing inversion method, device, system and storage medium. Background Art

[0002] Chlorophyll a (Chla) is the primary pigment in phytoplankton. It is crucial for determining phytoplankton biomass, which is used to estimate the trophic status of water bodies. Existing scientific literature examining chlorophyll a concentration inversion reveals that traditional Chla inversion algorithms are unable to distinguish Chla from suspended particulate matter (SPM) and colored dissolved organic matter (CDOM), leading to invariant uncertainties in these water components. Machine learning has also been used to construct Chla inversion algorithms across water body types, with multi-layer perceptrons (MLPs), Gaussian process regression (GPR), support vector regression (SVR), and random forest regression (RFR) demonstrating potential for Chla inversion. However, CNN-based methods for Chla inversion at different trophic levels have yet to be investigated. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a chlorophyll a remote sensing inversion method, device, system and storage medium in response to the deficiencies of the existing technology.

[0004] The present invention solves the above-mentioned technical problem with the following technical solution: A chlorophyll a remote sensing inversion method comprises the following steps:

[0005] S1: Acquire multiple chlorophyll a concentrations from the SeaBASS validation system and multiple satellite sensor measurements from the Moderate Resolution Imaging Spectroradiometer;

[0006] S2: Matching each of the chlorophyll a concentrations and each of the satellite sensor measurement values respectively to obtain satellite in-situ matching pairs of each of the chlorophyll a concentrations, and forming a set of satellite in-situ matching pairs;

[0007] S3: Analyze and screen the satellite in situ matching pair set to obtain a screened matching pair sample set;

[0008] S4: Dividing the screened matching pair sample set into a matching pair training subset and a matching pair test subset according to a preset ratio;

[0009] S5: constructing a one-dimensional convolutional neural network and a support vector machine regression model, and training the one-dimensional convolutional neural network and the support vector machine regression model using the matching pair training subset and the chlorophyll a concentration corresponding to each of the satellite in situ matching pairs in the matching pair training subset to obtain a trained one-dimensional convolutional neural network and a trained support vector machine regression model;

[0010] S6: predicting each of the satellite in-situ matching pairs in the matching pair test subset using the trained one-dimensional convolutional neural network and the trained support vector machine regression model to obtain a predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair test subset;

[0011] S7: analyzing the trained one-dimensional convolutional neural network and the trained support vector machine regression model for a target inversion model according to the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair training subset and the predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair testing subset to obtain a target inversion model;

[0012] S8: Importing the remote sensing satellite reflectivity data to be processed, performing inversion processing on the remote sensing satellite reflectivity data to be processed using the target inversion model, and obtaining a chlorophyll a remote sensing inversion result.

[0013] Another technical solution of the present invention to solve the above technical problem is as follows: a chlorophyll a remote sensing inversion device, comprising:

[0014] A parameter acquisition module to acquire multiple chlorophyll a concentrations from the SeaBASS validation system and multiple satellite sensor measurements from the Moderate Resolution Imaging Spectrometer;

[0015] a data matching module, configured to match each of the chlorophyll a concentrations with each of the satellite sensor measurements, to obtain satellite in-situ matching pairs of each of the chlorophyll a concentrations, and to form a set of satellite in-situ matching pairs;

[0016] A screening and analysis module is used to analyze and screen the satellite in situ matching pair set to obtain a screened matching pair sample set;

[0017] A matching pair division module, configured to divide the screened matching pair sample set into a matching pair training subset and a matching pair test subset according to a preset ratio;

[0018] a model training module for constructing a one-dimensional convolutional neural network and a support vector machine regression model, and training the one-dimensional convolutional neural network and the support vector machine regression model using the matching pair training subset and the chlorophyll a concentration corresponding to each of the satellite in situ matching pairs in the matching pair training subset to obtain a trained one-dimensional convolutional neural network and a trained support vector machine regression model;

[0019] a model testing module, configured to predict each of the satellite in-situ matching pairs in the matching pair test subset using the trained one-dimensional convolutional neural network and the trained support vector machine regression model, to obtain a predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair test subset;

[0020] a target inversion model analysis module, configured to analyze the trained one-dimensional convolutional neural network and the trained support vector machine regression model for a target inversion model based on the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair training subset and the predicted chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair testing subset, to obtain a target inversion model;

[0021] The remote sensing inversion result acquisition module is used to import the remote sensing satellite reflectivity data to be processed, perform inversion processing on the remote sensing satellite reflectivity data to be processed through the target inversion model, and obtain the chlorophyll a remote sensing inversion result.

[0022] Based on the above-mentioned chlorophyll a remote sensing inversion method, the present invention also provides a chlorophyll a remote sensing inversion system.

[0023] Another technical solution of the present invention to solve the above-mentioned technical problem is as follows: a chlorophyll a remote sensing inversion system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor; when the processor executes the computer program, the chlorophyll a remote sensing inversion method as described above is implemented.

[0024] Based on the above-mentioned chlorophyll a remote sensing inversion method, the present invention also provides a computer-readable storage medium.

[0025] Another technical solution of the present invention to solve the above technical problem is as follows: a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the chlorophyll a remote sensing inversion method as described above is implemented.

[0026] The beneficial effects of the present invention are as follows: a satellite in-situ matching pair set is obtained by matching chlorophyll a concentration with a satellite sensor measurement value, a screening matching pair sample set is obtained by analyzing and screening the satellite in-situ matching pair set, the screening matching pair sample set is divided into a matching pair training subset and a matching pair test subset according to a preset ratio, a one-dimensional convolutional neural network and a support vector machine regression model are trained using the matching pair training subset and the chlorophyll a concentration in the matching pair training subset to obtain a trained one-dimensional convolutional neural network and a trained support vector machine regression model, a predicted value is obtained by predicting the satellite in-situ matching pair through the trained one-dimensional convolutional neural network and the trained support vector machine regression model, and a prediction value is obtained according to the chlorophyll a concentration in the matching pair training subset and the chlorophyll a concentration in the matching pair training subset. The target inversion model of the trained one-dimensional convolutional neural network and the trained support vector machine regression model is analyzed based on the a concentration and predicted value to obtain the target inversion model. The chlorophyll a remote sensing inversion result is obtained by inverting the remote sensing satellite reflectance data to be processed through the target inversion model, which fills the deficiency of the CNN-based algorithm in inverting Chla at the nutrient level, avoids the need to combine the OWT-based inversion algorithm to establish a unified chlorophyll a inversion model for different water body types, solves the problem that the existing chlorophyll a inversion algorithm has insufficient input features and training samples, and the inversion accuracy is easily affected by the quality of the extended features. It also solves the shortcomings of artificially constructed extensions, noise intolerance and insufficient inversion results. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A schematic diagram of a process for remote sensing inversion of chlorophyll a provided by an embodiment of the present invention;

[0028] Figure 2 A schematic flow chart of another chlorophyll a remote sensing inversion method provided by an embodiment of the present invention;

[0029] Figure 3 This is a module block diagram of a chlorophyll a remote sensing inversion device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0030] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0031] Figure 1 A schematic flow chart of a chlorophyll a remote sensing inversion method provided in an embodiment of the present invention.

