Multi-source data fusion and analysis method for remote sensing monitoring of rivers and lakes

Abstract

The invention relates to the field of river and lake water quality monitoring, and discloses a river and lake remote sensing monitoring multi-source data fusion and analysis method, which comprises the following steps: firstly, collecting satellite remote sensing images, unmanned aerial vehicle aerial images and ground monitoring station data of a river and lake area, and preprocessing to form a feature data set; and then, constructing an image classification model based on a convolutional neural network (CNN), a water quality evaluation model based on a support vector machine (SVM) and a water quality prediction model based on time sequence analysis, and respectively classifying, evaluating and predicting water quality. And finally, by defining a multi-source data fusion algorithm, carrying out fusion analysis on a model output result, obtaining a comprehensive water quality evaluation result, and generating a river and lake remote sensing monitoring report. The advantages of multiple data sources are fully utilized, comprehensive, rapid and accurate monitoring of the water quality of rivers and lakes is achieved, powerful support is provided for environmental protection and water resource management, and the method has wide application prospects and important practical value.

Description

Technical Field

[0001] The present invention relates to the field of river and lake water quality monitoring, and in particular to a multi-source data fusion and analysis method for river and lake remote sensing monitoring. Background Art

[0002] As environmental problems become increasingly severe, river and lake water quality monitoring has become an important part of environmental protection and water resources management. Traditional river and lake water quality monitoring methods mainly rely on regular sampling and laboratory analysis at ground monitoring stations. Although this method can provide relatively accurate water quality data, it has problems such as limited monitoring range, long monitoring cycle, and high cost, and it is difficult to meet the needs of large-scale, real-time monitoring. In recent years, the development of remote sensing technology has provided new solutions for river and lake water quality monitoring. Remote sensing technology has the advantages of wide monitoring range, strong timeliness, and relatively low cost, and can achieve rapid and large-scale monitoring of river and lake water quality. However, a single remote sensing data source often has problems such as insufficient information and limited accuracy, which makes it difficult to meet the needs of complex water quality monitoring. Therefore, how to use multi-source remote sensing data for fusion and analysis to improve the accuracy and reliability of water quality monitoring has become a research hotspot in the field of remote sensing water quality monitoring.

[0003] At present, remote sensing monitoring technology for rivers and lakes has made significant progress, but there are still some challenges. On the one hand, remote sensing data from different sources differ in format, resolution, observation angle, etc., and effective preprocessing and feature extraction are required to form a feature data set that can be used for subsequent analysis. On the other hand, a single water quality classification or assessment model is often difficult to fully reflect the complex conditions of river and lake water quality, and a combination of multiple models and methods is required for comprehensive analysis. In addition, the water quality of rivers and lakes is affected by many factors, such as meteorological conditions, hydrological conditions, human activities, etc. The changes in these factors are time-series. Therefore, how to use time series analysis methods to predict the changing trend of river and lake water quality is also an important direction of current research. Summary of the invention

[0004] The purpose of the present invention is to provide a multi-source data fusion and analysis method for river and lake remote sensing monitoring to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a multi-source data fusion and analysis method for river and lake remote sensing monitoring, the method comprising: S1: Collect remote sensing data of river and lake areas, including satellite remote sensing images, drone aerial images and ground monitoring station data; S2: Preprocess the collected multi-source data, including data format conversion, coordinate system 1, cloud and fog removal, data denoising and data calibration, and then perform feature extraction to form a feature data set; S3: Build an image classification model based on convolutional neural network (CNN), which is trained using satellite remote sensing images and drone aerial image feature datasets to classify river and lake water quality; S4: Construct a water quality assessment model based on support vector machine (SVM), which uses the feature data in the preprocessed multi-source dataset to learn and identify patterns of different water quality types; S5: Construct a water quality prediction model based on time series analysis, which uses historical water quality data and environmental factors to predict the water quality trend of rivers and lakes in the future; S6: Input the feature data set into the image classification model, water quality assessment model and water quality prediction model respectively, and obtain the image classification results based on CNN, the water quality assessment results based on SVM and the water quality change trend based on time series analysis; S7: Define a multi-source data fusion algorithm, perform fusion analysis on the model output results, and obtain comprehensive water quality assessment results; S8: Generate a remote sensing monitoring report on rivers and lakes based on the comprehensive water quality assessment results, including water quality type, pollution status, future change trends and corresponding recommended measures.

[0006] Preferably, the satellite remote sensing images include optical remote sensing images and radar remote sensing images; the UAV aerial images include visible light images and infrared images; and the ground monitoring station data include water quality indicators such as water temperature, pH value, dissolved oxygen, turbidity, ammonia nitrogen, total phosphorus and total nitrogen.

