River and Lake Intelligent Monitoring and Management System Based on Remote Sensing and GIS

Abstract

This invention relates to the field of environmental monitoring and remote sensing information processing technology, specifically disclosing a river and lake intelligent monitoring and management system based on remote sensing and GIS. The system constructs a reflectance data cube with consistent spatiotemporal dimensions through a data preprocessing module; extracts temporal features such as fluctuation amplitude and duration using a temporal feature analysis module; performs multi-scale segmentation and morphological analysis through a spatial feature extraction module; integrates spectral, spatiotemporal, and spatial features to generate multi-dimensional feature vectors using a feature fusion module; and finally, utilizes an intelligent judgment module combining multi-level decision trees, fuzzy reasoning, and case matching to achieve intelligent judgment.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring and remote sensing information processing technology, specifically to a smart river and lake monitoring and management system based on remote sensing and GIS. Background Technology

[0002] While ground-based manual sampling and laboratory analysis offer high accuracy, they suffer from inherent drawbacks such as low monitoring frequency, limited spatial coverage, and high cost. In recent years, remote sensing-based monitoring methods, utilizing spectral characteristic indices and threshold segmentation techniques, have achieved rapid monitoring of large-scale water bodies to a certain extent, providing a new technological approach for water environment protection.

[0003] However, existing technologies have significant shortcomings: traditional methods mainly rely on single-temporal remote sensing images for analysis, which cannot effectively solve the spectral confusion problems caused by "same material, different spectra" and "different material, same spectra." Specifically, areas with high suspended matter formed by ship-driven sediment stirring up bottom sediment are highly similar in spectral characteristics to algal blooms, making accurate differentiation difficult based solely on single-temporal image features, leading to an increased misjudgment rate. These technical deficiencies severely restrict the in-depth application and operational promotion of remote sensing technology in the field of intelligent river and lake monitoring. Summary of the Invention

[0004] The purpose of this invention is to provide a smart monitoring and management system for rivers and lakes based on remote sensing and GIS, so as to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] The intelligent monitoring and management system for rivers and lakes based on remote sensing and GIS includes:

[0007] The data preprocessing module is used to acquire multi-temporal remote sensing data of the target water area, perform radiometric calibration, atmospheric correction and spatiotemporal registration on the multi-temporal remote sensing data, and generate a reflectance data cube with consistent spatiotemporal dimensions.

[0008] The time-series feature analysis module is used to extract the reflectance time series of each pixel point from the reflectance data cube and calculate the time domain change characteristics, including the fluctuation amplitude and duration period index of reflectance change.

[0009] The spatial feature extraction module is used to identify high reflectivity anomaly areas based on reflectivity data cubes, and to extract the morphological features of the anomaly areas and their spatial relationship features with shoreline and sewage outlet geographical elements.

[0010] The feature fusion module is used to fuse the reflectance reference value, temporal variation features and spatial morphological relationship features of each pixel to generate a multidimensional feature vector for each pixel.

[0011] The intelligent analysis module is used to analyze and judge multi-dimensional feature vectors and output the type of water quality anomaly event and the corresponding confidence level.

[0012] As a further aspect of the present invention: the specific process of generating a reflectivity data cube with consistent spatiotemporal dimensions is as follows:

[0013] Radiometric calibration was performed on the acquired multi-temporal remote sensing data to convert the raw digital quantization values ​​into apparent reflectance and eliminate differences in observation geometry.

[0014] By using a radiative transfer model combined with real-time meteorological parameters to eliminate atmospheric absorption and scattering effects, the true surface reflectance data can be obtained.

[0015] By using feature point matching and geometric precision correction techniques, reflectance data from different time periods are unified into the same geographic coordinate system, generating a reflectance data cube with consistent spatiotemporal dimensions.

[0016] As a further aspect of the present invention: the extraction of the reflectance time series of each pixel point specifically includes:

[0017] The reflectance observations of each pixel in all time phases are obtained from the reflectance data cube to form an initial time series. The dynamic time warping algorithm is used to align the observation time points between different time phases to eliminate the time offset caused by the inconsistency of satellite revisit periods. The data of the missing time periods are reconstructed to generate a complete and time-aligned reflectance time series.

[0018] As a further aspect of the present invention: the specific process for calculating the time-domain variation characteristics includes:

[0019] Based on the reflectance time series, the local change intensity at each time phase point is calculated using the sliding window method to obtain the fluctuation amplitude index; at the same time, the duration period index is determined by detecting the length of abnormal periods that continuously exceed the threshold; finally, combined with time series waveform analysis, the abrupt change points and trend characteristics of reflectance changes are extracted to form a comprehensive time domain change characteristic.

