Optical complex water body suspended silt monitoring method based on human-computer cooperation

Abstract

The application provides a kind of optical complex water body suspended silt monitoring method based on man-machine cooperation, belongs to the technical field of remote sensing water environment monitoring. Contain: collect the multispectral remote sensing image of target area and carry out pretreatment after space-time fusion, obtain high space-time resolution spectral data set;Spectral data set is manually labeled, constructs suspended silt concentration classification data set, and a few class samples are expanded, and balanced training set is generated;The suspended silt concentration inversion model is trained using the balanced training set, the initial suspended silt concentration prediction value is obtained, and the spectral feature weight and classification boundary of the suspended silt concentration inversion model are dynamically optimized based on expert knowledge;Based on the hydrodynamic law and spatial consistency constraint, the initial suspended silt concentration prediction value is locally corrected and diffused to the global range, and the global suspended silt concentration prediction value is obtained.The method combines remote sensing image analysis, machine learning and expert knowledge, and realizes high-precision inversion of SSC.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of remote sensing of water environment, and particularly relates to a method for monitoring suspended sediment in optically complex water bodies based on human-computer cooperation. BACKGROUND

[0002] Global coastal suspended sediment concentration (SSC) plays a crucial role in regulating erosion, deposition and geomorphic evolution processes. However, due to limited observation means, current large-scale, high-resolution SSC monitoring data is still scarce, which greatly restricts the study of coastal evolution mechanisms, especially the quantitative analysis based on empirical models.

[0003] Existing suspended sediment monitoring methods mainly include field sampling, acoustic remote sensing and optical remote sensing. The field sampling method has high precision, but is limited by high cost and limited spatial coverage, making it difficult to meet the monitoring needs of a large range and long time series. Acoustic remote sensing technology (such as Doppler current profiler and multi-beam sonar) is affected by sediment scattering and absorption effects in high-sediment water bodies, resulting in decreased measurement accuracy and expensive and complex operation. Optical remote sensing technology uses spectral information of satellite images to retrieve SSC, has the advantages of wide spatial coverage and fast data acquisition, and has become an important means of SSC monitoring. However, in optically complex water bodies (such as high-sediment, organic-rich or algal water bodies), traditional optical remote sensing retrieval models face the following challenges: 1) the strong scattering and absorption effects of suspended sediment interfere with the reflectance characteristics of the water body, leading to increased SSC retrieval error. 2) The water optical parameters (such as absorption coefficient, backscattering coefficient) of different sea areas differ significantly, and traditional global empirical models are difficult to adapt to different regional environmental characteristics. 3) SSC retrieval methods based on a single satellite sensor or empirical formula are difficult to ensure accuracy in cases of data scarcity or complex water conditions, especially when there is a lack of sufficient field data for model calibration.

[0004] In view of the above problems, the adaptability of traditional SSC monitoring methods in large-scale, data-scarce and optically complex environments is low, and it is difficult to meet the needs of global SSC dynamic change monitoring, so the present application is proposed. SUMMARY

[0005] In view of the problem of current large-scale, high-resolution suspended sediment concentration monitoring data scarcity, the present application proposes a method for monitoring suspended sediment in optically complex water bodies based on human-computer cooperation, which combines remote sensing image analysis, machine learning and expert knowledge to achieve high-precision retrieval of SSC.

[0006] The present application provides a method for monitoring suspended sediment in optically complex water bodies based on human-computer cooperation, which comprises:

[0007] collecting a multispectral remote sensing image of a target area and preprocessing the multispectral remote sensing image, performing spatio-temporal fusion on the preprocessed multispectral remote sensing image to obtain a high spatio-temporal resolution spectral data set;

[0008] performing manual labeling on the spectral data set to construct a suspended sediment concentration classification data set, and performing synthetic expansion on a few class samples based on a hydrodynamic similarity constraint to generate an equalized training set;

[0009] training a suspended sediment concentration inversion model using the equalized training set to obtain an initial suspended sediment concentration prediction value, and dynamically optimizing spectral feature weights and classification boundaries of the suspended sediment concentration inversion model based on expert knowledge, wherein the suspended sediment concentration inversion model adopts a regularized machine learning classifier model;

[0010] based on the hydrodynamic law and the spatial consistency constraint, locally correcting the initial suspended sediment concentration prediction value and diffusing it to the global range to obtain a global suspended sediment concentration prediction value.

[0011] In an embodiment of the present application, the preprocessed multispectral remote sensing image is spatio-temporal fused to obtain a high spatio-temporal resolution spectral data set, which specifically includes:

[0012] An improved target-based spatio-temporal adaptive reflectance fusion model is used to fuse the preprocessed multispectral remote sensing image to generate a fused image with high spatial resolution and high temporal resolution, wherein the calculation formula of the spatio-temporal adaptive reflectance fusion model is:

[0013] ,

[0014] wherein, represents the fused image; is a high spatial resolution Sentinel image; and respectively represent the ground reflectance of the MODIS image at time t and t-1, and the multispectral remote sensing image includes a Sentinel image and a MODIS image;

[0015] Based on the fused image, multi-band spectral features are extracted to obtain the high spatio-temporal resolution spectral data set.

