An image recognition-based water environment monitoring method and system

Abstract

This invention relates to the field of image recognition and discloses a method and system for monitoring aquatic environments based on image recognition. The method includes: data normalization of a multi-source image dataset to obtain a standardized image data sequence; feature deconstruction of the standardized image data sequence to obtain multi-scale feature data, and semantic category mapping to obtain environmental element identification result data; dynamic evolution analysis of the environmental element identification result data to obtain element-level spatiotemporal evolution trajectory data; network construction of the element-level spatiotemporal evolution trajectory data to obtain dynamic relationship graph data, and anomaly topology extraction of the dynamic relationship graph data to obtain environmental evolution feature data; dynamic risk assessment of the environmental evolution feature data to obtain comprehensive assessment data; and structured encapsulation of the comprehensive assessment data to obtain a comprehensive monitoring report. This invention can improve the monitoring efficiency of an image recognition-based aquatic environment monitoring method.

Description

Technical Field

[0001] This invention relates to the field of image recognition, and in particular to a method and system for monitoring aquatic environments based on image recognition. Background Technology

[0002] Currently, image recognition-based aquatic environment monitoring technology has been applied to some extent, but existing methods still have significant shortcomings in terms of the completeness of technical logic and the effectiveness of monitoring results. Existing technical solutions mostly focus on the identification and classification of environmental elements in single or static images, resulting in a clearly fragmented monitoring process. These methods typically only achieve the identification of "what" elements or simple location of "where," outputting discrete, fragmented, static snapshot information. Due to the lack of systematic tracking and analysis of the dynamic evolution of elements, and the failure to deeply explore the interrelationships between elements, it is impossible to accurately depict the dynamic development patterns and systemic risks of environmental events such as pollution diffusion and algal bloom migration. Monitoring results often lag behind the actual situation, resulting in insufficient timeliness and foresight in early warning.

[0003] The limitations of existing technologies are also reflected in the broken chain from data to decision-making. Most methods stop at providing identification results or simple statistical indicators, failing to construct a complete, closed-loop analytical system from raw image data to risk assessment conclusions. Due to the lack of in-depth mining of the dynamic behavior patterns and networked relationships of elements, existing technologies are unable to support the accurate assessment of complex and interconnected environmental risks. The monitoring reports they output are usually just a list or simple summary of raw identification data, lacking in-depth explanation and structured presentation of the causes, evolution trends, and scope of impact of risks. As a result, the final environmental management decisions still heavily rely on human experience and judgment, and the level of intelligence and automation needs to be improved, failing to meet the urgent needs of accurate monitoring and efficient management of modern aquatic environments. Therefore, how to improve the monitoring efficiency of aquatic environments has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a method and system for monitoring aquatic environments based on image recognition, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for monitoring aquatic environments based on image recognition, comprising:

[0006] S1. Perform data normalization on the multi-source image dataset of the target water area to obtain a standardized image data sequence of the target water area;

[0007] S2. Perform feature deconstruction on the standardized image data sequence to obtain multi-scale feature data of the standardized image data sequence, and perform semantic category mapping on the multi-scale feature data to obtain environmental element identification result data of the target water area;

[0008] S3. Perform dynamic evolution analysis on the environmental element identification result data to obtain element-level spatiotemporal evolution trajectory data of the target water area;

[0009] S4. The element-level spatiotemporal evolution trajectory data is networked to obtain dynamic relationship map data of the element-level spatiotemporal evolution trajectory data, and the abnormal topology structure of the dynamic relationship map data is extracted to obtain the environmental evolution characteristic data of the target water area.

[0010] S5. Based on a preset environmental risk assessment knowledge base, perform dynamic risk assessment on the environmental evolution characteristic data to obtain comprehensive assessment data of the environmental evolution characteristic data.

[0011] S6. The comprehensive assessment data is structured and packaged to obtain a comprehensive monitoring report of the target water area.

[0012] In a preferred embodiment, the step of data normalizing the multi-source image dataset of the target water area to obtain a standardized image data sequence of the target water area includes:

[0013] Target regions are extracted from the multi-source image dataset of the target water area to obtain the effective image dataset of the target water area;

[0014] The effective image dataset is optimized for image sharpness to obtain an enhanced image dataset of the target water area;

[0015] Based on a predefined spatial reference datum, the enhanced image dataset is scale-normalized to obtain a set of corrected image data for the target water area;

[0016] The corrected image dataset is normalized to obtain a standardized image data sequence of the target water area.