[0020] As a further aspect of the present invention: the process of identifying high reflectivity anomalous regions includes:

[0021] Multi-scale segmentation of the reflectance data cube is performed to generate image objects with spectral homogeneity. Dynamic threshold segmentation combined with local reflectance statistical features and spatial context information is used to identify potential high reflectance anomalous regions. Morphological filtering is used to optimize the initial detection results, eliminate discrete noise points and retain connected anomalous regions, form accurate boundaries of high reflectance anomalous regions, and generate accurate high reflectance anomalous regions.

[0022] As a further aspect of the present invention: the extraction of morphological features of abnormal areas and their spatial relationship with shorelines and sewage outlet geographical elements specifically includes:

[0023] Based on the identified high reflectivity anomaly areas, their morphological characteristics, including regional compactness and boundary tortuosity, are calculated. At the same time, through spatial buffer analysis and topological relationship judgment, the distance and orientation relationships between the anomaly areas and geographical elements such as shorelines and sewage outlets are quantified. Finally, spatial overlay analysis is used to establish a spatial correlation matrix between the anomaly areas and geographical elements, generating comprehensive spatial relationship characteristics.

[0024] As a further aspect of the present invention: the fusion of the reflectance reference value, temporal variation characteristics, and spatial morphological relationship characteristics of each pixel specifically includes:

[0025] A normalization processing channel is established between the reflectance benchmark value, time-domain variation characteristics, and spatial morphological relationship characteristics to eliminate the dimensional differences between different feature dimensions; weights are assigned according to the importance of each feature dimension to the identification of water quality anomalies; and redundant information between features is eliminated and complementary features are enhanced to form an optimized fusion feature set.

[0026] As a further aspect of the present invention: the generation of the multidimensional feature vector for each pixel specifically includes:

[0027] Based on the fusion feature set, principal component analysis is used to compress and reduce the dimensionality of high-dimensional features, retaining the feature components with the largest variance. Then, feature recombination technology is used to arrange the dimensionality-reduced feature components in a structured manner according to the spatiotemporal dimension. Finally, a multidimensional feature vector with a fixed dimension is generated, which simultaneously contains comprehensive information of spectral features, temporal variation features, and spatial relationship features.

[0028] As a further aspect of the present invention: the analysis and judgment of the multidimensional feature vectors specifically includes:

[0029] The multidimensional feature vectors are input into the decision tree model, and preliminary classification is performed based on the preset spatiotemporal feature thresholds. Fuzzy logic reasoning is used to perform joint probability analysis on the features of each dimension in the feature vectors to calculate the probability of belonging to various types of water quality anomalies. The current feature pattern is matched with historical typical event cases for similarity matching to complete the comprehensive analysis and judgment.

[0030] As a further aspect of the present invention: the output of water quality anomaly event types and corresponding confidence levels specifically includes:

[0031] Based on the analysis results, the system integrates multi-dimensional analysis conclusions to eliminate decision-making uncertainty; calculates the reliability of the final judgment result and generates a confidence level; establishes a mapping table between event types and confidence levels, and outputs analysis conclusions that include specific water quality anomaly event types and their corresponding confidence levels.

[0032] The beneficial effects of this invention are:

[0033] (1) This invention overcomes the limitations of traditional single-phase analysis by constructing a multi-phase reflectance data cube. The system uses a dynamic time warping algorithm to reconstruct the complete time series, accurately captures the temporal pattern of reflectance changes, calculates fluctuation amplitude indicators through sliding window analysis, identifies continuous abnormal periods to determine the duration period, and extracts abrupt change features by combining time series waveform decomposition. In the spatial dimension, the boundaries of abnormal areas are accurately defined through multi-scale segmentation and morphological optimization, quantifies their compactness, tortuosity and other morphological indicators, and establishes a spatial correlation matrix with geographical elements such as shorelines and sewage outlets. Finally, the spatiotemporal features and spectral features are integrated through the feature fusion module, and the attention mechanism is used for weighted fusion. The results are then input into the intelligent judgment module for multi-method collaborative analysis: decision tree for preliminary classification, fuzzy logic for calculating membership probability, case reasoning for similarity matching, and finally, high-confidence judgment results are output through evidence theory synthesis. It solves the problem of accurately distinguishing between spectrally similar phenomena such as ship turbidity (strong transient nature and linear distribution) and algal blooms (strong persistence and patchy distribution), thus elevating water quality anomaly identification from traditional qualitative judgment to a stage of quantitative and typological precision identification.