[0007] Preferably, the image classification model adopts a VGG network structure, extracts image features through multi-layer convolution and pooling operations, and finally outputs the classification results through a fully connected layer. Its network structure includes: Input layer: receives preprocessed satellite remote sensing images and drone aerial images, and the image size is unified as [H, W, C], where H is the height, W is the width, and C is the number of channels; Convolution layer: multi-layer convolution operation is used, and each convolution layer is followed by a ReLU activation function to extract image features; the convolution kernel size is set to 3x3 and the step size is set to 1; Pooling layer: Use maximum pooling or average pooling operation after the convolution layer to reduce the size of the feature map and the amount of calculation while retaining important features; Fully connected layer: Flatten the feature map output by the pooling layer into a one-dimensional vector and classify it through multiple layers of fully connected layers; the number of output nodes of the last fully connected layer is the number of classification categories, and the Softmax function is used to output the probability of each category; Output layer: output the results of water quality classification.

[0008] Preferably, the training step of the image classification model includes: A1: Initialize the parameters of the CNN model, including convolution kernel weights, bias terms, and learning rates; A2: Divide the preprocessed image feature dataset into a training set and a validation set. The training set is used for model training, and the validation set is used for model validation. A3: Use the training set to iteratively train the CNN model. In each iteration, the forward propagation calculates the model output and calculates the loss between the predicted value and the true value according to the loss function; the loss function uses the cross entropy loss function, and its calculation formula is: in, is the true category label, is the category probability predicted by the model, and N is the number of samples; A4: Update the model parameters based on the loss value through the back propagation algorithm and gradient descent method; A5: Repeat steps A3 to A4 until the loss converges or the preset number of iterations is reached; A6: Use the validation set to validate the trained CNN model and evaluate the model’s classification accuracy, precision, and recall. A7: When the model performance reaches the preset standard, save the model parameters and obtain the trained image classification model.

[0009] Preferably, the water quality assessment model adopts RBF kernel function, separates samples of different water quality categories by finding the optimal hyperplane, and realizes the assessment of water quality status. Its network structure includes: Input layer: Receives pre-processed ground monitoring station data features, including water temperature, pH value, dissolved oxygen, turbidity, ammonia nitrogen, total phosphorus and total nitrogen water quality indicators; Kernel function layer: uses RBF kernel function to map input features to high-dimensional space; Decision layer: Find the optimal hyperplane in high-dimensional space to separate samples of different water quality categories; Output layer: output water quality assessment results.

[0010] Preferably, the steps of training the water quality assessment model include: B1: Use the preprocessed multi-source dataset as the training set and select water quality type as the label; B2: Use grid search and cross-validation methods to determine the optimal penalty parameter C and Gaussian kernel parameter γ; B3: Use the optimal parameters to train the SVM model until the model converges; B4: Use the validation set to validate the SVM model and evaluate the classification accuracy and generalization ability of the model; B5: When the model performance reaches the preset standard, the model parameters are saved to obtain the trained water quality assessment model.

[0011] Preferably, the water quality prediction model is constructed using a long short-term memory network LSTM algorithm, and its training steps include: C1: Filter out historical water quality data and environmental factor data from the preprocessed multi-source data set to form a time series data set; C2: Perform a stationarity test on the time series data set. If it is not stationary, perform difference processing until it is stable; C3: Parameter estimation of the LSTM model, including parameters of trend term, seasonal term and random term; C4: Use the training set to train the model and minimize the prediction error through the gradient descent method; C5: Use the validation set to validate the model and evaluate the prediction accuracy and stability of the model; C6: When the model performance reaches the preset standard, the model parameters are saved to obtain the trained water quality prediction model.

[0012] Preferably, the specific implementation steps of defining the multi-source data fusion algorithm in step S7 include: Step 1: Define the water quality type output by the image classification model as D_CNN, the water quality assessment result output by the water quality assessment model as D_SVM, and the water quality change trend output by the water quality prediction model as D_Trend; Step 2: For D_CNN and D_SVM, if the water quality types or assessment results output by the two are inconsistent, a weighted average is performed according to the confidence or accuracy of each model to obtain the fused water quality type or assessment result D_Fusion_Type; Step 3: For D_Trend, combine it with the fused water quality type or assessment result D_Fusion_Type to form a comprehensive water quality assessment result that includes the current water quality status and future change trends; Step 4: If there are multiple water quality indicators or pollution factors, the above fusion process is performed on each indicator or factor to obtain their respective comprehensive evaluation results; Step 5: Based on the comprehensive evaluation results of all water quality indicators or pollution factors, the weighted average method is used to obtain the final comprehensive water quality evaluation result D_Final; Step 6: Compare D_Final with the preset water quality standards or thresholds to determine the water quality level, pollution status and future change trend of rivers and lakes.