[0032] like Figure 1 As shown, a chlorophyll a remote sensing inversion method includes the following steps:

[0033] S1: Acquire multiple chlorophyll a concentrations from the SeaBASS validation system and multiple satellite sensor measurements from the Moderate Resolution Imaging Spectroradiometer;

[0034] S2: Matching each of the chlorophyll a concentrations and each of the satellite sensor measurement values respectively to obtain satellite in-situ matching pairs of each of the chlorophyll a concentrations, and forming a set of satellite in-situ matching pairs;

[0035] S3: Analyze and screen the satellite in situ matching pair set to obtain a screened matching pair sample set;

[0036] S4: Dividing the screened matching pair sample set into a matching pair training subset and a matching pair test subset according to a preset ratio;

[0037] S5: constructing a one-dimensional convolutional neural network and a support vector machine regression model, and training the one-dimensional convolutional neural network and the support vector machine regression model using the matching pair training subset and the chlorophyll a concentration corresponding to each of the satellite in situ matching pairs in the matching pair training subset to obtain a trained one-dimensional convolutional neural network and a trained support vector machine regression model;

[0038] S6: predicting each of the satellite in-situ matching pairs in the matching pair test subset using the trained one-dimensional convolutional neural network and the trained support vector machine regression model to obtain a predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair test subset;

[0039] S7: analyzing the trained one-dimensional convolutional neural network and the trained support vector machine regression model for a target inversion model according to the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair training subset and the predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair testing subset to obtain a target inversion model;

[0040] S8: Importing the remote sensing satellite reflectivity data to be processed, performing inversion processing on the remote sensing satellite reflectivity data to be processed using the target inversion model, and obtaining a chlorophyll a remote sensing inversion result.

[0041] Preferably, the preset ratio may be 7:3.

[0042] It should be understood that the chlorophyll a concentration may be measured by a chlorophyll measuring device.

[0043] It should be understood that the chlorophyll a concentrations used herein are from the NOMAD V2 dataset of the SeaBASS website validation system provided by NASA's Ocean Biological Processing Group (OBPG).

[0044] It should be understood that the satellite sensor measurements may be obtained from any one of the three sensors: SeaWIFS (1997-2010), Moderate Resolution Imaging Spectroradiometer (MODIS Aqua, 2002-present), and Moderate Resolution Imaging Spectroradiometer (MERIS, 2002-2012).

[0045] It should be understood that the Moderate-resolution Imaging Spectroradiometer (MODIS) is a large-scale space remote sensing instrument developed by NASA to understand global climate changes and the impact of human activities on the climate.

[0046] It will be appreciated that the satellite sensor (ie the satellite sensor measurements) and the in situ measurements (ie the chlorophyll a concentrations) match.

[0047] It will be appreciated that field data (ie the chlorophyll a concentration) is matched with Rrs (ie the satellite sensor measurement).

[0048] Specifically, CHLNET includes a one-dimensional convolutional neural network 1DCNN (i.e., the one-dimensional convolutional neural network) for automatic feature extraction and a Chla concentration fitting support vector machine regression algorithm SVR (i.e., the support vector machine regression model) for global Chla concentration inversion.

[0049] It should be understood that all samples are first divided into training samples (ie, the matching pair training subset) and testing samples (ie, the matching pair testing subset) according to a ratio of 7:3.

[0050] It should be understood that the training (i.e., the matching pair training subset) and test data sets (i.e., the matching pair test subset) are randomly split in a ratio of 7:3, and the total number of training (i.e., the matching pair training subset) and test samples (i.e., the matching pair test subset) are 1505 and 646, respectively.

[0051] In the above embodiment, a satellite in-situ matching pair set is obtained by matching chlorophyll a concentration with a satellite sensor measurement value, and a screened matching pair sample set is obtained by analyzing and screening the satellite in-situ matching pair set. The screened matching pair sample set is divided into a matching pair training subset and a matching pair test subset according to a preset ratio. The matching pair training subset and the chlorophyll a concentration in the matching pair training subset are used to train a one-dimensional convolutional neural network and a support vector machine regression model to obtain a trained one-dimensional convolutional neural network and a trained support vector machine regression model. The satellite in-situ matching pairs are predicted by the trained one-dimensional convolutional neural network and the trained support vector machine regression model to obtain a predicted value. According to the chlorophyll a concentration, the predicted value is obtained. The target inversion model of the trained one-dimensional convolutional neural network and the trained support vector machine regression model is analyzed based on the degree and predicted value to obtain the target inversion model. The chlorophyll a remote sensing inversion result is obtained by inverting the remote sensing satellite reflectance data to be processed through the target inversion model, which fills the deficiency of the CNN-based algorithm in inverting Chla at the nutrient level, avoids the need to combine the OWT-based inversion algorithm to establish a unified chlorophyll a inversion model for different water body types, solves the problem that the input features and training samples of the existing chlorophyll a inversion algorithm are insufficient, and the inversion accuracy is easily affected by the quality of the extended features. It also solves the shortcomings of artificial structure extension, noise intolerance and poor inversion results.

[0052] Optionally, as an embodiment of the present invention, the process of step S3 includes:

[0053] According to a plurality of preset chlorophyll a concentration intervals, all satellite in-situ matching pairs of the chlorophyll a concentration obtained from the SeaBASS website verification system are divided into a plurality of categories; wherein each category includes at least one satellite in-situ matching pair;

[0054] Satellite sensor measurement values that are greater than or equal to a preset judgment value are screened out from the satellite sensor measurement values corresponding to all the categories of satellite in-situ matching pairs, and the satellite in-situ matching pairs corresponding to the screened satellite sensor measurement values are used as screened matching pair samples. All the screened matching pair samples are combined to obtain a screened matching pair sample set.