[0017] In a preferred embodiment, the step of deconstructing the standardized image data sequence to obtain multi-scale feature data of the standardized image data sequence, and performing semantic category mapping on the multi-scale feature data to obtain environmental element identification result data of the target water area, includes:

[0018] The standardized image data sequence is deconstructed to obtain a multi-scale feature representation of the standardized image data sequence;

[0019] Cross-scale feature interaction is performed on the multi-scale feature representation to obtain the fused feature map of the standardized image data sequence;

[0020] Based on the fused feature map, the pixel features in the fused feature map are subjected to category assignment mapping to obtain the pixel category probability distribution of the fused feature map;

[0021] Spatial consistency optimization is performed on the pixel category probability distribution to obtain the environmental element identification result data of the target water area.

[0022] In a preferred embodiment, the step of performing dynamic evolution analysis on the environmental element identification result data to obtain element-level spatiotemporal evolution trajectory data of the target water area includes:

[0023] Perform cross-time frame instance matching on the environmental element identification result data to obtain the associated element instance set of the environmental element identification result data;

[0024] Based on the set of associated element instances, the spatiotemporal location sequence of the element instances in the set of associated element instances is reconstructed to obtain the smooth spatiotemporal trajectory of the element instances.

[0025] The smoothed spatiotemporal trajectory is subjected to motion feature parameter quantization to obtain the temporal motion feature data of the element instance;

[0026] Evolutionary pattern analysis is performed on the temporal motion characteristic data to obtain element-level spatiotemporal evolution trajectory data of the target water area.

[0027] In a preferred embodiment, the step of performing evolution pattern analysis on the temporal motion feature data to obtain element-level spatiotemporal evolution trajectory data of the target water area includes:

[0028] The temporal motion feature data is subjected to change trend feature extraction to obtain a multi-dimensional change trend description of the temporal motion feature data;

[0029] Based on the historical evolution pattern library of the target water area, the evolution pattern matching of the multi-dimensional change trend description is performed to obtain the evolution pattern matching result of the time-series motion feature data.

[0030] Based on the preset confidence judgment rules, the confidence of the evolution pattern matching results is evaluated to obtain the evolution pattern results of the temporal motion feature data;

[0031] The evolution model results are sequence-tuned to obtain element-level spatiotemporal evolution trajectory data of the target water area.

[0032] In a preferred embodiment, the step of constructing a network from the element-level spatiotemporal evolution trajectory data to obtain dynamic relationship graph data of the element-level spatiotemporal evolution trajectory data, and extracting abnormal topology from the dynamic relationship graph data to obtain environmental evolution characteristic data of the target water area, includes:

[0033] The feature-level spatiotemporal evolution trajectory data is converted into graph nodes to obtain the feature nodes of the feature-level spatiotemporal evolution trajectory data.

[0034] Based on the spatiotemporal proximity and motion correlation in the feature-level spatiotemporal evolution trajectory data, the feature nodes are associated with edges to obtain the basic association graph of the feature-level spatiotemporal evolution trajectory data.

[0035] The basic association graph is serialized into a time-series dynamic graph to obtain the dynamic relationship map data of the element-level spatiotemporal evolution trajectory data;

[0036] Based on the historical topological baseline of the target water area, the topological deviation of the dynamic relationship map data is extracted to obtain the topological deviation of the dynamic relationship map data.

[0037] Anomaly interval detection is performed on the topological deviation, and the intervals with deviations exceeding a preset threshold and the corresponding dynamic relationship graph data are structurally aggregated to obtain the environmental evolution characteristic data of the target water area.

[0038] In a preferred embodiment, the step of detecting abnormal intervals in the topological deviation and structurally aggregating the intervals where the deviation exceeds a preset threshold and the corresponding dynamic relationship graph data to obtain the environmental evolution characteristic data of the target water area includes:

[0039] Anomaly interval detection is performed on the topological deviation to obtain a sequence of candidate anomaly intervals whose deviation exceeds a preset threshold;

[0040] The candidate abnormal interval sequence is subjected to interval merging processing to obtain the abnormal interval of the topological deviation.

[0041] Based on the abnormal intervals, interval data extraction is performed on the dynamic relationship graph data to obtain the abnormal time period sub-graph set of the dynamic relationship graph data;

[0042] The abnormal time period sub-map set and the abnormal interval are structurally aggregated to obtain the environmental evolution characteristic data of the target water area.

[0043] In a preferred embodiment, the step of dynamically assessing the environmental evolution characteristic data based on a preset environmental risk assessment knowledge base to obtain comprehensive assessment data of the environmental evolution characteristic data includes:

[0044] Based on the scenario pattern set in the preset environmental risk assessment knowledge base, the environmental evolution feature data is pattern matched to obtain the initial scenario matching result of the environmental evolution feature data.

[0045] Based on the initial scenario matching results, the environmental risk assessment knowledge base is subjected to associated data extraction to obtain the potential risk field data of the environmental evolution characteristic data.