[0013] Preferably, the data denoising includes processing missing values ​​in the data, and the processing method is: For numerical features, the median of the non-missing values ​​of the feature is used to fill the missing values; For categorical features, the missing values ​​are filled with the mode of the non-missing values ​​of the feature; For missing values ​​in time series data, the average value of the adjacent values ​​is used to fill them; For features with obvious business meanings and few missing values, they are manually filled in according to business logic.

[0014] Preferably, the data calibration processing method includes: For remote sensing image data, geometric calibration is performed, including image registration, distortion correction and projection transformation; For numerical data, value calibration is performed. By comparing with the benchmark data, the deviation and error of the remote sensing data are adjusted to match the benchmark data.

[0015] Compared with the prior art, the present invention has the following beneficial effects: By integrating satellite remote sensing images, drone aerial images and ground monitoring station data, the present invention greatly expands the scope of water quality monitoring and realizes comprehensive monitoring from local to global, from static to dynamic. At the same time, the rapid response capability of remote sensing technology enables the present invention to provide water quality information in real time or near real time, meeting the needs of large-scale, real-time monitoring.

[0016] The fusion of multi-source data effectively makes up for the lack of information and accuracy of a single data source. The present invention ensures the consistency and accuracy of the data through sophisticated preprocessing and feature extraction steps. Combining the image classification model of convolutional neural network (CNN), the water quality assessment model of support vector machine (SVM) and the water quality prediction model of time series analysis, the present invention can more comprehensively and accurately reflect the actual situation of river and lake water quality, and improve the accuracy and reliability of monitoring. The present invention not only pays attention to the current water quality situation, but also predicts the changing trend of river and lake water quality through the time series analysis model, providing forward-looking decision support for environmental protection and water resources management. At the same time, the application of multi-source data fusion algorithm enables the present invention to comprehensively consider multiple influencing factors, obtain more comprehensive water quality assessment results, and provide a strong basis for formulating scientific and reasonable governance measures. Compared with traditional ground monitoring methods, the present invention uses remote sensing technology to achieve remote and non-contact monitoring, which greatly reduces the monitoring cost and improves the monitoring efficiency. At the same time, the automated data processing and analysis process reduces manual intervention, making the monitoring work more efficient and convenient. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A processing step diagram of a multi-source data fusion and analysis method for river and lake remote sensing monitoring according to the present invention; Figure 2 Flowchart for building an image classification model using the VGG network; Figure 3Flowchart of multi-source data fusion algorithm. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] See also Figure 1-3 The present invention provides a technical solution: a multi-source data fusion and analysis method for river and lake remote sensing monitoring, the method comprising: S1: Data collection. Satellite remote sensing images of river and lake areas are obtained through satellite remote sensing technology, drone aerial images are obtained through aerial photography, and ground monitoring data provided by ground monitoring stations are collected to form a multi-source data set.

[0020] S2: Data preprocessing. The collected multi-source data are converted to ensure uniform data formats; coordinate system processing is performed to make all data in the same coordinate system; cloud and fog removal is performed to improve image quality; data denoising and data calibration are performed to eliminate noise and errors in the data; finally, feature extraction is performed to form a feature data set for subsequent analysis.

[0021] S3: Build an image classification model. Based on the convolutional neural network (CNN) technology, build an image classification model. Use the pre-processed satellite remote sensing image and drone aerial image feature datasets to train the model so that it can accurately classify river and lake water quality.

[0022] S4: Construct a water quality assessment model. Use support vector machine (SVM) technology to construct a water quality assessment model. Using the feature data in the preprocessed multi-source dataset, train the model to learn and identify patterns of different water quality types to provide support for water quality assessment.

[0023] S5: Construct a water quality prediction model. Based on time series analysis technology, a water quality prediction model is constructed. By inputting historical water quality data and environmental factors, the model can predict the water quality change trend of rivers and lakes in the future and provide early warning for water quality management.

[0024] S6: Model application and result acquisition. The preprocessed feature data set is input into the image classification model, water quality assessment model and water quality prediction model respectively to obtain the image classification results based on CNN, the water quality assessment results based on SVM and the water quality change trend based on time series analysis.

[0025] S7: Data fusion analysis. Define the multi-source data fusion algorithm and perform fusion analysis on the obtained model output results. Comprehensively consider the results of different models to obtain more comprehensive and accurate comprehensive water quality assessment results.