[0055] Preferably, the preset judgment value may be 0.

[0056] Specifically, the satellite in-situ matching samples (i.e., the satellite in-situ matching pairs) are divided into 12 segments according to the Chla concentration (i.e., the preset chlorophyll a concentration range), and the matching pairs with larger noise in each segment are eliminated according to the trend of the spectral curve. At the same time, the matching pairs containing negative Rrs(λ) (i.e., the satellite in-situ matching pairs) are eliminated. After the above steps, the matching noise points are removed, thereby generating a more consistent spectral reflectance curve within the segmented range.

[0057] It should be understood that the matched Rrs (ie, the satellite sensor measurement value) has a positive or negative sign.

[0058] It should be understood that the determination of noise is as follows: taking all Rrs (ie, the satellite sensor measurement values) in the same segment (ie, the category), calculating the average Rrs change curve, and if there is a difference in a certain Rrs change curve, it is determined to be noise.

[0059] In the above embodiment, the analysis and screening of the satellite in-situ matching pair set obtains a screened matching pair sample set, eliminating matching pairs with large noise and matching pairs with negative Rrs(λ) in each segment, so that the matching noise points can be removed, thereby generating a more consistent spectral reflectance curve within the segment range.

[0060] Optionally, as an embodiment of the present invention, the process of step S5 includes:

[0061] S51: constructing a one-dimensional convolutional neural network, and constructing a support vector machine regression model based on an SVR support vector machine regression algorithm, wherein the one-dimensional convolutional neural network includes a fully connected layer and multiple convolutional layers;

[0062] S52: performing normalization processing on each of the satellite in-situ matching pairs in the matching pair training subset to obtain normalized satellite in-situ matching pairs;

[0063] S53: performing data enhancement on each of the normalized satellite in-situ matching pairs through the fully connected layer to obtain enhanced satellite in-situ matching pairs;

[0064] S54: extracting features from each of the enhanced satellite in-situ matching pairs through the multiple convolutional layers to obtain training matching features;

[0065] S55: performing fitting modeling on the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair training subset and the training matching features, respectively, using the support vector machine regression model to obtain a fitting value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs;

[0066] S56: performing an evaluation index analysis based on the chlorophyll a concentrations corresponding to all the satellite in-situ matching pairs in the matching pair training subset and the fitting values of the chlorophyll a concentrations to obtain a first determination coefficient and a first loss value;

[0067] S57: updating parameters of the one-dimensional convolutional neural network and the support vector machine regression model respectively according to the first loss value;

[0068] S58: After the parameters are updated, the loop returns to execute S52-S57 until the preset number of iterations is reached, and the one-dimensional convolutional neural network and support vector machine regression model after the last parameter update are respectively used as the trained one-dimensional convolutional neural network and the trained support vector machine regression model.

[0069] Preferably, the preset number of iterations may be 100.

[0070] It should be understood that 1DCNN is divided into three stages: stage 1 is the input feature preparation stage, using the fully connected layer to construct new input features; stage 2 is the feature extraction stage; and stage 3 is a linear regression stage.

[0071] It should be understood that the first determination coefficient may be stored.

[0072] It should be understood that the fully connected layer merely expands the number of inputs. Its purpose is to integrate the entire convolutional model into a cohesive whole. If not used, the features of individual repeated samples are identical. For example, if the original Rrs (i.e., the normalized satellite in-situ matching pairs) currently only have 5, it can be repeated 100 times to obtain 500 Rrs features (i.e., the enhanced satellite in-situ matching pairs), which has the same function as a fully connected layer with 100 neurons.

[0073] Specifically, nonlinear regression acts on the third step of the CHLNET model, so the choice of regression algorithm is also crucial. Among regression algorithms, linear regression, LassoCV, RFR, and SVR are widely used. The present invention compares the performance of these algorithms and ultimately determines support vector machine regression as a nonlinear algorithm.

[0074] It should be understood that during the training phase, the five Rrs(λ) (i.e., the satellite in-situ matching pairs in the matching pair training subset) are used as input features (X), normalized, and then fed into the 1DCNN (i.e., the one-dimensional convolutional neural network) for training. The label data (Y) represents the Chla concentration (i.e., the chlorophyll a concentration) in logarithmic space. Once the model converges, the 1DCNN (i.e., the one-dimensional convolutional neural network) has completed training.

[0075] In the above embodiment, the one-dimensional convolutional neural network and the support vector machine regression model are trained using the matching pair training subset and the chlorophyll a concentration to obtain the trained one-dimensional convolutional neural network and the trained support vector machine regression model, which avoids the need to combine the OWT-based inversion algorithms to establish a unified chlorophyll a inversion model for different water body types, and also solves the problems of poor noise tolerance and inaccurate inversion results of traditional algorithms.

[0076] Optionally, as an embodiment of the present invention, the process of step S55 includes:

[0077] Performing logarithmic transformation on the training matching features corresponding to each of the satellite in-situ matching pairs in the matching pair training subset to obtain a logarithmic transformation value of the training matching features;

[0078] The support vector machine regression model is used to fit the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair training subset and the logarithmic transformation value of the training matching features to obtain the fitting value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs.

[0079] It should be understood that the features extracted in stage 2 (ie, the training matching features) are matched with the original Chla concentration (ie, the chlorophyll a concentration), and the training of stage 3 is performed.

[0080] It should be understood that the regression algorithm, whose input (X') is the logarithmically transformed value of the 1DCNN output in the second stage (ie, the logarithmically transformed value of the training matching feature), uses X' and Y to implement the training of the regression algorithm (SVR).

[0081] In the above embodiment, the training matching features are logarithmically transformed to obtain the logarithmic transformation value of the training matching features, and the chlorophyll a concentration and the logarithmic transformation value of the training matching features are fitted by the support vector machine regression model to obtain the fitting value, thereby avoiding the need to combine the OWT-based inversion algorithm to establish a unified chlorophyll a inversion model for different water body types, and also solving the problems of poor noise tolerance and inaccurate inversion results of traditional algorithms.