[0034] (2) This invention constructs a fully automated processing system from raw data to intelligent decision-making, completely reconstructing the traditional business process of river and lake monitoring. The system automatically completes radiometric calibration, atmospheric correction, and spatiotemporal registration of multi-source remote sensing data through the data preprocessing module, generating a standardized reflectance data cube; the time series feature analysis module automatically extracts pixel-level time series and calculates dynamic indicators such as fluctuation amplitude and duration period; the spatial feature extraction module realizes automatic identification and feature quantification of abnormal areas through multi-scale segmentation and morphological optimization; the feature fusion module automatically integrates multi-dimensional features using a weighted fusion algorithm based on an attention mechanism; and the intelligent judgment module automatically outputs judgment results through multi-model collaborative reasoning. The entire processing process requires no manual intervention, reducing reliance on professional personnel and making it possible to conduct routine and accurate monitoring of large-scale water areas and to respond quickly to sudden water environment events. Attached Figure Description

[0035] The invention will now be further described with reference to the accompanying drawings.

[0036] Figure 1 This is a flowchart of the system of the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Please see Figure 1 As shown, this invention is a river and lake intelligent monitoring and management system based on remote sensing and GIS, comprising:

[0039] The data preprocessing module is used to acquire multi-temporal remote sensing data of the target water area, perform radiometric calibration, atmospheric correction and spatiotemporal registration on the multi-temporal remote sensing data, and generate a reflectance data cube with consistent spatiotemporal dimensions.

[0040] The time-series feature analysis module is used to extract the reflectance time series of each pixel point from the reflectance data cube and calculate the time domain change characteristics, including the fluctuation amplitude and duration period index of reflectance change.

[0041] The spatial feature extraction module is used to identify high reflectivity anomaly areas based on reflectivity data cubes, and to extract the morphological features of the anomaly areas and their spatial relationship features with shoreline and sewage outlet geographical elements.

[0042] The feature fusion module is used to fuse the reflectance reference value, temporal variation features and spatial morphological relationship features of each pixel to generate a multidimensional feature vector for each pixel.

[0043] The intelligent analysis module is used to analyze and judge multi-dimensional feature vectors and output the type of water quality anomaly event and the corresponding confidence level.

[0044] The data preprocessing module converts the raw digital quantized values ​​acquired by satellite sensors into physically meaningful apparent reflectance. First, sensor calibration parameters, including gain coefficients and offsets, are obtained; these parameters are provided by the satellite data provider in the metadata. Then, a solar elevation angle correction model is used to eliminate the effects of differences in observation geometry, based on the solar azimuth and elevation angle data at the time of imaging. Specifically, a linear transformation formula is applied to the digital quantized value of each pixel to convert it into the apparent reflectance value of the top layer of the atmosphere. This step ensures the comparability of data acquired at different times and from different sensors, laying a solid foundation for subsequent processing.

[0045] This is a crucial step in eliminating atmospheric absorption and scattering effects to obtain accurate surface reflectance data. This system employs a radiative transfer model-based approach combined with real-time meteorological parameters for precise atmospheric correction. First, meteorological data collected during imaging is gathered, including parameters such as aerosol optical thickness, water vapor content, and ozone content. Then, an improved dark pixel adaptive correction algorithm, specifically optimized for water body characteristics, is used. This algorithm first identifies dark pixels in the image (such as clear, deep water areas), then retrieves atmospheric parameters based on the reflectance characteristics of these pixels, thereby establishing an atmospheric correction model. During implementation, atmospheric correction coefficients are calculated separately for each band to ensure accurate correction across different spectral ranges. This process significantly improves the accuracy of water reflectance data, providing a reliable data foundation for subsequent analysis.

[0046] A crucial step in unifying reflectance data from different time phases to the same geographic coordinate system is the application of a feature point matching algorithm. First, a stable feature point (such as river inflection points or island outlines) is automatically identified on the imagery of each time phase. Then, geometric precision correction techniques are used to establish a transformation model between the images based on these feature points. Specifically, a quadratic polynomial transformation model is employed, and the transformation parameters are calculated using the least squares method to accurately register the imagery of each time phase to a unified geographic coordinate system. Finally, resampling techniques are used to resample the imagery of all time phases to the same spatial resolution, generating a reflectance data cube with completely consistent spatiotemporal dimensions. This data cube maintains the pixel-level correspondence between the data from different time phases, ensuring the accuracy of subsequent time-series analysis.

[0047] In the time-series feature analysis module, which is one of the core analysis modules of this system, it is responsible for extracting feature information with temporal regularity from the preprocessed reflectance data cube. This module realizes time-series feature analysis through two key processing stages: reflectance time series extraction and time-domain variation feature calculation. The specific implementation methods and technical details of each stage are described in detail below:

[0048] Reflectance time series extraction is a fundamental step in time series analysis, aiming to construct a complete and time-aligned reflectance observation sequence for each cell. This process comprises three specific implementation steps:

[0049] First, reflectance observations for each pixel across all time phases are extracted from a spatiotemporally consistent reflectance data cube. Each pixel in the data cube contains reflectance data from multiple time phases, which have undergone radiometric calibration, atmospheric correction, and geometric registration, possessing consistent physical meaning and spatial location. The system iterates through the time dimension of the data cube, sequentially reading the pixel values ​​for each time phase to form an initial time series. This initial series contains reflectance observations for the pixel at different time points; however, due to factors such as satellite revisit cycles and cloud cover, the series may contain uneven time intervals and missing data.