[0026] S8: Report generation. Generate a river and lake remote sensing monitoring report based on the comprehensive water quality assessment results. The report includes water quality type, pollution status, future change trends, and recommended measures for these situations, providing scientific basis and decision-making support for environmental protection and water resources management.

[0027] The present invention will be further described below in conjunction with Examples 1 to 5: Embodiment 1: This embodiment describes the collection method of multiple data sources and the method of building an image classification model. : The specific methods of data collection include: Satellite remote sensing images: including optical remote sensing images, which are used to capture the natural color and reflective characteristics of the surface of rivers and lakes, and radar remote sensing images, which are used to obtain information such as the roughness of the surface of rivers and lakes, water level changes, etc.

[0028] Drone aerial images: including visible light images, which are used to visually display the surface conditions of rivers and lakes, such as color, floating objects, etc., and infrared images, which are used to detect the temperature distribution on the surface of rivers and lakes and identify possible pollution hotspots.

[0029] Ground monitoring station data: including key water quality indicators such as water temperature, pH value, dissolved oxygen, turbidity, ammonia nitrogen, total phosphorus and total nitrogen, providing direct water quality parameter information.

[0030] The VGG network is used to build an image classification model, and its network structure is as follows: Input layer: Receives preprocessed satellite remote sensing images and drone aerial images. The image size is unified as [H, W, C], where H represents the image height, W represents the image width, and C represents the number of image channels (such as RGB three channels).

[0031] Convolutional layer: Design multiple layers of convolution operations, and each layer of convolution is followed by a ReLU activation function to extract image features in a nonlinear way. The convolution kernel size is uniformly set to 3x3, and the step size is set to 1 to ensure the effective extraction of fine-grained features.

[0032] Pooling layer: After the convolution layer, the maximum pooling or average pooling operation is used to reduce the size of the feature map, reduce the computational burden, and retain key feature information.

[0033] Fully connected layer: Flatten the feature map output by the pooling layer into a one-dimensional vector, and further integrate and classify the features through multiple layers of fully connected layers. The number of output nodes in the last fully connected layer corresponds to the number of water quality classification categories, and the Softmax function is used to calculate the probability distribution of each category.

[0034] Output layer: Determine the final result of water quality classification based on the output of the Softmax function.

[0035] The specific steps for training the image classification model are as follows: A1: Initialize the parameters of the CNN model, including the weights of the convolution kernel, bias terms, and learning rates, to ensure that the starting point of the model training is reasonable. A2: Divide the preprocessed image feature data set into a training set and a validation set in a certain proportion. The training set is used for the model training process, and the validation set is used to evaluate the performance of the model. A3: Use the training set to iteratively train the CNN model. In each iteration, the output of the model is calculated through forward propagation, and the loss between the predicted value and the true value is calculated based on the cross entropy loss function. The calculation formula of the cross entropy loss function is: in, is the true category label, is the category probability predicted by the model, and N is the number of samples. A4: According to the calculated loss value, update the model parameters through the back propagation algorithm and the gradient descent method to minimize the loss function. A5: Repeat steps A3 to A4 until the loss function converges or reaches the preset number of iterations to ensure that the model is fully trained. A6: Use the validation set to verify the trained CNN model and evaluate the model's key performance indicators such as classification accuracy, precision, and recall to verify the model's generalization ability. A7: When the model's performance reaches the preset standard, save the model's parameters and obtain the trained image classification model for subsequent water quality classification tasks.

[0036] Embodiment 2: The water quality assessment model is constructed based on support vector machine (SVM), and its core lies in using radial basis function (RBF) as kernel function to achieve efficient water quality classification and assessment. This embodiment describes in detail the implementation steps and network structure of the water quality assessment model, as follows: The water quality assessment model network structure includes: Input layer: This layer is responsible for receiving the pre-processed ground monitoring station data features, including key water quality indicators such as water temperature, pH value, dissolved oxygen, turbidity, ammonia nitrogen, total phosphorus and total nitrogen. These indicators serve as the input features of the model and provide a direct basis for water quality assessment.

[0037] Kernel function layer: RBF kernel function, also known as Gaussian kernel function, is used, and its form is K(x, y) = exp(-gamma* ||xy||^2); where: x and y are input data points, which can be multidimensional vectors. ||xy||^2 represents the square of the Euclidean distance between x and y. Gamma (γ) is a hyperparameter used to control the shape of the RBF kernel function. The larger the value of gamma, the more selective the function is, that is, the smaller the influence range of a single training sample; conversely, the smaller the value of gamma, the weaker the selectivity of the function, and the larger the influence range of a single training sample. . The RBF kernel function can map the input features to a high-dimensional space, making the originally linearly inseparable data linearly separable in the high-dimensional space, thereby enhancing the classification ability of the model.