[0082] Optionally, as an embodiment of the present invention, the process of step S56 includes:

[0083] Based on the first formula, a first determination coefficient is calculated according to the chlorophyll a concentrations corresponding to all the satellite in-situ matching pairs in the matching pair training subset and the fitted values of the chlorophyll a concentrations to obtain the first determination coefficient. The first formula is:

[0084]

[0085] Based on the second formula, a first loss value is calculated according to the chlorophyll a concentrations corresponding to all the satellite in-situ matching pairs in the matching pair training subset and the fitted values of the chlorophyll a concentrations to obtain the first loss value. The second formula is:

[0086]

[0087] Among them, R 2 is the first determination coefficient, MAE is the first loss value, mean is the mean function, M i is the chlorophyll a concentration corresponding to the i-th satellite in-situ matching pair in the matching pair training subset, E i is the fitted value of the chlorophyll a concentration corresponding to the i-th satellite in-situ matching pair in the matching pair training subset, and n is the number of the satellite in-situ matching pairs in the matching pair training subset.

[0088] It should be understood that goodness of fit refers to the degree of fit of the regression line to the observed values. The statistic that measures goodness of fit is the coefficient of determination (also known as the coefficient of determination) R 2 (ie the first determination coefficient). R 2 The maximum value is 1. 2 The closer the value of R is to 1, the better the regression line fits the observed value; on the contrary, 2 The smaller the value of , the worse the regression line fits the observed values.

[0089] It should be understood that R 2 is the square of R.

[0090] It should be understood that the performance evaluation index for Chla inversion using the CHLNET algorithm is established, and the performance evaluation is performed using the original and logarithmically transformed Chla index.

[0091] In the above embodiment, the first determination coefficient and the first loss value are obtained according to the evaluation index analysis of the chlorophyll a concentration and the fitting value, which fills the deficiency of the CNN-based algorithm in inverting Ch la at the nutrient level and avoids the need to combine the OWT-based inversion algorithm to establish a unified chlorophyll a inversion model for different water types.

[0092] Optionally, as an embodiment of the present invention, the process of step S6 includes:

[0093] performing standardization processing on each of the satellite in-situ matching pairs in the matching pair test subset to obtain standardized satellite in-situ matching pairs;

[0094] Inputting each of the standardized satellite in-situ matching pairs into the trained one-dimensional convolutional neural network for prediction processing to obtain an initial predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair test subset;

[0095] Normalizing the initial predicted values of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair test subset to obtain normalized predicted values of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair test subset;

[0096] The normalized predicted value of the chlorophyll a concentration corresponding to each of the satellite in situ matching pairs in the matching pair test subset is input into the trained support vector machine regression model for prediction processing to obtain the predicted value of the chlorophyll a concentration corresponding to each of the satellite in situ matching pairs in the matching pair test subset.

[0097] It should be understood that in the testing phase, the samples (i.e., the satellite in-situ matching pairs in the matching pair test subset) are first standardized, and then the first two stages of 1DCNN (i.e., the trained one-dimensional convolutional neural network) are used to predict the samples (i.e., the satellite in-situ matching pairs in the matching pair test subset). Secondly, the prediction results (i.e., the initial prediction values) are normalized and then put into the regression model (i.e., the trained support vector machine regression model) for prediction to obtain the final inversion result (i.e., the prediction value).

[0098] In the above embodiment, the predicted values are obtained by predicting the satellite in-situ matching pairs through the trained one-dimensional convolutional neural network and the trained support vector machine regression model, which solves the problems of poor noise tolerance and inaccurate inversion results of the traditional algorithm.

[0099] Optionally, as an embodiment of the present invention, the process of step S7 includes:

[0100] Based on the third formula, a second determination coefficient is calculated according to the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair training subset and the predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair testing subset to obtain the second determination coefficient. The third formula is:

[0101]

[0102] Based on the fourth formula, a second loss value is calculated according to the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair training subset and the predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair testing subset to obtain a second loss value, and the second determination coefficient and the second loss value are stored. The fourth formula is:

[0103]

[0104] Among them, R' 2 is the second determination coefficient, MAE' is the second loss value, mean is the mean function, M i is the chlorophyll a concentration corresponding to the i-th satellite in-situ matching pair in the matching pair training subset, E' j is the predicted value of the chlorophyll a concentration corresponding to the jth satellite in-situ matching pair in the matching pair test subset, n is the number of satellite in-situ matching pairs in the matching pair training subset, and m is the number of satellite in-situ matching pairs in the matching pair test subset;

[0105] Parameters of the trained one-dimensional convolutional neural network and the trained support vector machine regression model are updated respectively according to the second loss value. After the parameters are updated, the process returns to S6 and executes in a loop until the preset number of iterations is reached. Then, a second determination coefficient curve graph and a second loss value curve graph are drawn respectively according to all the stored second determination coefficients and all the second loss values.

[0106] Determine whether the second determination coefficient curve graph and the second loss value curve graph meet the conditions, wherein the conditions are that the second determination coefficient curve graph shows a downward or stable trend, and the second loss value curve graph shows an upward or stable trend; if not, return to S6 and execute in a loop; if so, obtain the target inversion model through the trained one-dimensional convolutional neural network and the trained support vector machine regression model after the last parameter update.

[0107] It should be understood that if after 100 iterations, the curve (ie, the second determination coefficient curve) shows a downward trend, but the decline is not obvious, then it can be considered that the model (ie, the target inversion model) has achieved optimal performance.

[0108] It should be understood that goodness of fit refers to the degree of fit of the regression line to the observed values. The statistic that measures goodness of fit is the coefficient of determination (also known as the coefficient of determination) R 2 (ie, the R' 2 ). R 2 The maximum value is 1. 2The closer the value of R is to 1, the better the regression line fits the observed value; on the contrary, 2 The smaller the value of , the worse the regression line fits the observed values.

[0109] It should be understood that R 2 (ie, the R' 2 ) is the square of R'.

[0110] It should be understood that R 2 (ie, the R' 2 ) is no longer rising, and MAE (that is, the second loss value) is no longer falling, which means that the curves of the two indicators are smoother.

[0111] In the above embodiment, the target inversion model is obtained by analyzing the target inversion model of the trained one-dimensional convolutional neural network and the trained support vector machine regression model based on the chlorophyll a concentration and the predicted value, avoiding the need to combine the OWT-based inversion algorithms to establish a unified chlorophyll a inversion model for different water types.