[0016] In an embodiment of the present application, the spectral data set is manually labeled to construct a suspended sediment concentration classification data set, which includes:

[0017] Through a gridded labeling interface, an expert visually classifies the spectral data set, and divides the suspended sediment concentration mode into three categories: clear, translucent, and turbid;

[0018] By combining spectral and spatial features, suspended sediment concentration categories are divided into different groups, and a suspended sediment concentration classification dataset is constructed.

[0019] In one embodiment of the present invention, an expert-guided SMOTE variant is used to synthesize and expand minority class samples based on hydrodynamic similarity constraints, comprising:

[0020] For each minority class sample in the suspended sediment concentration classification dataset, the k nearest neighbor samples of that minority class sample are identified within a region of similar hydrodynamics, and the feature difference between the minority class sample and its nearest neighbor samples is calculated. This feature difference is expressed as:

[0021] ,

[0022] in, and Let i and j represent the feature vectors of minority class sample i and nearest neighbor sample j, respectively, and Δd be the interpolation difference.

[0023] Randomly generate interpolation coefficients. Based on the difference between these interpolation coefficients and the original interpolation, calculate the new sample and its feature vector, where:

[0024] The new sample is represented as: i + θ × Δd;

[0025] The feature vector of the new sample is represented as: ;

[0026] Where θ is the interpolation coefficient, θ∈(0,1), used to control the feature vector of the new sample. exist and The interpolation positions between them.

[0027] In one embodiment of the present invention, the spectrum of the suspended sediment concentration inversion model is based on expert knowledge.

[0028] Feature weights and classification boundaries are dynamically optimized, including:

[0029] Several key spectral features were selected from the balanced training set through spectral sensitivity analysis. Based on expert knowledge, these key spectral features were ranked according to their importance. Combined with hydrodynamic laws, spectral features that may lead to misclassification were identified, and the weights of the corresponding spectral features were adjusted.

[0030] Expert collaborative annotation is introduced in the water body category boundary area to correct the classification boundary of the suspended sediment concentration inversion model.

[0031] In one embodiment of the present invention, based on hydrodynamic laws and spatial consistency constraints, the initial predicted value of suspended sediment concentration is locally corrected to obtain a local predicted value of suspended sediment concentration, including:

[0032] The initial predicted suspended sediment concentration is traversed through a sliding window of a given size. Based on the principle of sediment transport, a sediment transport threshold is calculated within each window, and outliers exceeding the threshold are detected.

[0033] The outliers were manually corrected by experts based on hydrodynamic principles and measured values ​​of suspended sediment concentration, and the outliers were adjusted to conform to hydrodynamic principles.

[0034] The adjusted data was marked as high-confidence baseline data, and the suspended sediment concentration was re-entered.

[0035] Parameter optimization of the evolution model;

[0036] By incorporating spatial consistency constraints into the loss function of the suspended sediment concentration inversion model, the predicted values ​​of local suspended sediment concentrations are controlled to satisfy spatial consistency constraints within the neighborhood.

[0037] In one embodiment of the present invention, the process of locally correcting the initial predicted value of suspended sediment concentration further includes:

[0038] Numerical transformation processing is performed on the equalized training set, including dynamic filtering to remove or correct low-confidence data, adjusting the range of data using normalization or standardization methods, and / or denoising existing noise.

[0039] The processed equalized training set is marked as high-confidence baseline data and re-inputted into the suspended sediment concentration inversion model for parameter optimization.

[0040] Residual analysis is introduced to correct the error between the predicted and measured values ​​of suspended sediment concentration in the local area.

[0041] Establish a dynamic feedback mechanism to update and correct the high-confidence benchmark data;

[0042] Newly acquired multispectral remote sensing images are added to update the equalized training set.

[0043] In one embodiment of the present invention, the predicted value of the local suspended sediment concentration is expressed as:

[0044] ,

[0045] in, This represents the predicted value of local suspended sediment concentration; α is the weighting adjustment coefficient. is a measured value of the suspended sediment concentration; fi(x) is an optimized suspended sediment concentration inversion model, and x is a feature vector of the i th sample in the equalization training set.

[0046] In an embodiment of the present application, an adaptive diffusion mechanism is used to propagate local corrections to a global range, including:

[0047] A dynamic balance factor is introduced to control the influence weight of the local correction on the global range, and the global suspended sediment concentration prediction value is obtained, which is expressed as:

[0048] ,

[0049] wherein, is the global suspended sediment concentration prediction value, is the initial suspended sediment concentration

[0050] prediction value, is the local suspended sediment concentration prediction value, and P is a dynamic balance factor.

[0051] A statistical consistency optimization method is used to verify the data distribution rationality by calculating the mean, standard deviation and divergence of the data before and after correction, and the dynamic balance factor is adjusted in combination with the chi-square test;

[0052] A global consistency constraint is introduced in the model loss function, and the hyperparameters are dynamically adjusted in combination with the gradient descent optimization method to meet the hydrodynamic law and spatial consistency constraint.

[0053] In an embodiment of the present application, it further includes verifying the performance of the suspended sediment concentration inversion model, including:

[0054] The global suspended sediment concentration prediction value is visualized by using Taylor diagram, scatter plot and violin plot to evaluate the performance of the suspended sediment concentration inversion model;

[0055] The NSE efficiency is calculated to evaluate the fitting ability of the suspended sediment concentration inversion model;

[0056] The MK trend test is calculated to evaluate the significant change of the global suspended sediment concentration prediction value in the time dimension;

[0057] The conversion rate is calculated to quantify the concentration change rate of the global suspended sediment concentration prediction value between different seasonal changes;

[0058] The global suspended sediment concentration prediction value is evaluated by gridding analysis to evaluate the change characteristics of the global suspended sediment concentration prediction value with different seasonal changes.