[0038] Decision layer: In high-dimensional space, the SVM algorithm separates samples of different water quality categories by finding the optimal hyperplane. This hyperplane is determined by the support vector, which maximizes the interval between different categories, thereby achieving effective evaluation of water quality.

[0039] Output layer: According to the optimal hyperplane found by the decision layer, the model outputs the water quality assessment result, that is, determines which water quality category the current water quality sample belongs to.

[0040] The steps to train a water quality assessment model include: B1. Data preparation: Use the preprocessed multi-source dataset as the training set, where each sample contains water quality indicator characteristics such as water temperature, pH value, dissolved oxygen, and the corresponding water quality type label. These labels are determined based on actual water quality monitoring results and are used to guide the model training process.

[0041] B2. Parameter optimization: Grid search and cross-validation methods are used to determine the optimal penalty parameter C and Gaussian kernel parameter γ. Grid search finds the parameter combination that optimizes model performance by traversing a series of candidate values ​​in the parameter space. Cross-validation is used to evaluate the performance of the model under each parameter combination to avoid overfitting and underfitting problems.

[0042] B3. Model training: Use the optimal parameters C and γ to train the SVM model. During the training process, the model continuously adjusts internal parameters through an iterative optimization algorithm until the convergence condition is reached, that is, the model performance is no longer significantly improved or the preset number of iterations is reached.

[0043] B4. Model validation: Use the validation set to validate the trained SVM model and evaluate the classification accuracy and generalization ability of the model. The validation set is a data set independent of the training set and is used to test the model's ability to predict unknown data. By calculating the accuracy, precision, recall and other indicators of the model on the validation set, the performance of the model can be fully evaluated.

[0044] B5. Model saving: When the model performance reaches the preset standard, the model parameters and configuration information are saved to obtain a trained water quality assessment model. This model can be deployed in the actual water quality monitoring system to provide accurate water quality assessment results for river and lake remote sensing monitoring.

[0045] Embodiment 3: In the field of water quality monitoring and management, in order to achieve effective prediction of future water quality conditions, the present invention constructs a water quality prediction model based on a long short-term memory network (LSTM). The LSTM algorithm is particularly suitable for predicting time series data such as water quality, which is affected by multiple factors and has complex changes, because it can capture long-term dependencies in time series data. This embodiment describes in detail the training steps and implementation methods of the water quality prediction model, specifically including: The water quality prediction model training step comprises: C1. Data preparation and time series construction: Carefully select historical water quality data (such as water temperature, pH value, dissolved oxygen and other key indicators) and related environmental factor data (such as temperature, rainfall, wind speed, etc.) from the preprocessed multi-source data set. Arrange these data in chronological order to form a time series data set.

[0046] C2. Time series stationarity test and processing: Perform a stationarity test on the constructed time series data set. Common methods include ADF (Augmented Dickey-Fuller) test or KPSS (Kwiatkowski-Phillips-Schmidt-Shin) test. If the test result shows that the time series is not stationary, differential processing (such as first difference, second difference, etc.) is used until the time series becomes stable. Stationarity is crucial to the accuracy and stability of the time series prediction model.

[0047] C3. LSTM model parameter estimation: Before building the LSTM model, the model parameters need to be preliminarily estimated. These parameters include trend terms (reflecting the long-term trend of the time series), seasonal terms (reflecting the seasonal changes of the time series), and random terms (reflecting the random fluctuations in the time series). Parameter estimation can be performed through statistical methods or empirical settings to provide reasonable initial values ​​for subsequent model training.

[0048] C4. Model training and parameter optimization: Use the training set to train the LSTM model. During the training process, the model parameters are continuously adjusted through the gradient descent method Adam to minimize the prediction error. The prediction error can be measured by loss functions, such as mean square error (MSE), mean absolute error (MAE), etc. Through continuous iterative training, the model gradually learns the inherent laws and characteristics of time series data.

[0049] C5. Model validation and performance evaluation: Use the validation set to validate the trained LSTM model. The validation set is a data set independent of the training set and is used to evaluate the prediction accuracy and stability of the model. By calculating the prediction error, accuracy, R² score and other indicators of the model on the validation set, the performance of the model is comprehensively evaluated. At the same time, a comparison chart between the predicted value and the true value can be drawn to intuitively display the prediction effect of the model.

[0050] C6. Model preservation and deployment: When the model performance reaches the preset standard, save the model parameters and configuration information to obtain a trained water quality prediction model. Deploy the trained model to the actual water quality monitoring system to provide strong support for future water quality prediction. By inputting new water quality data and environmental factor data in real time, the model can output future water quality prediction results to provide a basis for water quality management and decision-making.