[0050] Secondly, a dynamic time warping algorithm is employed to align observation time points across different time phases. This algorithm automatically adjusts the time axis of the time series, eliminating time shifts caused by inconsistent satellite revisit periods. Specifically, the algorithm first establishes a reference time axis based on uniform time intervals (e.g., daily or every five days). Then, it uses dynamic programming to calculate the optimal correspondence between actual observation time points and the reference time points, thus realigning the time series. This process not only considers the matching of time points but also maintains the physical consistency of reflectivity values.

[0051] Finally, the data for the missing periods are reconstructed to generate a complete time series. The system employs a spatiotemporal co-interpolation method, utilizing both temporal and spatial correlations to estimate missing values. For short-term missing data (e.g., 1-2 time phases), a cubic spline interpolation method in the time domain is used; for long-term missing data (e.g., more than 3 time phases), co-interpolation is performed by combining spatial information from adjacent pixels. A quality control mechanism is implemented during the interpolation process: when the duration of missing data exceeds 30% of the total sequence length, the pixel is marked as low-quality data and excluded from subsequent analysis.

[0052] The calculation of time-domain variation features involves quantifying the characteristic patterns of reflectance changes over time based on the extracted time series data. This process is achieved through three levels of analysis:

[0053] The first level employs a sliding window method to calculate the intensity of local changes and obtain a fluctuation amplitude index. The size of the sliding window is set according to specific application requirements, typically 3 to 5 time points. For each time point in the sequence, the reflectance statistical characteristics within its window are calculated, including the mean, standard deviation, and extreme values. The fluctuation amplitude index is obtained by calculating the difference in reflectance between the current window and the previous window, specifically using a relative rate of change calculation method. To enhance the robustness of the index, the system also smooths the calculation results, employing Gaussian filtering to eliminate the influence of random noise.

[0054] The second level determines the persistence period index by detecting the length of abnormal periods that continuously exceed a threshold. The abnormal threshold is set based on historical statistical characteristics, taking the 95th percentile of the historical reflectance value of the pixel. The system traverses the entire time series, identifies all time points that exceed the threshold, and then calculates the length of the period that continuously exceeds the threshold. The persistence period index considers not only the absolute length of the period but also factors such as the magnitude of the exceedance and the trend of change for a comprehensive evaluation. To distinguish between different types of persistence patterns, the system also defines two modes: sudden persistence and gradual persistence, each corresponding to different discrimination rules.

[0055] The third level involves combining time-series waveform analysis to extract abrupt changes and trend characteristics in reflectance changes. The time-series waveform analysis employs a multi-scale decomposition method, breaking down the original sequence into trend components, periodic components, and residual components. The trend component is extracted using the moving average method to characterize the long-term trend of reflectance changes; the periodic component is extracted using Fourier analysis to identify potential periodic change patterns; and the residual component contains short-term abrupt change information. Abrupt change detection is based on the statistical properties of the residual component; when the residual value exceeds three times its standard deviation, the point is considered to have abrupt change characteristics. The system also defines a quantitative index for trend strength, characterizing the significance of the trend by calculating the magnitude of change per unit time.

[0056] In the spatial feature extraction module, the reflectance data cube is first segmented at multiple scales. This multi-scale segmentation employs a region-growing-based algorithm, dividing the image into spectrally homogeneous image objects by calculating the spectral similarity and spatial proximity between pixels. Three different scale parameters are set during the segmentation process: small scale (segmentation scale parameter 10) for identifying small anomalies, medium scale (segmentation scale parameter 30) for identifying medium-sized anomaly regions, and large scale (segmentation scale parameter 50) for identifying large-scale anomaly regions. Segmentation results at each scale level are retained, forming a multi-scale set of segmented objects. The segmentation process also considers the reflectance characteristics of different spectral bands, assigning higher weights to the near-infrared band, as this band is most sensitive to anomalies in water bodies.

[0057] Secondly, a dynamic threshold segmentation algorithm combining local reflectance statistical features and spatial context information is employed to identify potential high-reflectance anomaly regions. The dynamic threshold is set based on the reflectance statistical features within a local window, with the window size set to 11×11 pixels. For each window, its mean and standard deviation of reflectance are calculated, and the threshold is set to the mean plus twice the standard deviation. Simultaneously, spatial context information is introduced, considering the spectral characteristics of adjacent regions to avoid misclassifying isolated high-reflectance points as anomalies. For regions near the shoreline, the threshold is appropriately increased to avoid misclassifying shoreline reflection as water anomalies. Region growing technology is also used during the identification process, gradually expanding from a seed point to ensure the spatial continuity of detected anomaly regions.