[0059] From the above scheme, the advantages of the present application are:

[0060] The human-computer cooperation-based optical complex water body suspended sediment monitoring method disclosed by the application obtains a high-temporal-and-spatial-resolution spectral data set by collecting multispectral remote sensing images of a target area and performing pretreatment and spatiotemporal fusion; the spectral data set is manually labeled to construct a suspended sediment concentration classification data set, and a few samples are synthesized and expanded to generate an equalized training set; the suspended sediment concentration inversion model is trained by using the equalized training set to obtain an initial suspended sediment concentration prediction value, and the spectral feature weight and classification boundary of the suspended sediment concentration inversion model are dynamically optimized based on expert knowledge; the initial suspended sediment concentration prediction value is locally corrected and diffused to a global range based on the hydrodynamic law and spatial consistency constraint to obtain a global suspended sediment concentration prediction value. The method combines remote sensing image analysis, machine learning and expert knowledge to realize high-precision inversion of SSC. Moreover, the method iteratively optimizes the inversion result by the least artificial correction, thereby reducing the dependence on traditional field measurement, improving the monitoring precision and applicability. Meanwhile, the application follows the human-computer interaction principle, embeds human intelligence in SSC inversion, enhances the interpretability of the model, and ensures that the SSC inversion conforms to the hydrodynamic law. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 is the overall flowchart of the human-computer cooperation-based optical complex water body suspended sediment monitoring method provided by the embodiment of the application;

[0062] Figure 2 shows a spatial distribution diagram of measured suspended sediment concentration and estimated suspended sediment concentration obtained by executing the method of the application in a specific example; wherein, A diagram is a measured result of suspended sediment concentration, and B diagram is a suspended sediment concentration prediction value estimated by the inversion model;

[0063] Figure 3 is a performance comparison diagram of the suspended sediment concentration inversion models of the application and prior art for this example; A diagram uses Taylor diagram to compare the performance of the models, and B-E diagrams show the model scatter diagrams and NSE values, wherein, B diagram is model 1, C diagram is model 2, D diagram is model 3, and E diagram is model 4;

[0064] Figure 4 is a suspended sediment concentration distribution diagram estimated by the suspended sediment concentration inversion model of this example for the years from 2017 to 2024;

[0065] Figure 5Figures showing the spatial heterogeneity of annual SSC characteristics for this example; Figure a shows the Mann-Kendall (MK) trend test Z-score for SSC concentration in the coastal waters of China from 2017 to 2024, Figure b shows an enlarged view of the Yellow River Delta (A, 38°N) and the Yangtze River Estuary (B, 31°N) in 2017 and 2024, and the fusion of Sentinel-2 AMSI and MOD09A1 Terra surface reflectance products, and the superposition of the local Z-score from 2017 to 2024 on the satellite images of the Yellow River Delta and the Yangtze River Estuary in 2024;

[0066] Figure 6 Figures showing the seasonal heterogeneity of SSC characteristics for this example, where Figure A shows the seasonal distribution of SSC concentration from 2017 to 2024, and Figure B shows the SSC conversion rate between seasons from 2017 to 2024. DETAILED DESCRIPTION

[0067] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. In the following description, many specific details are set forth in order to provide a thorough understanding of the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the scope of the present application, so the present application is not limited to the specific implementations disclosed below.

[0068] In view of the current problem of lack of large-scale and high-resolution SSC monitoring data, the present application provides an optical complex water SSC monitoring method based on human-computer cooperation, which combines remote sensing image analysis, machine learning and expert knowledge to realize high-precision inversion of SSC. This method uses public high-resolution satellite images, estimates based on the relationship between backscattering coefficient and SSC, and iteratively optimizes the inversion results through minimum artificial correction, thereby reducing the dependence on traditional field measurements, improving monitoring accuracy and applicability. This method follows the Human-in-the-Loop (HITL) principle, embeds human intelligence in the three key stages of SSC inversion, enhances the interpretability of the model, and ensures that the SSC inversion conforms to the laws of hydrodynamics. Specifically, please refer to Figure 1 as shown, Figure 1 Figure 1 shows the overall flowchart of the optical complex water SSC monitoring method based on human-computer cooperation provided by an embodiment of the present application.

[0069] The optical complex water SSC monitoring method based on human-computer cooperation specifically comprises the following steps:

[0070] Step S1: Acquire multispectral remote sensing images of the target area and perform preprocessing. Then, perform spatiotemporal fusion on the preprocessed multispectral remote sensing images to obtain a high spatiotemporal resolution spectral dataset.

[0071] In one embodiment, several multispectral remote sensing images of the target area are first collected, such as Sentinel images and MODIS images. Sentinel images include multiple bands such as blue, green, red, near-infrared, and shortwave infrared to obtain detailed surface feature information; MODIS images have high temporal resolution and can be used to monitor short-term water body change trends. The selected images are ensured to cover the spatiotemporal range of the target area to guarantee data integrity and continuity. Simultaneously, measured data of the target area are collected, including measured values ​​of suspended sediment concentration at different locations, geographic coordinates (including latitude and longitude), and other information. Sampling time, sampling location, and data quality information are recorded.