[0112] Alternatively, as another embodiment of the present invention, the present invention uses the original and logarithmically transformed Chla indicators for performance evaluation. These indicators include the following formulas:

[0113]

[0114]

[0115]

[0116]

[0117] When there are multiple evaluation indicators, it is difficult to judge the quality of an algorithm from a single indicator. This paper improves the star map visualization method and comprehensively evaluates the performance of the algorithm. The maximum and minimum values of each metric are set as equations (6) and (7), and the maximum and minimum values of the scalar are set as the maximum and minimum values of each metric, respectively. For those indicators (RMSE, RMLSE, MAE), the smaller the value, the better the performance.

[0118] Maximum=Max(P i )+10%(Max(P i )-Min(P i )) (6)

[0119] Minimum=Min(P i )-10%(Max(P i )-Min(P i )) (7).

[0120] It should be understood that M refers to the measured value (i.e., the chlorophyll a concentration), and E refers to the estimated value (i.e., the fitted value or the predicted value). Since there are multiple measured values, the value i in Mi refers to the number of chlorophyll a concentrations. Here, P is the result of formulas 1-5. Since there are multiple indicators, it is not convenient to evaluate, so this evaluation method is designed.

[0121] Optionally, as another embodiment of the present invention, the steps of feature extraction, regression model construction and performance evaluation of the CHLNET model of the present invention are as follows:

[0122] Step 1: After preprocessing the data as described in step S3, the training and test data sets were randomly split in a ratio of 7:3. The total number of training and test samples was 1505 and 646 respectively. The training data set was trained in a 1DCNN network structure. The number of iterations of the model was set to 100, and the ratio of training to validation data was 7:3 during training. R was recorded during training. 2 If the curve still shows a downward trend after 100 iterations, but the decline is not obvious, then the model can be considered to have achieved the best performance.

[0123] Step 2: To further evaluate the stability of the 1DCNN structure, the training process was repeated 20 times, and the MAE and R of the training and test datasets were recorded each time. 2 index.

[0124] Step 3: Determine the regression algorithm for the CHLNET model. The results of the 1DCNN model were used as a baseline, and the test dataset was used as an evaluation dataset to compare the performance of various regression algorithms with the baseline. Each regression algorithm was run 20 times, and the average of each metric was taken. Experiments showed that SVR and RFR outperformed linear regression and LassoCV in most metrics. Although RFR slightly outperformed SVR in the bias metric, it lagged behind SVR in all other metrics. Furthermore, throughout the experiments, it was found that RFR took 10 times longer to train than SVR. Therefore, SVR was selected as the regression algorithm for CHLNET.

[0125] Step 4: Evaluate the CHLNET model and use the performance indicators of RFR, SVR, and MDN for horizontal comparison. All algorithms use the same training data (n=1505) and test data (n=646). Taking the SeaWiFS Chla product indicators as the baseline, the performance indicators of the four algorithms are normalized on all datasets and test datasets. When comparing and analyzing multiple algorithms, the larger the area covered by the algorithm, the better the performance. The slope and RMSE metrics of CHLNET are very close to those of MDN. On the test dataset, the difference between them in Slope and RMSE is only 0.016 and 1.073. CHLNET shows good performance on the test dataset and is superior to MDN in several indicators such as R. 2 , MWP and MAE) outperforms the other four algorithms.

[0126] Alternatively, as another embodiment of the present invention, the accuracy of CHLNET at different Chla concentrations is measured by the following steps:

[0127] To measure the accuracy of CHLNET at varying Chla concentrations, matched samples were categorized into three trophic levels: oligotrophic (Chla ≤ 0.1 mg / m³), mesotrophic (0.1 Chla ≤ 1 mg / m³), and eutrophic (chlorophyll a > 1 mg / m³). CHLNET's performance varied across trophic levels. In oligotrophic waters, CHLNET's performance metrics (RMSE, RMLSE, and MAE) slightly outperformed those of SeaWiFS products. For mesotrophic waters, CHLNET demonstrated a clear advantage across all metrics. The slope improved from 0.571 to 1.007, and the bias metric became closer to 1, ranging from 1.406 to 1.061. CHLNET also demonstrated a significant performance advantage over SeaWiFS products in eutrophic waters. It can be concluded that CHLNET exhibits high retrieval accuracy across different trophic levels and can avoid the issues associated with combining OWT-based retrievals to establish a consistent Chla retrieval model.

[0128] Optionally, as another embodiment of the present invention, the spatial mapping capability of CHLNET in the present invention is evaluated as follows:

[0129] The mapping capabilities of CHLNET were evaluated by comparing Chla concentrations retrieved using the CHLNET algorithm with SeaWIFS and MODISAqua products in regions with varying trophic levels. The experiments showed that the CHLNET model generally demonstrated superior mapping performance compared to the SeaWIFS Chla product in mesotrophic and eutrophic waters. However, mapping performance at low Chla concentrations requires improvement.

[0130] Optionally, as another embodiment of the present invention, the steps of quantifying the generalization ability of the CHLNET model for cross-sensor data in the present invention are as follows:

[0131] The MERIS and MODIS Aqua satellite images, after band shifting, were matched in situ and divided into training and test datasets in a 7:3 ratio. The CHLNET model trained on the SeaWiFS dataset was applied to the MERIS and MODIS Aqua test datasets. Experiments showed that by adding sensor-specific samples, CHLNET captured features in the MERIS and MODIS Aqua test datasets that were not extracted by the original CHLNET. This demonstrates that the CHLNET model can be used for cross-sensor generalization when target sensor samples are insufficient to build a 1D CNN model.

[0132] Optionally, as another embodiment of the present invention, the steps of evaluating the quality of the CHLNET algorithm in the present invention are as follows:

[0133] Because CHLNET lacks a well-defined functional form, the physical meaning of its features cannot be determined. This makes achieving Chla accuracy beyond the training range challenging when training data is insufficient. CHLNET outperforms the OCx algorithm in cross-sensor applications, providing a new approach for Chla inversion. When sensor sample size is limited, the feature weights learned by CHLNET in the original sensor (e.g., SeaWiFS) can be transferred to a new sensor (e.g., MERIS) through transfer learning, thereby optimizing the original CHLNET with a small number of new sensor samples and establishing high-performance inversion capabilities for the target sensor.