[0051] Embodiment 4: In the water quality monitoring method of multi-source data fusion and analysis, in order to obtain more comprehensive and accurate water quality assessment results, a multi-source data fusion algorithm is defined, which aims to integrate the output results of the image classification model, the water quality assessment model and the water quality prediction model to form a comprehensive water quality report including the current water quality status, future change trends and comprehensive assessment of various water quality indicators or pollution factors. This embodiment describes in detail the specific implementation steps of the multi-source data fusion algorithm: Step 1: Define model output The water quality type output by the image classification model (such as convolutional neural network CNN) is defined as D_CNN, which makes a preliminary judgment on the water quality category based on remote sensing images or water body images taken on site.

[0052] The water quality assessment result output by the water quality assessment model (such as support vector machine SVM) is defined as D_SVM. The model uses the water quality index data provided by the ground monitoring station to perform a more detailed water quality assessment.

[0053] The water quality change trend output by the water quality prediction model (such as the long short-term memory network LSTM) is defined as D_Trend, which predicts the possible change direction of water quality in the future.

[0054] Step 2: Integrate water quality type and assessment results For the outputs of D_CNN and D_SVM, if the water quality type or assessment results given by the two are inconsistent, a weighted average is performed based on the confidence or accuracy of each model. The confidence or accuracy can be evaluated through cross-validation during model training or performance in actual applications.

[0055] The formula for weighted averaging is: D_Fusion_Type=α×D_CNN+(1−α)×D_SVM, where α is the weighting coefficient, which is dynamically adjusted according to the model performance.

[0056] Step 3: Combine water quality trends The output of the water quality prediction model, D_Trend, is combined with the fused water quality type or assessment result, D_Fusion_Type, to form a comprehensive description of the current water quality status and future change trends. This helps decision makers understand the dynamic changes in water quality and take corresponding management measures.

[0057] Step 4: Multi-index fusion analysis If water quality monitoring involves multiple indicators or pollution factors (such as water temperature, pH value, dissolved oxygen, turbidity, ammonia nitrogen, etc.), the above fusion process is carried out for each indicator or factor separately to obtain their respective comprehensive evaluation results.

[0058] This means that for each water quality indicator, there is a fused current status assessment and a forecast of future change trends.

[0059] Step 5: Calculation of comprehensive evaluation results Based on the comprehensive evaluation results of all water quality indicators or pollution factors, the weighted average method is used to calculate the final comprehensive water quality evaluation result D_Final. The weighting coefficient can be determined according to the importance or relevance of each indicator to ensure the comprehensiveness and accuracy of the evaluation results.

[0060] Step 6: Determination of water quality level and pollution status D_Final is compared with preset water quality standards or thresholds, which are usually based on national or local environmental protection laws and regulations, industry standards or scientific research results. Based on the comparison results, the water quality level of rivers and lakes (such as excellent, good, slightly polluted, moderately polluted, heavily polluted, etc.), pollution status (such as major pollution factors, pollution degree, etc.) and future change trends (such as improvement, stability, deterioration, etc.) are determined.

[0061] Example 5: In the process of data processing and analysis, data denoising and data calibration are key steps to ensure data quality and improve analysis accuracy. In view of the characteristics of water quality monitoring data, this example describes in detail the specific implementation methods of data denoising and data calibration.

[0062] Data denoising aims to process missing values ​​in data to reduce data incompleteness and improve data availability. The following processing methods are used for different types of missing values: Processing of missing values ​​of numerical features: For numerical features (such as water temperature, pH value, dissolved oxygen concentration, etc.), first count the non-missing values ​​of the feature and calculate the median. Use the median to fill in the missing values ​​in the feature. As a highly robust statistic, the median can effectively avoid the impact of extreme values ​​on the filling results.

[0063] Missing value processing of categorical features: For categorical features (such as water body type, pollution source, etc.), count the non-missing values ​​of the feature and find the mode with the highest frequency. Use the mode to fill the missing values ​​in the feature. The mode reflects the most common value in the data and is suitable for missing value filling of categorical data.

[0064] Missing value processing of time series data: For time series data (such as daily water quality monitoring data), if there are missing values, the average value of the adjacent values ​​before and after is used to fill them. Specifically, for the two valid data points before and after the missing value, their average value is calculated and used as the filling value of the missing value. This method can maintain the continuity and smoothness of time series data.