[0072] In one embodiment, further preprocessing of the multispectral remote sensing imagery is required. First, image quality control and screening are performed, employing cloud detection and removal algorithms to eliminate images with severe cloud contamination, ensuring the image data quality meets analytical requirements. The radiometric consistency of the imagery is assessed, and data that does not meet quality standards is replaced or removed. Then, radiometric correction, atmospheric correction, and geometric correction are performed on the remote sensing imagery to ensure the quality and reliability of the image data. Radiometric correction removes sensor noise, restoring the actual radiance of the image and ensuring accurate and reliable radiance values. Atmospheric correction methods, such as the 6S model or other atmospheric correction methods, remove atmospheric interference factors such as aerosols and molecular scattering, improving the accuracy of the image's surface reflectance. Geometric correction, based on known geographic coordinates, performs orthorectification to eliminate geometric distortions caused by sensor location and terrain variations, ensuring the image is aligned with the actual geographic location.

[0073] Then, the preprocessed multispectral remote sensing images are spatiotemporally fused to obtain a high spatiotemporal resolution spectral dataset. Specifically, this involves using an improved target-based spatiotemporal adaptive reflectance fusion model (OS-STARFM) to fuse the preprocessed multispectral remote sensing images, generating a fused image with both high spatial and temporal resolution, thus improving the spatiotemporal continuity of surface information. The time...

[0074] The calculation formula for the spatial adaptive reflectivity fusion model is as follows:

[0075] (1),

[0076] in, Indicates fused imagery; For high spatial resolution Sentinel images; and respectively represent the surface reflectivity of the MODIS image at time t and t-1, and the multispectral remote sensing image includes a Sentinel image and a MODIS image.

[0077] Based on the fused image, multi-band spectral features are extracted, including blue, green, red, near-infrared, short-wave infrared, etc., to enhance the inversion capability of parameters such as suspended sediment concentration (SSC), and to construct a high spatio-temporal resolution spectral dataset.

[0078] In an embodiment, considering that the spectral bands of Sentinel-2 AMSI and MOD09A1 Terra are roughly equivalent, as shown in Table 1, the Sentinel image is selected from the Sentinel-2 AMSI image shown in Table 1, and the MODIS image is selected from the MOD09A1 Terra image.

[0079] Table 1

[0080]

[0081] Step S2, manually labeling the spectral dataset, constructing a suspended sediment concentration classification dataset, and synthesizing and expanding a few class samples based on hydrodynamic similarity constraints to generate an equalized training set.

[0082] In an embodiment, the spectral dataset is visually classified by an expert through a gridding labeling interface, the suspended sediment concentration mode is divided into three categories of clear (Clear), translucent (Translucent), and turbid (Turbid), as shown in formula (2), and the suspended sediment concentration categories are divided into different groups in combination with spectral-spatial features to construct a suspended sediment concentration classification dataset.

[0083] (2),

[0084] Wherein, C represents the suspended sediment concentration value, and the unit is g / m³.

[0085] In an embodiment, an expert-guided SMOTE variant is further used for the suspended sediment concentration classification dataset, a few class samples are synthesized and expanded based on hydrodynamic similarity constraints to improve data balance, to ensure that the interpolation process does not change the spatial boundary of the original sample, to improve the credibility of the generated data, and to generate an equalized training set.

[0086] Specifically, for each few class sample of the suspended sediment concentration classification dataset, k nearest neighbor samples of the few class sample are identified in a hydrodynamically similar region, and a feature difference between the few class sample and the nearest neighbor samples is calculated, and the feature difference is represented as: ; wherein, and respectively, and Δd is the interpolation difference.

[0087] Next, an interpolation coefficient is randomly generated, and a new sample and a feature vector of the new sample are calculated according to the interpolation coefficient and the interpolation difference, wherein the new sample is represented as i + θ × Δd, and the feature vector of the new sample is represented as: wherein θ is an interpolation coefficient, θ ∈ (0, 1), and is used to control the feature vector of the new sample. and between the interpolation positions. The process is repeatedly executed until each minority class sample is expanded to a data balanced state.

[0088] In step S3, the suspended sediment concentration inversion model is trained using the balanced training set to obtain an initial suspended sediment concentration prediction value, and the spectral feature weight and the classification boundary of the suspended sediment concentration inversion model are dynamically optimized based on expert knowledge, wherein the suspended sediment concentration inversion model adopts a regularized machine learning classifier model.

[0089] In an embodiment, a suspended sediment concentration inversion model is constructed based on a regularized machine learning classifier model. In an embodiment, the regularized machine learning classifier model adopts a random forest classifier model, and a λ regularization term is introduced in the training process to reduce the risk of overfitting. In order to ensure the robustness of the model, the hyperparameters of the random forest (RF) classifier are optimized to ensure the applicability of the suspended sediment concentration inversion model in different water environments and to improve the classification ability of complex water bodies. The random forest (RF) classifier is represented as:

[0090] (3), wherein, , , is a regularization coefficient, m is the number of trees in the random forest, represents the depth of the tree, represents the minimum sample split number, represents the minimum leaf node sample number, specifies the maximum number of features.

[0091] In an embodiment, the spectral feature weight and the classification boundary of the suspended sediment concentration inversion model are dynamically optimized based on expert knowledge to improve the precision and stability of the model. Through spectral sensitivity analysis and expert guided optimization, the applicability of the model in different water environments is ensured, and the errors caused by spectral complexity and hydrodynamic feature differences are reduced.