[0134] Alternatively, as another embodiment of the present invention, the present invention provides a new ensemble learning algorithm for marine chlorophyll a estimation based on a one-dimensional convolutional neural network (1DCNN) and a regression algorithm (SVR) - CHLNET, to establish a relationship between Rrs(λ) and Chla, to fill the deficiency of CNN-based algorithms in inverting Chla in cross-trophic waters. Performance evaluation shows that the performance of the present invention is better than the most advanced OCx, SVR, RFR and MDN algorithms. Applications in different trophic waters show that CHLNET avoids the need to combine OWT-based inversion algorithms to establish a unified chlorophyll a inversion model for different water types. CHLNET solves the problems of poor noise tolerance and inaccurate inversion results. At the same time, the application of in-situ matching of MERIS and MOIDS-Aqua satellites shows that CHLNET can significantly improve the performance of CHLNET in cross-sensor chlorophyll a inversion by adding a small number of target sensor matching samples to the training data set.

[0135] Alternatively, as another embodiment of the present invention, Figure 2 As shown, the steps of the present invention are as follows:

[0136] Step (1), satellite sensor and field measurement matching;

[0137] Step (2), matching sample preprocessing;

[0138] Step (3), model development of CHLNET;

[0139] Step (4), Chla inversion workflow of CHLNET;

[0140] Step (5), Chla inversion performance evaluation index;

[0141] Step (6), CHLNET model performs feature extraction, regression model construction and performance evaluation;

[0142] Step (7) measures the performance of nutrient water bodies at different levels and compares the performance of CHLNET with the SeaWIFS product indicators to analyze the accuracy of CHLNET at different Chla concentrations.

[0143] Step (8) CHLNET mapping capability assessment. Rrs(λ) is used as input data for CHLNET to invert global Chla concentrations. The SeaWiFS Chla product data from the same period is used as comparison data to verify the spatial mapping capability of CHLNET.

[0144] Step (9), cross-sensor capability evaluation. The CHLNET model trained on the SeaWiFS dataset is applied to the MERIS and MODIS Aqua test datasets to quantify the cross-sensor generalization capability of the CHLNET model.

[0145] Step (10): Analyze the advantages and disadvantages of the CHLNET algorithm.

[0146] Alternatively, as another embodiment of the present invention, compared with the prior art, the present invention has the following advantages: a high-precision chlorophyll a inversion method is implemented by extracting sample data and training a test model, combining the advantages of a one-dimensional convolutional neural network and a regression algorithm to construct a new integrated learning algorithm, CHLNET. The CHLNET algorithm fills the shortcomings of the CNN-based algorithm in inverting Chla at the nutrient level. By using CHLNET, the need to combine OWT-based inversion algorithms to establish a unified chlorophyll a inversion model for different water body types is avoided. In addition, the CHLNET model solves the problems of noise intolerance and insufficient inversion results when inverting Chla concentration.

[0147] Alternatively, as another embodiment of the present invention, the present invention compares the performance indicators of four Chla inversion algorithms, evaluates the performance under different trophic levels, and analyzes the global mapping capability. At the same time, the generalization ability of the CHLNET model is explored. Although different sensors (SeaWiFS, MERIS, and MODIS Aqua) have differences in atmospheric correction methods, wavelengths, bandwidths, and data processing, the model also reduces the influence of the central band by band shifting, and obtains better inversion results on the MERIS and MODIS Aqua datasets.

[0148] Optionally, as another embodiment of the present invention, the present invention relates to a new algorithm model for chlorophyll a inversion, especially for the shortcomings of the existing chlorophyll a (Chla) inversion algorithm in terms of input features and training samples, and the inversion accuracy is easily affected by the quality of extended features, especially the shortcomings of artificially constructed extensions.

[0149] Optionally, as another embodiment of the present invention, since CHLNET is composed of a convolutional neural network and a machine learning algorithm, its training and testing process is different from that of a standard conventional neural network.

[0150] Optionally, as another embodiment of the present invention, the present invention establishes a relationship between Rrs(λ) and Chla to fill the deficiency of CNN-based algorithms in inverting Chla in cross-trophic waters. Performance evaluation shows that the performance of the present invention is better than the most advanced OCx, SVR, RFR and MDN algorithms. Applications in different trophic waters show that CHLNET avoids the need to combine OWT-based inversion algorithms to establish a unified chlorophyll a inversion model for different water types. CHLNET solves the problems of poor noise tolerance and inaccurate inversion results. At the same time, the application of in-situ matching of MERIS and MOIDS-Aqua satellites shows that CHLNET can significantly improve the performance of CHLNET in cross-sensor chlorophyll a inversion by adding a small number of target sensor matching samples to the training data set.

[0151] Optionally, as another embodiment of the present invention, the advantages of the present invention are as follows:

[0152] 1. The constructed high-precision chlorophyll a (Chla) inversion method provides a new idea for Chla inversion.

[0153] 2. The relationship between Rrs(λ) and Chla was established based on the CNN algorithm to fill the deficiency of inverting Chla at the trophic level based on machine learning algorithms.

[0154] 3. The performance evaluation of the model shows that it outperforms the state-of-the-art OCx, SVR, RFR and MDN algorithms. The application of different nutrient water bodies shows that CHLNET avoids the need to combine OWT-based inversion algorithms to establish a unified chlorophyll a inversion model for different water body types.

[0155] Figure 3 This is a module block diagram of a chlorophyll a remote sensing inversion device provided by an embodiment of the present invention.