[0065] Missing value processing for features with business meaning: For features with obvious business meaning and few missing values ​​(such as specific event records, special water quality indicators, etc.), they are manually filled according to business logic. This requires an in-depth understanding of the business scenarios behind the data and the causes of missing values, and then reasonable filling based on business rules or expert experience.

[0066] Data calibration aims to ensure the accuracy and consistency of data, especially for remote sensing image data and numerical data, which require special calibration processing.

[0067] Remote sensing image data calibration: Geometric calibration: perform image registration on remote sensing images to ensure that the geographic location in the image matches the actual geographic location; perform distortion correction to eliminate image deformation caused by the sensor or shooting angle; perform projection conversion to convert the image from the original projection coordinate system to the target projection coordinate system for overlay analysis with other geographic data. The specific implementation of geometric calibration can be completed with the help of professional remote sensing image processing software or algorithm library, through steps such as selecting control points and performing polynomial fitting.

[0068] Numerical data calibration: Calibration: For numerical data (such as water quality parameters derived from remote sensing), the deviation and error of remote sensing data are evaluated by comparing with benchmark data (such as ground measured data, historical data or standard data).

[0069] Based on the evaluation results, the remote sensing data is adjusted to match the benchmark data. Adjustment methods may include linear regression, nonlinear correction, or calibration models based on machine learning. The purpose of value calibration is to eliminate systematic errors and random errors in remote sensing data and improve the accuracy and reliability of the data.

[0070] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0071] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-source data fusion and analysis method for river and lake remote sensing monitoring, characterized in that: The method comprises: S1: Collect remote sensing data of river and lake areas, including satellite remote sensing images, drone aerial images and ground monitoring station data; S2: Preprocess the collected multi-source data, including data format conversion, coordinate system 1, cloud and fog removal, data denoising and data calibration, and then perform feature extraction to form a feature data set; S3: Build an image classification model based on convolutional neural network (CNN), which is trained using satellite remote sensing images and drone aerial image feature datasets to classify river and lake water quality; S4: Construct a water quality assessment model based on support vector machine (SVM), which uses the feature data in the preprocessed multi-source dataset to learn and identify patterns of different water quality types; S5: Construct a water quality prediction model based on time series analysis, which uses historical water quality data and environmental factors to predict the water quality trend of rivers and lakes in the future; S6: Input the feature data set into the image classification model, water quality assessment model and water quality prediction model respectively, and obtain the image classification results based on CNN, the water quality assessment results based on SVM and the water quality change trend based on time series analysis; S7: Define a multi-source data fusion algorithm, perform fusion analysis on the model output results, and obtain comprehensive water quality assessment results; S8: Generate a remote sensing monitoring report on rivers and lakes based on the comprehensive water quality assessment results, including water quality type, pollution status, future change trends and corresponding recommended measures.

2. The multi-source data fusion and analysis method for river and lake remote sensing monitoring according to claim 1 is characterized in that: The satellite remote sensing images include optical remote sensing images and radar remote sensing images; the UAV aerial images include visible light images and infrared images; the ground monitoring station data include water quality indicators such as water temperature, pH value, dissolved oxygen, turbidity, ammonia nitrogen, total phosphorus and total nitrogen.

3. The multi-source data fusion and analysis method for river and lake remote sensing monitoring according to claim 1 is characterized in that: The image classification model adopts the VGG network structure, extracts image features through multi-layer convolution and pooling operations, and finally outputs the classification results through the fully connected layer. Its network structure includes: Input layer: receives preprocessed satellite remote sensing images and drone aerial images, and the image size is unified as [H, W, C], where H is the height, W is the width, and C is the number of channels; Convolution layer: multi-layer convolution operation is used, and each convolution layer is followed by a ReLU activation function to extract image features; the convolution kernel size is set to 3x3 and the step size is set to 1; Pooling layer: Use maximum pooling or average pooling operation after the convolution layer to reduce the size of the feature map and the amount of calculation while retaining important features; Fully connected layer: Flatten the feature map output by the pooling layer into a one-dimensional vector and classify it through multiple layers of fully connected layers; the number of output nodes of the last fully connected layer is the number of classification categories, and the Softmax function is used to output the probability of each category; Output layer: output the results of water quality classification.

4. The multi-source data fusion and analysis method for river and lake remote sensing monitoring according to claim 3 is characterized in that: The training steps of the image classification model include: A1: Initialize the parameters of the CNN model, including convolution kernel weights, bias terms, and learning rates; A2: Divide the preprocessed image feature dataset into a training set and a validation set. The training set is used for model training, and the validation set is used for model validation. A3: Use the training set to iteratively train the CNN model. In each iteration, the forward propagation calculates the model output and calculates the loss between the predicted value and the true value according to the loss function; the loss function uses the cross entropy loss function, and its calculation formula is: in, is the true category label, is the category probability predicted by the model, and N is the number of samples; A4: Update the model parameters based on the loss value through the back propagation algorithm and gradient descent method; A5: Repeat steps A3 to A4 until the loss converges or the preset number of iterations is reached; A6: Use the validation set to validate the trained CNN model and evaluate the model’s classification accuracy, precision, and recall. A7: When the model performance reaches the preset standard, save the model parameters and obtain the trained image classification model.