[0092] ​Specifically, by spectral sensitivity analysis, the influence of different wavebands on SSC inversion is evaluated, and several key spectral features are selected to construct an initial suspended sediment concentration (SSC) inversion model, as shown in equation (4),

[0093] (4),

[0094] wherein, represent several key spectral features of the water body.

[0095] In an embodiment, the several key spectral features are sorted according to importance based on expert knowledge, ensuring that the explanation ability of key wavebands and features to SSC changes conforms to the hydrodynamic law. In combination with the hydrodynamic law, spectral features that may lead to misclassification are identified, and the weights of the corresponding spectral features are adjusted to reduce model bias.

[0096] At the same time, in the boundary area of water body categories (such as the transition zone from nearshore to open sea), due to the complexity of hydrodynamics and optical properties, the model is prone to misclassification. To reduce the classification error in this area, expert collaborative labeling is introduced to correct the classification boundary of the suspended sediment concentration inversion model. Through expert collaborative labeling, the classification accuracy is improved. In combination with multi-temporal data, the SSC change trend is analyzed, and the classification strategy of the boundary area is adjusted to avoid errors caused by a single temporal image.

[0097] Step S4, based on the hydrodynamic law and spatial consistency constraint, locally corrects the initial suspended sediment concentration prediction value and diffuses it to the global range to obtain a global suspended sediment concentration prediction value.

[0098] In an embodiment, first, based on the initial suspended sediment concentration prediction value, local correction is performed in combination with the hydrodynamic principle and expert knowledge to obtain a local suspended sediment concentration prediction value, and data optimization is performed to improve the credibility and accuracy of the model. By introducing spatial consistency constraint and optimizing data quality, the inaccuracy caused by model error and data noise can be effectively reduced, thereby improving the stability and reliability of the inversion result.

[0099] Specifically, a sliding window of a given size (for example, 3x3) is used to traverse the initial suspended sediment concentration prediction value, and the sediment transport threshold value in each window is calculated based on the sediment transport principle. This process ensures that the change trend of suspended sediment concentration in the water body is consistent with the actual hydrodynamic process. Specifically, the transport of sediments is affected by factors such as water flow velocity and particle size distribution, and the SSC values in the neighborhood should have certain spatial continuity. Therefore, by using a sliding window, the sediment transport threshold value (see equation (5)) in each grid area is calculated to determine the reasonable fluctuation range of SSC in the area.

[0100] (5),

[0101] wherein, denotes the measured value of suspended sediment concentration, denotes the predicted value of suspended sediment concentration output by the model.

[0102] Then, outliers exceeding the threshold are detected, which are often caused by optical sensor errors, cloud cover, or noise in the transmission process. These outliers are marked as samples that need to be manually corrected, and further corrected through expert review. This process ensures the spatial consistency of the model results, avoiding misclassification or false prediction caused by individual outliers. The outliers are manually corrected by experts according to the laws of hydrodynamics and the measured values of suspended sediment concentration, adjusting the outliers to conform to the sediment transport laws of the actual water body, and marking the adjusted data as high-confidence reference data, re-inputting the suspended sediment concentration inversion model for parameter optimization, thereby improving the spatial consistency of the entire data set and improving the reliability of the inversion results.

[0103] At the same time, the spatial consistency constraint is introduced into the loss function of the suspended sediment concentration inversion model, controlling the local suspended sediment concentration prediction value to meet the spatial consistency constraint within the neighborhood range. In this way, the model can identify and correct local mutations and inconsistencies, effectively avoiding the influence of non-natural fluctuations on the model inversion results.

[0104] Further, the equalized training set is subjected to numerical transformation processing to optimize the distribution of training data, specifically including dynamically filtering out or correcting low-confidence data, adjusting the range of data using normalization or standardization methods, and / or denoising existing noise, etc. Specifically, in order to improve the training effect of the model, the original data is subjected to numerical transformation to remove the influence of low-confidence data and optimize the data distribution. First, low-confidence data is removed through dynamic filtering. These data may introduce noise due to optical interference, seasonal changes, or incompleteness of remote sensing images. By comparing expert-labeled data with model output, we filter out data points with low reliability and exclude or correct them. In order to optimize the data distribution, normalization or standardization methods are used to adjust the range of data, reducing the difference between feature values and improving the convergence speed and stability of model training. For noise that may exist in remote sensing data, wavelet transform or filtering algorithms are applied for denoising to improve the signal-to-noise ratio of the data, so that the model can more accurately learn effective information and improve the SSC inversion accuracy. The processed data is labeled as high-confidence reference data, which is re-input into the suspended sediment concentration inversion model for parameter optimization. Experts compare the remote sensing image inversion results with actual observation data to further adjust the SSC prediction values that do not conform to the physical laws or hydrodynamic model. These high-confidence reference data provide a reliable reference standard to help the model better adjust the prediction results and reduce errors.

[0105] In addition, in an embodiment, in order to reduce the deviation between the model estimated value and the true value, residual analysis is introduced to analyze the model error and correct the error between the local suspended sediment concentration prediction value and the measured suspended sediment concentration value. By observing the residual distribution, the estimation process of the model is adjusted to ensure that it is as close as possible to the actual SSC distribution of the water body. This process not only improves the accuracy of the model, but also enhances its adaptability in different types of water environments.