[0058] Finally, morphological filtering was used to optimize the initial detection results. A combination of opening and closing operations was employed. Opening used 3×3 structuring elements to eliminate small discrete noise points, while closing used 5×5 structuring elements to fill small holes within the region, maintaining connectivity. An area threshold was set during the filtering process to remove excessively small regions with an area less than 10 pixels, as these regions were likely noise or small-scale perturbations. Simultaneously, the boundaries were smoothed using a mathematical morphology boundary optimization algorithm to make the region boundaries more natural and smooth. This resulted in accurate boundaries for high-reflectivity anomaly regions, providing a precise spatial range for subsequent feature extraction.

[0059] Spatial relationship feature extraction involves identifying anomalous regions and quantifying their morphological and spatial distribution characteristics. This process is achieved through three levels of analysis:

[0060] The first level involves calculating morphological characteristic indicators of the anomalous regions. Regional compactness is assessed by calculating the ratio of the region's area to its perimeter; a ratio closer to 1 indicates a more compact shape. Boundary tortuosity is quantified by calculating the fractal dimension of the boundary line using box counting; a higher value indicates a more complex and tortuous boundary. The aspect ratio and orientation of the region are also calculated. The aspect ratio is obtained by the ratio of the longer side to the shorter side of the smallest bounding rectangle, and the orientation is represented by the angle between the region's principal axis and true north. These morphological characteristic indicators can effectively distinguish different types of anomalous regions; for example, algal blooms appear as clumps or sheets, while areas affected by sewage outlets often exhibit a banded or feather-like distribution.

[0061] The second level quantifies the spatial relationship between anomalous areas and geographic features through spatial buffer analysis and topological relationship judgment. First, geographic feature buffer zones are established for the shoreline and sewage outlets, with the shoreline buffer zone set at 100 meters and the sewage outlet buffer zone at 500 meters. Then, the spatial relationship between the anomalous areas and these buffer zones is calculated, including the shortest distance, overlapping area, and overlapping ratio. Distance relationships are quantified by calculating the shortest distance from the centroid of the anomalous area to the geographic feature. Orientation relationships are represented by calculating the azimuth angle of the anomalous area relative to the geographic feature, dividing the circumference into eight directions (north, northeast, east, southeast, south, southwest, west, and northwest), and statistically analyzing the distribution of anomalous areas in each direction. Simultaneously, the topological relationship between the anomalous areas and geographic features is analyzed, including intersecting, disjoint, and contained relationships.

[0062] The third level employs spatial overlay analysis to establish a spatial correlation matrix between anomalous areas and geographic features. Spatial overlay analysis uses layer overlay technology, superimposing the anomalous area layer with the geographic feature layer. For each anomalous area, its spatial correlation with each geographic feature is calculated. The correlation is calculated by weighted integration of distance, orientation, and topological relationship. The distance weight is set to 0.5, the orientation weight to 0.3, and the topological relationship weight to 0.2. The resulting spatial correlation matrix is ​​an n×m matrix, where n represents the number of anomalous areas, m represents the number of geographic features, and each element in the matrix represents the degree of spatial correlation between the corresponding anomalous area and the geographic feature. This matrix provides important spatial relationship characteristics for subsequent water quality anomaly tracing analysis.

[0063] In the feature fusion module, multi-feature fusion processing is the fundamental step in feature integration, aiming to unify features from different sources and with different dimensions into a common feature space. This process includes three specific implementation steps:

[0064] First, a normalization channel is established between the reflectance baseline value, time-domain variation characteristics, and spatial morphological relationship characteristics. The normalization process employs a modified z-score standardization method, calculating the mean and standard deviation for each feature dimension. For the reflectance baseline value, since its value ranges from 0 to 1, mean-variance normalization is used to convert the data into a standard distribution with a mean of 0 and a variance of 1. For time-domain variation characteristics, including fluctuation amplitude and duration indicators, min-max normalization is used to map the values ​​to the interval between 0 and 1. For spatial morphological relationship characteristics, such as regional compactness and boundary tortuosity, decimal scaling normalization is used, achieving normalization by shifting the decimal point. The normalization process for all features preserves the distribution characteristics of the original data while eliminating the influence of dimensional differences.

[0065] Secondly, dynamic weights are assigned based on the importance of each feature dimension in identifying water quality anomalies. The weight allocation employs an attention-based weighting method, automatically learning the importance of each feature through a neural network. Specifically, a three-layer attention network is first constructed, with normalized feature vectors as input and weight coefficients for each feature as output. Network training uses historical labeled data, and the weight parameters are optimized through backpropagation. The calculation of attention weights comprehensively considers the discriminative power and stability of features, assigning higher weights to features with high discriminative power and good stability. The final generated weight coefficients range from 0 to 1, and the sum of the weights of all features is 1, ensuring the comparability of the weighted features.