[0156] Alternatively, as another embodiment of the present invention, Figure 3 As shown, a chlorophyll a remote sensing inversion device includes:

[0157] A parameter acquisition module to acquire multiple chlorophyll a concentrations from the SeaBASS validation system and multiple satellite sensor measurements from the Moderate Resolution Imaging Spectrometer;

[0158] a data matching module, configured to match each of the chlorophyll a concentrations with each of the satellite sensor measurements, to obtain satellite in-situ matching pairs of each of the chlorophyll a concentrations, and to form a set of satellite in-situ matching pairs;

[0159] A screening and analysis module is used to analyze and screen the satellite in situ matching pair set to obtain a screened matching pair sample set;

[0160] A matching pair division module, configured to divide the screened matching pair sample set into a matching pair training subset and a matching pair test subset according to a preset ratio;

[0161] a model training module for constructing a one-dimensional convolutional neural network and a support vector machine regression model, and training the one-dimensional convolutional neural network and the support vector machine regression model using the matching pair training subset and the chlorophyll a concentration corresponding to each of the satellite in situ matching pairs in the matching pair training subset to obtain a trained one-dimensional convolutional neural network and a trained support vector machine regression model;

[0162] a model testing module, configured to predict each of the satellite in-situ matching pairs in the matching pair test subset using the trained one-dimensional convolutional neural network and the trained support vector machine regression model, to obtain a predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair test subset;

[0163] a target inversion model analysis module, configured to analyze the trained one-dimensional convolutional neural network and the trained support vector machine regression model for a target inversion model based on the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair training subset and the predicted chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair testing subset, to obtain a target inversion model;

[0164] The remote sensing inversion result acquisition module is used to import the remote sensing satellite reflectivity data to be processed, perform inversion processing on the remote sensing satellite reflectivity data to be processed through the target inversion model, and obtain the chlorophyll a remote sensing inversion result.

[0165] Alternatively, another embodiment of the present invention provides a chlorophyll a remote sensing inversion system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the chlorophyll a remote sensing inversion method described above is implemented. The system may be a computer or other system.

[0166] Optionally, another embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the chlorophyll a remote sensing inversion method as described above is implemented.

[0167] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0168] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0169] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is merely a logical functional division. In actual implementation, other division methods may be used, such as combining or integrating multiple units or components into another system, or ignoring or not implementing certain features.

[0170] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected based on actual needs to achieve the objectives of the embodiments of the present invention.

[0171] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0172] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program code.

[0173] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A chlorophyll a remote sensing inversion method, characterized in that: The steps include: S1: Acquire multiple chlorophyll a concentrations from the SeaBASS validation system and multiple satellite sensor measurements from the Moderate Resolution Imaging Spectroradiometer; S2: Matching each of the chlorophyll a concentrations and each of the satellite sensor measurement values respectively to obtain satellite in-situ matching pairs of each of the chlorophyll a concentrations, and forming a set of satellite in-situ matching pairs; S3: Analyze and screen the satellite in situ matching pair set to obtain a screened matching pair sample set; S4: Dividing the screened matching pair sample set into a matching pair training subset and a matching pair test subset according to a preset ratio; S5: constructing a one-dimensional convolutional neural network and a support vector machine regression model, and training the one-dimensional convolutional neural network and the support vector machine regression model using the matching pair training subset and the chlorophyll a concentration corresponding to each of the satellite in situ matching pairs in the matching pair training subset to obtain a trained one-dimensional convolutional neural network and a trained support vector machine regression model; S6: predicting each of the satellite in-situ matching pairs in the matching pair test subset using the trained one-dimensional convolutional neural network and the trained support vector machine regression model to obtain a predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair test subset; S7: analyzing the trained one-dimensional convolutional neural network and the trained support vector machine regression model for a target inversion model according to the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair training subset and the predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair testing subset to obtain a target inversion model; S8: importing the remote sensing satellite reflectivity data to be processed, performing inversion processing on the remote sensing satellite reflectivity data to be processed using the target inversion model, and obtaining a chlorophyll a remote sensing inversion result; The process of step S7 includes: Based on the third formula, a second determination coefficient is calculated according to the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair training subset and the predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair testing subset to obtain the second determination coefficient. The third formula is: Based on the fourth formula, a second loss value is calculated according to the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair training subset and the predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair testing subset to obtain a second loss value, and the second determination coefficient and the second loss value are stored. The fourth formula is: Among them, R' 2 is the second determination coefficient, MAE' is the second loss value, mean is the mean function, M i is the chlorophyll a concentration corresponding to the i-th satellite in-situ matching pair in the matching pair training subset, E' j is the predicted value of the chlorophyll a concentration corresponding to the jth satellite in-situ matching pair in the matching pair test subset, n is the number of satellite in-situ matching pairs in the matching pair training subset, and m is the number of satellite in-situ matching pairs in the matching pair test subset; Parameters of the trained one-dimensional convolutional neural network and the trained support vector machine regression model are updated according to the second loss value, and after the parameters are updated, the process returns to S6 and executes in a loop until a preset number of iterations is reached, and then a second determination coefficient curve graph and a second loss value curve graph are drawn according to all the stored second determination coefficients and all the second loss values, respectively; Determine whether the second determination coefficient curve graph and the second loss value curve graph meet the conditions, wherein the conditions are that the second determination coefficient curve graph shows a downward or stable trend, and the second loss value curve graph shows an upward or stable trend; if not, return to S6 and execute in a loop; if so, obtain the target inversion model through the trained one-dimensional convolutional neural network and the trained support vector machine regression model after the last parameter update.

2. The chlorophyll a remote sensing inversion method according to claim 1, characterized in that: The process of step S3 includes: According to a plurality of preset chlorophyll a concentration intervals, all satellite in-situ matching pairs of the chlorophyll a concentration obtained from the SeaBASS website verification system are divided into a plurality of categories; wherein each category includes at least one satellite in-situ matching pair; Satellite sensor measurement values that are greater than or equal to a preset judgment value are screened out from the satellite sensor measurement values corresponding to all the categories of satellite in-situ matching pairs, and the satellite in-situ matching pairs corresponding to the screened satellite sensor measurement values are used as screened matching pair samples. All the screened matching pair samples are combined to obtain a screened matching pair sample set.

3. The chlorophyll a remote sensing inversion method according to claim 1, characterized in that: The process of step S5 includes: S51: constructing a one-dimensional convolutional neural network, and constructing a support vector machine regression model based on an SVR support vector machine regression algorithm, wherein the one-dimensional convolutional neural network includes a fully connected layer and multiple convolutional layers; S52: performing normalization processing on each of the satellite in-situ matching pairs in the matching pair training subset to obtain normalized satellite in-situ matching pairs; S53: performing data enhancement on each of the normalized satellite in-situ matching pairs through the fully connected layer to obtain enhanced satellite in-situ matching pairs; S54: extracting features from each of the enhanced satellite in-situ matching pairs through the multiple convolutional layers to obtain training matching features; S55: performing fitting modeling on the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair training subset and the training matching features, respectively, using the support vector machine regression model to obtain a fitting value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs; S56: performing an evaluation index analysis based on the chlorophyll a concentrations corresponding to all the satellite in-situ matching pairs in the matching pair training subset and the fitting values of the chlorophyll a concentrations to obtain a first determination coefficient and a first loss value; S57: updating parameters of the one-dimensional convolutional neural network and the support vector machine regression model respectively according to the first loss value; S58: After the parameters are updated, the loop returns to execute S52-S57 until the preset number of iterations is reached, and the one-dimensional convolutional neural network and support vector machine regression model after the last parameter update are respectively used as the trained one-dimensional convolutional neural network and the trained support vector machine regression model.