[0106] In addition, in an embodiment, as the water environment changes, new remote sensing data and field observation data continue to emerge. Therefore, a dynamic feedback mechanism is established to update the high-confidence reference data. After each data update, the model is retrained and adjusted based on the updated high-confidence reference data to ensure that it always reflects the latest water body change trend. At the same time, experts manually correct new data during each feedback to ensure that it conforms to the hydrodynamic laws and sediment transport characteristics.

[0107] In addition, to further improve the adaptive ability of the model, the training data is dynamically enhanced by adding newly collected multispectral remote sensing images and field observation data to update the multispectral remote sensing images, and then update the equalized training set, so that it can adapt to different environmental and seasonal conditions, improve the precision of SSC inversion, and finally realize stable and reliable suspended sediment concentration inversion.

[0108] After the local correction of the above process, the local suspended sediment concentration prediction value is represented as:

[0109] (6),

[0110] wherein, the local suspended sediment concentration prediction value; a is a weight adjustment coefficient; is the measured value of the suspended sediment concentration; x is the input of several key spectral features, f i (x) is the optimized suspended sediment concentration (SSC) inversion model, and x is the feature vector of the i-th sample in the equalized training set.

[0111] In an embodiment, to ensure the global consistency of the SSC inversion result, an adaptive diffusion mechanism is adopted, a dynamic balance factor is introduced, the local correction is propagated to the global range, and the cascading error caused by the local correction is avoided. At the same time, by using the weighted feature space adjustment of the approximation ideal solution ranking method, the dynamic balance factor is recalibrated, the global suspended sediment concentration prediction value is optimized, and the applicability and accuracy of the model in the optical complex water body environment are improved.

[0112] Specifically, after introducing the local correction, the influence of the local correction on the global SSC estimation result is calculated to ensure that the local adjustment will not amplify the error or produce cumulative deviation in the subsequent iteration process. For this purpose, a dynamic balance factor is introduced to control the influence weight of the local correction on the global range, and the global suspended sediment concentration prediction value is obtained, which avoids cascading errors, as shown in equation (7). The global suspended sediment concentration prediction value is represented as:

[0113] (7),

[0114] wherein, is the global suspended sediment concentration prediction value, is the initial suspended sediment concentration

[0115] prediction value, is the local suspended sediment concentration prediction value, and P is the dynamic balance factor.

[0116] In addition, in an embodiment, in order to prevent local correction from affecting the overall data distribution, a statistical consistency optimization method is adopted. By calculating the mean, standard deviation and divergence of the data before and after correction, the rationality of the data distribution is verified, and the dynamic balance factor is adjusted combined with the chi-square test to ensure that the mean and standard deviation of the SSC estimated value before and after correction remain consistent. At the same time, the Kullback-Leibler divergence is used to evaluate the change of data distribution, and the chi-square test is performed to ensure the rationality of the distribution of the corrected data. In addition, combined with the weight adjustment mechanism driven by expert knowledge, the corrected data is more consistent with the hydrodynamic law, and the credibility of the model is improved.

[0117] In addition, in an embodiment, in the global correction process, in order to reduce the spectral interference caused by optically complex water bodies (such as eutrophic water bodies, algal dense areas, and turbid estuary water areas), multispectral and short-wave infrared (SWIR) are combined to improve the adaptability of the model to high turbidity water bodies. At the same time, an adaptive spectral normalization method is used to reduce the interference of water color changes on SSC retrieval, and the prediction accuracy is improved combined with hydrodynamic simulation. In addition, an abnormal area recognition mechanism is established to dynamically adjust the correction propagation strategy, ensuring the stability of SSC retrieval in different environments.

[0118] In addition, in an embodiment, after the local correction is propagated to the global range, the loss function of the model is adjusted, and global consistency constraints are introduced into the model loss function to ensure the stability of the final SSC retrieval result. Combined with the gradient descent optimization method, the hyperparameters of the model are dynamically adjusted to meet the hydrodynamic law and spatial consistency constraints, and trend analysis is performed combined with long-term observation data to prevent systematic bias in SSC retrieval.

[0119] In addition, in an embodiment, in the global correction process, in order to continuously optimize the SSC retrieval model, a dynamic data feedback system is constructed, and error analysis is performed after each data update to adjust the local correction strategy and improve the accuracy of long-term prediction. At the same time, we introduce an automated anomaly detection algorithm combined with an active learning mechanism to improve the reliability of the model, and establish a long-term SSC monitoring system based on multi-temporal data to ensure the stability and adaptability of the model.

[0120] In addition, in an embodiment, a variety of statistical and visualization methods are further used to verify the performance of the suspended sediment concentration retrieval model and analyze its spatiotemporal variation trend.

[0121] Specifically, for model performance evaluation, in an embodiment, Taylor diagram and scatter plot are used to visualize the global suspended sediment concentration prediction value, and the performance of the model is evaluated. Among them, Taylor diagram comprehensively displays the correlation coefficient (r), standard deviation (Std. Dev) and center root mean square error (CRMSE) to measure the closeness of the model to the measured value.

[0122] In an embodiment, the Nash-Sutcliffe efficiency (NSE) is calculated to evaluate the fitting ability of the model. The closer the NSE value is to 1, the more accurate the model estimation is.