[0066] Finally, a multi-source feature cross-validation algorithm is employed to eliminate redundant information between features and enhance complementary features. This algorithm evaluates feature redundancy based on the correlation coefficient and mutual information between features. First, the Pearson correlation coefficient between all pairs of features is calculated. When the correlation coefficient exceeds 0.8, the two features are considered highly redundant, and one of them needs to be removed. The removal principle is to retain features that are more strongly correlated with the target variable. Simultaneously, the mutual information between each feature and the target variable is calculated, and features with a mutual information value greater than 0.1 are retained. For complementary features, feature combination techniques are used to generate new cross-features, such as multiplying time-domain variation features with spatial morphology features to generate spatiotemporal joint features. The final fused feature set reduces feature dimensionality while enhancing the feature representation ability.

[0067] Multidimensional feature vector generation builds upon feature fusion to construct the final feature representation used for classification and recognition. This process is achieved through three levels of analysis:

[0068] The first level involves using Principal Component Analysis (PCA) to compress and reduce the dimensionality of high-dimensional features. PCA first calculates the covariance matrix of the fused feature set, then solves for the eigenvalues ​​and eigenvectors of this matrix. The top k principal components with the largest eigenvalues ​​are retained; the value of k is determined using the cumulative variance contribution rate method, stopping when the cumulative contribution rate reaches 85%. Each principal component is a linear combination of the original features, and its coefficients are determined by the corresponding eigenvector. This process compresses the original high-dimensional feature space into a low-dimensional space while retaining most of the variation information of the original data. The dimensionality-reduced feature components not only reduce computational complexity but also eliminate correlations between features, improving the performance of subsequent classifiers.

[0069] The second level involves structuring the reduced-dimensional feature components according to spatiotemporal dimensions using feature recombination techniques. Feature recombination employs a semantic-based arrangement strategy, dividing features into three groups: spectral feature group, temporal feature group, and spatial feature group. The spectral feature group includes the reflectance baseline value and its derived features; the temporal feature group includes time-domain variation features such as fluctuation amplitude and duration period; and the spatial feature group includes regional morphology and spatial relationship features. Features within each group are arranged in descending order of importance, determined by feature weights. Feature type identifiers are also added during the recombination process to facilitate subsequent feature tracing and analysis. This structured arrangement not only preserves the physical meaning of the features but also facilitates subsequent feature selection and interpretation.

[0070] The third level involves generating multidimensional feature vectors with fixed dimensions. The feature vectors are uniformly set to 50 dimensions, with spectral features comprising 20 dimensions, temporal features 15 dimensions, and spatial features 15 dimensions. For insufficient features, feature duplication or interpolation methods are used to supplement the dimensions; for excessive features, feature selection methods are used to retain the most important features. Each feature value is standardized to ensure it falls within the range of 0 to 1. The final generated multidimensional feature vector simultaneously contains comprehensive information from spectral features, temporal variation features, and spatial relationship features, while maintaining relative independence and interpretability among the feature components. The feature vectors are stored as floating-point arrays with accompanying feature description information for easier processing by subsequent machine learning algorithms.

[0071] In the intelligent assessment module, multi-dimensional feature vector analysis is the core component, aiming to accurately identify the type of water quality anomalies through in-depth analysis of feature vectors. This process involves three levels of analysis and judgment:

[0072] First, the multidimensional feature vectors are input into a multi-level decision tree model for preliminary classification. The decision tree model is constructed using the C4.5 algorithm and contains 15 decision nodes across 3 levels. Each decision node corresponds to a feature dimension threshold, which is determined based on historical data statistical analysis. For example, for the reflectivity variation feature, the threshold is set to 1.5 times the standard deviation of the historical mean; for the spatial clustering feature, the threshold is set to 0.7 (range 0-1). The branching rules of the decision tree comprehensively consider the combination of spatiotemporal features. For example, if the conditions of duration greater than 3 days and spatial range greater than 100 square meters are met simultaneously, it is judged as a suspected algal bloom event. The preliminary classification results generate an initial probability estimate for each abnormal event type, with the estimate ranging from 0 to 1.

[0073] Secondly, a fuzzy logic inference mechanism is employed to perform joint probability analysis on the feature vectors. The fuzzy inference system contains 25 inference rules, each corresponding to a specific feature combination pattern. The system first fuzzifies the input features, converting precise feature values ​​into fuzzy membership degrees. The membership function adopts a triangular function form, with three fuzzy levels set for each feature: low, medium, and high. For example, the fuzzy classification threshold for the duration feature is: low (0-2 days), medium (2-5 days), and high (more than 5 days). The inference process uses the Mamdani fuzzy model, performing inference calculations through if-then rules in the rule base. Finally, the system outputs the membership probability for each event type, representing the degree of probability that the feature vector belongs to various types of water quality anomalies.