4. The chlorophyll a remote sensing inversion method according to claim 3, characterized in that: The process of step S55 includes: Performing logarithmic transformation on the training matching features corresponding to each of the satellite in-situ matching pairs in the matching pair training subset to obtain a logarithmic transformation value of the training matching features; The support vector machine regression model is used to fit the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair training subset and the logarithmic transformation value of the training matching features to obtain the fitting value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs.

5. The chlorophyll a remote sensing inversion method according to claim 3, characterized in that: The process of step S56 includes: Based on the first formula, a first determination coefficient is calculated according to the chlorophyll a concentrations corresponding to all the satellite in-situ matching pairs in the matching pair training subset and the fitted values of the chlorophyll a concentrations to obtain the first determination coefficient. The first formula is: Based on the second formula, a first loss value is calculated according to the chlorophyll a concentrations corresponding to all the satellite in-situ matching pairs in the matching pair training subset and the fitted values of the chlorophyll a concentrations to obtain the first loss value. The second formula is: Among them, R 2 is the first determination coefficient, MAE is the first loss value, mean is the mean function, M i is the chlorophyll a concentration corresponding to the i-th satellite in-situ matching pair in the matching pair training subset, E i is the fitted value of the chlorophyll a concentration corresponding to the i-th satellite in-situ matching pair in the matching pair training subset, and n is the number of the satellite in-situ matching pairs in the matching pair training subset.

6. The chlorophyll a remote sensing inversion method according to claim 1, characterized in that: The process of step S6 includes: performing standardization processing on each of the satellite in-situ matching pairs in the matching pair test subset to obtain standardized satellite in-situ matching pairs; Inputting each of the standardized satellite in-situ matching pairs into the trained one-dimensional convolutional neural network for prediction processing to obtain an initial predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair test subset; Normalizing the initial predicted values of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair test subset to obtain normalized predicted values of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair test subset; The normalized predicted value of the chlorophyll a concentration corresponding to each of the satellite in situ matching pairs in the matching pair test subset is input into the trained support vector machine regression model for prediction processing to obtain the predicted value of the chlorophyll a concentration corresponding to each of the satellite in situ matching pairs in the matching pair test subset.

7. A chlorophyll a remote sensing inversion device, characterized in that: include: A parameter acquisition module to acquire multiple chlorophyll a concentrations from the SeaBASS validation system and multiple satellite sensor measurements from the Moderate Resolution Imaging Spectrometer; a data matching module, configured to match each of the chlorophyll a concentrations with each of the satellite sensor measurements, to obtain satellite in-situ matching pairs of each of the chlorophyll a concentrations, and to form a set of satellite in-situ matching pairs; A screening and analysis module is used to analyze and screen the satellite in situ matching pair set to obtain a screened matching pair sample set; A matching pair division module, configured to divide the screened matching pair sample set into a matching pair training subset and a matching pair test subset according to a preset ratio; a model training module for constructing a one-dimensional convolutional neural network and a support vector machine regression model, and training the one-dimensional convolutional neural network and the support vector machine regression model using the matching pair training subset and the chlorophyll a concentration corresponding to each of the satellite in situ matching pairs in the matching pair training subset to obtain a trained one-dimensional convolutional neural network and a trained support vector machine regression model; a model testing module, configured to predict each of the satellite in-situ matching pairs in the matching pair test subset using the trained one-dimensional convolutional neural network and the trained support vector machine regression model, to obtain a predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair test subset; a target inversion model analysis module, configured to analyze the trained one-dimensional convolutional neural network and the trained support vector machine regression model for a target inversion model based on the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair training subset and the predicted chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair testing subset, to obtain a target inversion model; A remote sensing inversion result acquisition module is used to import the remote sensing satellite reflectivity data to be processed, perform inversion processing on the remote sensing satellite reflectivity data to be processed through the target inversion model, and obtain the chlorophyll a remote sensing inversion result; The target inversion model analysis module is specifically used for: Based on the third formula, a second determination coefficient is calculated according to the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair training subset and the predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair testing subset to obtain the second determination coefficient. The third formula is: Based on the fourth formula, a second loss value is calculated according to the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair training subset and the predicted value of the chlorophyll a concentration corresponding to each of the satellite in-situ matching pairs in the matching pair testing subset to obtain a second loss value, and the second determination coefficient and the second loss value are stored. The fourth formula is: Among them, R' 2 is the second determination coefficient, MAE' is the second loss value, mean is the mean function, M i is the chlorophyll a concentration corresponding to the i-th satellite in-situ matching pair in the matching pair training subset, E' j is the predicted value of the chlorophyll a concentration corresponding to the jth satellite in-situ matching pair in the matching pair test subset, n is the number of satellite in-situ matching pairs in the matching pair training subset, and m is the number of satellite in-situ matching pairs in the matching pair test subset; Parameters of the trained one-dimensional convolutional neural network and the trained support vector machine regression model are updated according to the second loss value, and after the parameters are updated, the process returns to S6 and executes in a loop until a preset number of iterations is reached, and then a second determination coefficient curve graph and a second loss value curve graph are drawn according to all the stored second determination coefficients and all the second loss values, respectively; Determine whether the second determination coefficient curve graph and the second loss value curve graph meet the conditions, wherein the conditions are that the second determination coefficient curve graph shows a downward or stable trend, and the second loss value curve graph shows an upward or stable trend; if not, return to S6 and execute in a loop; if so, obtain the target inversion model through the trained one-dimensional convolutional neural network and the trained support vector machine regression model after the last parameter update.

8. A chlorophyll a remote sensing inversion system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the chlorophyll a remote sensing inversion method according to any one of claims 1 to 6 is implemented.

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

Images