[0123] For the trend of spatial and temporal changes, in an embodiment, the Mann-Kendall (MK) trend test is used to evaluate the significant change of SSC in the time dimension. When |Z| > 1.96, p < 0.05, it indicates that there is a statistically significant trend.

[0124] In an embodiment, the conversion rate (CR) is calculated to quantify the concentration change rate between seasons. The CR is determined by calculating the percentage change of the global suspended sediment concentration prediction value between the current season and the next season for each grid cell.

[0125] In an embodiment, the global suspended sediment concentration prediction value is evaluated by gridding analysis to reveal the spatial distribution trend of different regions.

[0126] As shown in Figures 2-6 Figure 2 Fig. 1 shows the spatial distribution of the measured and estimated suspended sediment concentration values obtained by performing the method of the present application, wherein the left area A of the figure is the measured result of the suspended sediment concentration, and the right area B of the figure is the predicted value of the suspended sediment concentration estimated by the inversion model; Figure 3 Performance comparison of the suspended sediment concentration inversion model, model 1 is the man-machine collaborative model based on multi-source remote sensing data fusion of the present application, model 2 is the prior art man-machine collaborative model, model 3 is the empirical algorithm model based on multi-source remote sensing data fusion of the prior art, model 4 is the empirical algorithm model of the prior art, A graph uses Taylor diagram to compare the performance of the model, B-E graphs show the model scatter plot and NSE value, B graph is model 1, C graph is model 2, D graph is model 3, and E graph is model 4; Figure 4 Fig. 2 shows the annual suspended sediment concentration distribution from 2017 to 2024 estimated by the model of the example, and the mean value graph is the average value of the 8-year suspended sediment concentration distribution; Figure 5 ​For example, the spatial heterogeneity evaluation of the annual suspended sediment concentration characteristics is shown in the left figure, which shows the Mann-Kendall (MK) trend test Z score of the suspended sediment concentration in the coastal area of China from 2017 to 2024, showing a statistically significant trend (|Z|>1.96, p<0.05), the red dot indicates the increase of SSC during this period, and the blue dot indicates the decrease of SSC during this period, the enlarged view on the right shows the fusion satellite images of Sentinel-2A MSI and MOD09A1 Terra surface reflectance products in the Yellow River Delta (A, 38 °N) and the Yangtze River Estuary (B, 31 °N) in 2017 and 2024, and the local Z score from 2017 to 2024 is superimposed on the satellite images of the Yellow River Delta and the Yangtze River Estuary in 2024, showing that about 80% of the pixels in the Yellow River Delta significantly decreased (p<0.05), while the grid cells with increased SSC seem to be concentrated in the Yangtze River Estuary and its surrounding waters; Figure 6 For example, the seasonal heterogeneity evaluation of the suspended sediment concentration characteristics is shown in the figure, which shows the seasonal distribution of the suspended sediment concentration estimated by the model of the present application from 2017 to 2024, and the SSC conversion rate between seasons. It can be seen that the present method can accurately predict the suspended sediment concentration, meeting the demand for monitoring the dynamic changes of global SSC.

[0127] In summary, the optical complex water suspended sediment monitoring method based on human-computer cooperation disclosed in the present application combines remote sensing image analysis, machine learning and expert knowledge to realize high-precision inversion of SSC. The method uses public high-resolution satellite images to estimate based on the relationship between backscattering coefficient and SSC, and iteratively optimizes the inversion result through minimum artificial correction, thereby reducing the dependence on traditional field measurement, improving the monitoring accuracy and applicability. The present method follows the principle of human-computer interaction, embeds human intelligence in the SSC inversion, enhances the explainability of the model, and ensures that the SSC inversion conforms to the hydrodynamic law.

[0128] The above specific embodiments do not constitute a limitation on the scope of protection of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for monitoring suspended sediment in complex water bodies based on human-machine collaboration, characterized in that, The method includes: Multispectral remote sensing images of the target area are acquired and preprocessed. The preprocessed multispectral remote sensing images are then spatiotemporally fused to obtain a high spatiotemporal resolution spectral dataset. The spectral dataset was manually labeled to construct a suspended sediment concentration classification dataset, and minority class samples were synthetically expanded based on hydrodynamic similarity constraints to generate a balanced training set. The suspended sediment concentration inversion model is trained using the equalized training set to obtain the initial predicted value of suspended sediment concentration. The spectral feature weights and classification boundaries of the suspended sediment concentration inversion model are dynamically optimized based on expert knowledge. The suspended sediment concentration inversion model adopts a regularized machine learning classifier model. Based on hydrodynamic principles and spatial consistency constraints, the initial predicted suspended sediment concentration is locally corrected and diffused to the global range to obtain the global predicted suspended sediment concentration.

2. The method according to claim 1, characterized in that, The preprocessed multispectral remote sensing images are spatiotemporally fused to obtain a high spatiotemporal resolution spectral dataset, specifically including: An improved target-based spatiotemporal adaptive reflectance fusion model is used to fuse preprocessed multispectral remote sensing images, generating a fused image with both high spatial and temporal resolution. The calculation formula for the spatiotemporal adaptive reflectance fusion model is as follows: ， in, Indicates fused imagery; For high spatial resolution Sentinel images; and These represent the surface reflectance of the MODIS image at times t and t−1, respectively. The multispectral remote sensing image includes Sentinel image and MODIS image. Based on the fused image, multi-band spectral features are extracted to obtain the high spatiotemporal resolution spectral dataset.