[0046] A multidimensional risk index assessment is performed on the potential risk field data to obtain the fusion risk index of the environmental evolution characteristic data;

[0047] Based on the threshold ranges and level determination rules in the environmental risk assessment knowledge base, the fusion risk index is compared with the threshold to obtain the risk level of the fusion risk index, and the data is integrated to obtain the comprehensive assessment data of the environmental evolution characteristic data.

[0048] In a preferred embodiment, the step of structuring and encapsulating the comprehensive assessment data to obtain a comprehensive monitoring report for the target water area includes:

[0049] The core conclusions are extracted from the comprehensive evaluation data to obtain the report summary data of the comprehensive evaluation data;

[0050] Evidence chains are established by linking the report summary data, the environmental element identification result data, the element-level spatiotemporal evolution trajectory data, and the environmental evolution characteristic data to obtain a traceable data association network for the comprehensive evaluation data.

[0051] Visualize and render the traceable data association network to obtain a draft evaluation report of the comprehensive evaluation data;

[0052] Based on a preset report template, the draft assessment report is structured and organized to obtain a comprehensive monitoring report for the target water area.

[0053] To address the aforementioned problems, the present invention also provides an image recognition-based aquatic environment monitoring system, the system comprising:

[0054] The data normalization module is used to normalize the multi-source image dataset of the target water area to obtain a standardized image data sequence of the target water area;

[0055] The feature recognition module is used to deconstruct the standardized image data sequence to obtain multi-scale feature data of the standardized image data sequence, and to perform semantic category mapping on the multi-scale feature data to obtain environmental feature recognition result data of the target water area.

[0056] The evolution analysis module is used to perform dynamic evolution analysis on the environmental element identification result data to obtain element-level spatiotemporal evolution trajectory data of the target water area;

[0057] The relationship mining module is used to construct a network from the element-level spatiotemporal evolution trajectory data to obtain dynamic relationship graph data of the element-level spatiotemporal evolution trajectory data, and to extract abnormal topology from the dynamic relationship graph data to obtain environmental evolution feature data of the target water area.

[0058] The risk assessment module is used to perform dynamic risk assessment on the environmental evolution characteristic data based on a preset environmental risk assessment knowledge base, and obtain comprehensive assessment data of the environmental evolution characteristic data.

[0059] The report generation module is used to encapsulate the comprehensive assessment data in a structured manner to obtain a comprehensive monitoring report of the target water area.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] 1. This method achieves a leap in monitoring dimensions, moving from static element identification to dynamic system analysis. By performing temporal correlation and trajectory reconstruction on identified environmental elements, the method can accurately depict the dynamic behavioral trajectory of each element and further explore the network of relationships between multiple elements over time, thereby systematically capturing potential, interconnected, and abnormal evolution patterns within the environmental system. This allows monitoring results to go beyond simply identifying "what is there" and "where it is," profoundly revealing the inherent laws governing "how the environment changes" and "how different elements interact," significantly enhancing the depth of monitoring and the ability to understand complex environmental processes.

[0062] 2. A complete and automated analysis loop, from raw data to intelligent decision-making, has been constructed. The method intelligently matches and integrates the evolutionary characteristics obtained from in-depth analysis with specific environmental risk scenarios, impact projection models, and decision-making rules through a pre-set knowledge base, ultimately generating structured risk assessment conclusions and traceable monitoring reports. This process automatically transforms image data into high-value decision-making information containing risk levels, impact predictions, and remediation recommendations, achieving full-process intelligentization of monitoring, analysis, assessment, and report generation, greatly improving the timeliness, accuracy, and action support capabilities of environmental supervision. Attached Figure Description

[0063] Figure 1 This is a flowchart illustrating an image recognition-based method for monitoring aquatic environments according to an embodiment of the present invention.

[0064] Figure 2 A functional block diagram of an image recognition-based aquatic environment monitoring system provided in an embodiment of the present invention;

[0065] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0066] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0067] This application provides an image recognition-based method for monitoring aquatic environments. The executing entity of this image recognition-based aquatic environment monitoring method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the image recognition-based aquatic environment monitoring method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0068] Reference Figure 1 The diagram shown is a flowchart illustrating an image recognition-based aquatic environment monitoring method according to an embodiment of the present invention. In this embodiment, the image recognition-based aquatic environment monitoring method includes:

[0069] S1. Perform data normalization on the multi-source image dataset of the target water area to obtain a standardized image data sequence of the target water area;

[0070] In this embodiment of the invention, the step of data normalizing the multi-source image dataset of the target water area to obtain a standardized image data sequence of the target water area includes:

[0071] Target regions are extracted from the multi-source image dataset of the target water area to obtain the effective image dataset of the target water area;

[0072] The effective image dataset is optimized for image sharpness to obtain an enhanced image dataset of the target water area;

[0073] Based on a predefined spatial reference datum, the enhanced image dataset is scale-normalized to obtain the corrected image dataset of the target water area;

[0074] The corrected image dataset is normalized to obtain a standardized image data sequence of the target water area.