[0074] Finally, historical similarity matching is performed using a case-based reasoning engine. The case database stores 500 typical historical event cases, each containing a feature vector and the final confirmed event type. Similarity calculation employs a weighted Euclidean distance algorithm, with the weights of different features dynamically adjusted based on their discriminative power. Discriminative power is determined by calculating the variance of features across different event types; the larger the variance, the higher the weight. When a historical case with a similarity exceeding 0.85 is found, the nearest neighbor algorithm directly outputs the matching result; when the similarity is between 0.7 and 0.85, the K-nearest neighbor algorithm (K=5) is used for majority voting; when the similarity is below 0.7, no matching case is considered, and the judgment primarily relies on the results of the first two methods.

[0075] The results output phase is responsible for integrating the analysis results and generating the final credibility assessment. This process is achieved through three steps:

[0076] First, an evidence theory synthesis algorithm is employed to integrate multi-dimensional judgment conclusions. Decision tree classification results, fuzzy inference outputs, and case matching results are considered as three independent sources of evidence, which are then synthesized using Dempster-Shafer evidence theory. The credibility weight of each evidence source is dynamically adjusted based on its historical accuracy: decision tree weight is 0.3, fuzzy inference weight is 0.4, and case matching weight is 0.3. During the synthesis process, a basic probability assignment is calculated for each event type, and inconsistencies between evidence are eliminated through orthogonal summation formulas. Finally, a comprehensive probability value for each event type is generated, ranging from 0 to 1.

[0077] Secondly, the reliability of the final judgment result is calculated using a confidence propagation network. The confidence network employs a Bayesian network structure, containing 20 nodes and 35 edges. The network parameters are obtained through training on historical data, achieving an accuracy of over 85%. The confidence calculation considers the following factors: feature quality score (30% weight), method consistency (40% weight), and historical accuracy (30% weight). When the results of the three judgment methods are consistent, the confidence score is directly assigned to 0.9 or higher; when two methods are consistent, the confidence score is adjusted to 0.7-0.8; when all three methods are inconsistent, the confidence score drops below 0.6, requiring manual review.

[0078] Finally, a mapping table between event types and confidence levels was established. The mapping table uses a two-dimensional matrix format, with rows representing event types (including eight categories such as algal blooms, oil spills, wastewater discharge, and suspended sediment), and columns representing confidence levels (high, medium, and low). Each cell contains specific judgment rules; for example, when the overall probability of an algal bloom event is greater than 0.8 and the feature quality score is greater than 0.7, the confidence level is high; when the probability is between 0.6 and 0.8, the confidence level is medium; and when the probability is less than 0.6, the confidence level is low. The final output includes specific water quality anomaly event types and their corresponding confidence levels, as well as the main feature evidence supporting the conclusion.

[0079] The working principle of this invention is as follows: A data preprocessing module performs radiometric calibration, atmospheric correction, and spatiotemporal registration on multi-source remote sensing data to generate a reflectance data cube with consistent spatiotemporal dimensions. A time-series feature analysis module extracts pixel-level time series data and calculates temporal features such as fluctuation amplitude and duration. A spatial feature extraction module performs multi-scale segmentation and morphological analysis to identify anomalous areas and quantify their morphological characteristics and spatial relationships. A feature fusion module integrates spectral, spatiotemporal, and spatial features to generate multi-dimensional feature vectors. Finally, an intelligent judgment module combines multi-level decision trees, fuzzy inference, and case matching to perform fusion judgment, outputting the type and confidence level of water quality anomalies, thus achieving fully automated, high-precision intelligent identification and assessment of river and lake water quality anomalies.