3. The method according to claim 1, characterized in that, The spectral dataset was manually labeled to construct a suspended sediment concentration classification dataset, which includes: Through a gridded annotation interface, experts visually classified the spectral dataset, dividing the suspended sediment concentration patterns into three categories: clear, translucent, and turbid. By combining spectral and spatial features, suspended sediment concentration categories are divided into different groups, and a suspended sediment concentration classification dataset is constructed.

4. The method according to claim 3, characterized in that, Using an expert-guided variant of SMOTE, a synthetic expansion of minority class samples is performed based on hydrodynamic similarity constraints, including: For each minority class sample in the suspended sediment concentration classification dataset, the k nearest neighbor samples of that minority class sample are identified within a region of similar hydrodynamics, and the feature difference between the minority class sample and its nearest neighbor samples is calculated. This feature difference is expressed as: in, and Let i and j represent the feature vectors of minority class sample i and nearest neighbor sample j, respectively, and Δd be the interpolation difference. Randomly generate interpolation coefficients. Based on the difference between these interpolation coefficients and the original interpolation, calculate the new sample and its feature vector, where: The new sample is represented as: i + θ × Δd; The feature vector of the new sample is represented as: ; Where θ is the interpolation coefficient, θ∈(0,1), used to control the feature vector of the new sample. exist and The interpolation positions between them.

5. The method according to claim 4, characterized in that, Based on expert knowledge, the suspended sludge The spectral feature weights and classification boundaries of the sand concentration inversion model are dynamically optimized, including: Several key spectral features were selected from the equalized training set through spectral sensitivity analysis. Based on expert knowledge, these key spectral features were ranked according to their importance. Combined with hydrodynamic laws, spectral features that may lead to misclassification were identified, and the weights of the corresponding spectral features were adjusted. Expert collaborative annotation is introduced in the water body category boundary area to correct the classification boundary of the suspended sediment concentration inversion model.

6. The method according to claim 5, characterized in that, Based on hydrodynamic principles and spatial consistency constraints, the initial predicted suspended sediment concentration is locally corrected to obtain a local predicted suspended sediment concentration, including: The initial predicted suspended sediment concentration is traversed through a sliding window of a given size. Based on the principle of sediment transport, a sediment transport threshold is calculated within each window, and outliers exceeding the threshold are detected. The outliers were manually corrected by experts based on hydrodynamic principles and measured values ​​of suspended sediment concentration, and the outliers were adjusted to conform to hydrodynamic principles. The adjusted data was marked as high-confidence baseline data, and the suspended sediment concentration was re-entered. Parameter optimization of the evolution model; By incorporating spatial consistency constraints into the loss function of the suspended sediment concentration inversion model, the predicted values ​​of local suspended sediment concentrations are controlled to satisfy spatial consistency constraints within the neighborhood.

7. The method according to claim 6, characterized in that, Regarding the initial suspended sediment concentration The process of locally correcting the degree prediction value further includes: Numerical transformation processing is performed on the equalized training set, including dynamic filtering to remove or correct low-confidence data, adjusting the range of data using normalization or standardization methods, and / or denoising existing noise. The processed equalized training set is marked as high-confidence baseline data and re-inputted into the suspended sediment concentration inversion model for parameter optimization. Residual analysis is introduced to correct the error between the predicted and measured values ​​of suspended sediment concentration in the local area. Establish a dynamic feedback mechanism to update and correct the high-confidence benchmark data; Newly acquired multispectral remote sensing images are added to update the equalized training set.

8. The method according to claim 7, characterized in that, The local suspended sediment concentration The predicted value is expressed as: ， in, This represents the predicted value of local suspended sediment concentration; α is the weighting adjustment coefficient. is the measured value of suspended sediment concentration; fi(x) is the optimized suspended sediment concentration inversion model, and x is the feature vector of the i-th sample in the equalization training set.

9. The method according to claim 8, characterized in that, Adopting an adaptive diffusion mechanism, Local corrections are propagated to the global extent, including: A dynamic balance factor is introduced to control the weight of the local correction on the global range, resulting in the predicted global suspended sediment concentration, which is expressed as follows: ， in, This is the predicted value for the global suspended sediment concentration. Initial suspended sediment concentration Predicted value P represents the predicted local suspended sediment concentration, and P is the dynamic equilibrium factor. The statistical consistency optimization method was adopted to verify the rationality of the data distribution by calculating the mean, standard deviation and divergence of the data before and after correction, and the dynamic balance factor was adjusted by combining the chi-square test. A global consistency constraint is introduced into the model loss function, and the hyperparameters are dynamically adjusted by combining the gradient descent optimization method to satisfy the hydrodynamic laws and spatial consistency constraints.

10. The method according to claim 1, characterized in that, It also includes: verifying the performance of the suspended sediment concentration inversion model, including: Taylor plots, scatter plots, and violin plots were used to visualize the global suspended sediment concentration predictions and to evaluate the performance of the suspended sediment concentration inversion model. Calculate the NSE efficiency and evaluate the fitting ability of the suspended sediment concentration inversion model; Calculate the MK trend test to assess the significance of the global suspended sediment concentration prediction over time. Calculate the conversion rate to quantify the rate of change of the global suspended sediment concentration prediction values ​​across different seasonal variations; The variation characteristics of the global suspended sediment concentration prediction values ​​with different seasons were evaluated through gridded analysis.

Images