5. The multi-source data fusion and analysis method for river and lake remote sensing monitoring according to claim 1 is characterized in that: The water quality assessment model adopts RBF kernel function to separate samples of different water quality categories by finding the optimal hyperplane to evaluate the water quality. Its network structure includes: Input layer: Receives pre-processed ground monitoring station data features, including water temperature, pH value, dissolved oxygen, turbidity, ammonia nitrogen, total phosphorus and total nitrogen water quality indicators; Kernel function layer: uses RBF kernel function to map input features to high-dimensional space; Decision layer: Find the optimal hyperplane in high-dimensional space to separate samples of different water quality categories; Output layer: output water quality assessment results.

6. The multi-source data fusion and analysis method for river and lake remote sensing monitoring according to claim 5 is characterized in that: The steps to train a water quality assessment model include: B1: Use the preprocessed multi-source dataset as the training set and select water quality type as the label; B2: Use grid search and cross-validation methods to determine the optimal penalty parameter C and Gaussian kernel parameter γ; B3: Use the optimal parameters to train the SVM model until the model converges; B4: Use the validation set to validate the SVM model and evaluate the classification accuracy and generalization ability of the model; B5: When the model performance reaches the preset standard, the model parameters are saved to obtain the trained water quality assessment model.

7. The multi-source data fusion and analysis method for river and lake remote sensing monitoring according to claim 1 is characterized in that: The water quality prediction model is constructed using the long short-term memory network LSTM algorithm, and its training steps include: C1: Filter out historical water quality data and environmental factor data from the preprocessed multi-source data set to form a time series data set; C2: Perform a stationarity test on the time series data set. If it is not stationary, perform difference processing until it is stable; C3: Parameter estimation of the LSTM model, including parameters of trend term, seasonal term and random term; C4: Use the training set to train the model and minimize the prediction error through the gradient descent method; C5: Use the validation set to validate the model and evaluate the prediction accuracy and stability of the model; C6: When the model performance reaches the preset standard, the model parameters are saved to obtain the trained water quality prediction model.

8. The multi-source data fusion and analysis method for river and lake remote sensing monitoring according to claim 1 is characterized in that: The specific implementation steps of defining the multi-source data fusion algorithm in step S7 include: Step 1: Define the water quality type output by the image classification model as D_CNN, the water quality assessment result output by the water quality assessment model as D_SVM, and the water quality change trend output by the water quality prediction model as D_Trend; Step 2: For D_CNN and D_SVM, if the water quality types or assessment results output by the two are inconsistent, a weighted average is performed according to the confidence or accuracy of each model to obtain the fused water quality type or assessment result D_Fusion_Type; Step 3: For D_Trend, combine it with the fused water quality type or assessment result D_Fusion_Type to form a comprehensive water quality assessment result that includes the current water quality status and future change trends; Step 4: If there are multiple water quality indicators or pollution factors, the above fusion process is performed on each indicator or factor to obtain their respective comprehensive evaluation results; Step 5: Based on the comprehensive evaluation results of all water quality indicators or pollution factors, the weighted average method is used to obtain the final comprehensive water quality evaluation result D_Final; Step 6: Compare D_Final with the preset water quality standards or thresholds to determine the water quality level, pollution status and future change trend of rivers and lakes.

9. The multi-source data fusion and analysis method for river and lake remote sensing monitoring according to claim 1 is characterized in that: The data denoising includes processing missing values ​​in the data, and the processing method is as follows: For numerical features, the median of the non-missing values ​​of the feature is used to fill the missing values; For categorical features, the missing values ​​are filled with the mode of the non-missing values ​​of the feature; For missing values ​​in time series data, the average value of the adjacent values ​​is used to fill them; For features with obvious business meanings and few missing values, they are manually filled in according to business logic.

10. The multi-source data fusion and analysis method for river and lake remote sensing monitoring according to claim 1 is characterized in that: The data calibration processing method includes: For remote sensing image data, geometric calibration is performed, including image registration, distortion correction and projection transformation; For numerical data, value calibration is performed. By comparing with the benchmark data, the deviation and error of the remote sensing data are adjusted to match the benchmark data.

Images