[0080] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A river and lake intelligent monitoring and management system based on remote sensing and GIS, characterized in that, Comprise: A data preprocessing module for obtaining multi-temporal remote sensing data of a target water area, performing radiation calibration, atmospheric correction and space-time registration processing on the multi-temporal remote sensing data, and generating a reflectivity data cube with consistent space-time dimensions; A time series feature analysis module for extracting a reflectivity time series of each pixel point from the reflectivity data cube and calculating time domain variation characteristics, including reflectivity variation amplitude and duration period indicators; A spatial feature extraction module for identifying high reflectivity anomaly regions based on the reflectivity data cube, extracting morphological features of the high reflectivity anomaly regions and their spatial morphological relationship features with shorelines and sewage outlet geographic elements, specifically including: Performing multi-scale segmentation processing on the reflectivity data cube to generate image objects with spectral homogeneity; using dynamic threshold segmentation combined with local reflectivity statistical features and spatial context information to identify potential high reflectivity anomaly regions; optimizing the initial detection results through morphological filtering operations to eliminate discrete noise points and retain connected high reflectivity anomaly regions, forming accurate high reflectivity anomaly region boundaries and generating accurate high reflectivity anomaly regions; Based on the identified high reflectivity anomaly regions, calculating their morphological feature indicators including region compactness and boundary tortuosity; at the same time, through spatial buffer analysis and topological relationship judgment, quantifying the distance relationship and orientation relationship of high reflectivity anomaly regions with shorelines and sewage outlet geographic elements; finally, using spatial overlay analysis method, establishing the spatial association matrix of high reflectivity anomaly regions and geographic elements, and generating comprehensive spatial morphological relationship features; A feature fusion module for fusing the reflectivity reference value, time domain variation characteristics and spatial morphological relationship features of each pixel to generate a multi-dimensional feature vector for each pixel; An intelligent research and judgment module for analyzing and judging the multi-dimensional feature vector, outputting water quality anomaly event types and corresponding confidence levels. 2.The remote sensing and GIS-based intelligent monitoring and management system for rivers and lakes according to claim 1, characterized in that, The specific process of generating a reflectivity data cube with consistent space-time dimensions is: Performing radiation calibration processing on the obtained multi-temporal remote sensing data to convert the original digital quantization values to apparent reflectivity and eliminate differences in observation geometry; Using a radiative transfer model combined with real-time meteorological parameters to eliminate atmospheric absorption and scattering effects and obtain ground true reflectivity data; Through feature point matching and geometric precision correction techniques, unify the reflectivity data of different time phases to the same geographic coordinate system to generate a reflectivity data cube with consistent space-time dimensions. 3.The remote sensing and GIS-based intelligent monitoring and management system for rivers and lakes according to claim 1, characterized in that, The specific process of extracting a reflectivity time series of each pixel point includes: Obtain the reflectivity observation values of each pixel at all time phases from the reflectivity data cube to form an initial time series; use dynamic time warping algorithm to align the observation time points between different time phases to eliminate time offset caused by inconsistent satellite revisit periods; reconstruct the missing period data to generate a complete and time-aligned reflectivity time series. 4.The remote sensing and GIS-based intelligent monitoring and management system for rivers and lakes according to claim 1, characterized in that, The specific process of calculating time domain variation characteristics includes: Based on the reflectivity time series, the local change intensity of each time phase point is calculated by using a sliding window method to obtain a fluctuation amplitude index; meanwhile, the length of an abnormal period that continuously exceeds a threshold is detected to determine a duration cycle index; finally, the mutation point and trend characteristics of reflectivity change are extracted by combining time series waveform analysis to form comprehensive time domain change characteristics. 5.The remote sensing and GIS-based intelligent monitoring and management system for rivers and lakes according to claim 1, characterized in that, The fusing of the reflectivity reference value, time domain change characteristics and spatial form relationship characteristics of each pixel specifically includes: A normalization processing channel is established among the reflectivity reference value, time domain change characteristics and spatial form relationship characteristics to eliminate the dimensional differences of different characteristic dimensions; weights are assigned according to the importance degree of each characteristic dimension to water quality anomaly identification; and redundant information among the characteristics is eliminated and complementary characteristics are enhanced to form an optimized fusion feature set. 6.The remote sensing and GIS-based intelligent monitoring and management system for rivers and lakes according to claim 5, characterized in that, The generating of the multi-dimensional feature vector of each pixel specifically includes: Based on the fusion feature set, a principal component analysis method is used to compress and reduce the high-dimensional features to retain the feature components with the largest variance; then, the feature components after dimension reduction are structured and arranged according to the time and space dimensions through feature recombination technology; finally, a multi-dimensional feature vector with fixed dimensions is generated, which contains comprehensive information of spectral characteristics, time series change characteristics and spatial form relationship characteristics. 7.The remote sensing and GIS-based intelligent monitoring and management system for rivers and lakes according to claim 1, characterized in that, The analyzing and judging of the multi-dimensional feature vector specifically includes: The multi-dimensional feature vector is input into a decision tree model, and a preliminary classification is performed according to a preset time and space feature threshold; fuzzy logic reasoning is used to perform joint probability analysis on the characteristics in each dimension of the feature vector to calculate the possibility of belonging to each type of water quality anomaly event; similarity matching is performed between the current feature mode and historical typical event cases to complete comprehensive analysis and judgment. 8.The remote sensing and GIS-based intelligent monitoring and management system for rivers and lakes according to claim 1, characterized in that, The outputting of the type of water quality anomaly event and the corresponding confidence level specifically includes: Based on the judgment and analysis results, the multi-dimensional judgment conclusions are integrated to eliminate decision uncertainty; the reliability of the final determination result is calculated to generate a confidence level; a mapping relationship table of event type and confidence is established to output the judgment conclusion containing the specific type of water quality anomaly event and the corresponding confidence level.

Images