[0075] The target region is extracted from the multi-source image dataset of the target water area using a threshold segmentation method based on preset color features or texture features. A set of images containing typical target water areas is selected from the multi-source image dataset as training samples. The target water area and non-target background area in the images are manually labeled. The distribution range of pixel values ​​or the numerical range of local texture feature descriptors of the target water area in a specific color space channel in the training samples are statistically analyzed as preset threshold standards. Then, this threshold standard is applied to each image in the multi-source image dataset. Pixel areas that meet the target water feature threshold standard are retained as foreground, while pixel areas that do not meet the threshold standard are set as background or removed, thus obtaining an effective image dataset containing only pixels of the target water area.

[0076] An image sharpness optimization method based on histogram equalization is used to process the effective image dataset. Each image in the effective image dataset is read and transformed from its original color image representation to a color space that separates luminance and color information. The luminance component image is extracted, and the pixel grayscale value statistical histogram of the luminance component image is calculated. This histogram reflects the distribution of the number of pixels at each grayscale level. Based on the histogram, a cumulative distribution function is calculated, and the original grayscale value of each pixel in the luminance component image is mapped and transformed using the cumulative distribution function to generate new grayscale values. This makes the grayscale histogram of the output image as uniformly distributed as possible throughout the dynamic range. Then, the processed luminance component and the original color component are merged back into a color image. This process improves the overall contrast and detail visibility of the image, and finally, an enhanced image dataset of the target water area is obtained.

[0077] Based on a scale normalization method using georegistration and resampling, the augmented image dataset is processed. A unified spatial reference datum is predefined, which includes a clear geodetic coordinate system, map projection method, and standard spatial resolution. For each image in the augmented image dataset, the known coordinate control points corresponding to its image corners or feature points in the actual geographic space are identified. Using the correspondence between these control point coordinates and image pixel coordinates, a spatial transformation relationship is calculated through a polynomial transformation model. This transformation relationship is applied to perform geometric correction on the original image, aligning its pixel coordinates with the predefined spatial reference datum. Subsequently, the corrected image is resampled according to the standard spatial resolution to ensure that all output images have the same pixel size and geographic coverage accuracy, ultimately yielding the corrected image dataset of the target water area.

[0078] The corrected image dataset is processed using a data normalization method based on fixed-size cropping and uniform format conversion. A standard image size specification is set, which defines the number of rows and columns of pixels in the image. For each image in the corrected image dataset, its size is checked to see if it is consistent with the standard specification. If not, bilinear interpolation is used to scale the image to the standard size. During the scaling process, the new pixel value is calculated based on the position of the target pixel in the original image, using the gray values ​​of its four neighboring original pixels according to distance weight. After size unification, the color mode of all images is converted to a preset uniform format. For example, images containing an alpha channel are converted to three-channel images without an alpha channel, or images with different bit depths are uniformly converted to eight-bit bit depth images. Finally, the images are sorted and stored according to the time sequence of image acquisition to form a standardized image data sequence of the target water area.

[0079] This step improves data specificity by extracting effective images containing only the target water area through threshold segmentation based on color or texture features from multi-source images. Histogram equalization is then applied to optimize the contrast and detail of the effective images, resulting in clearer enhanced images. Next, the enhanced images are georegistered and resampled according to a preset spatial reference, achieving uniform correction in scale and coordinates for all images. Finally, the corrected images are standardized to standard sizes and formats through fixed-size cropping and uniform format conversion, and then sorted chronologically to form a standardized image sequence that can be directly used for subsequent analysis. The entire process progressively improves the quality, consistency, and usability of the image data.

[0080] S2. Perform feature deconstruction on the standardized image data sequence to obtain multi-scale feature data of the standardized image data sequence, and perform semantic category mapping on the multi-scale feature data to obtain environmental element identification result data of the target water area;

[0081] In this embodiment of the invention, the step of deconstructing the standardized image data sequence to obtain multi-scale feature data of the standardized image data sequence, and performing semantic category mapping on the multi-scale feature data to obtain environmental element identification result data of the target water area, includes:

[0082] The standardized image data sequence is deconstructed to obtain a multi-scale feature representation of the standardized image data sequence;

[0083] Cross-scale feature interaction is performed on the multi-scale feature representation to obtain the fused feature map of the standardized image data sequence;

[0084] Based on the fused feature map, the pixel features in the fused feature map are subjected to category assignment mapping to obtain the pixel category probability distribution of the fused feature map;

[0085] Spatial consistency optimization is performed on the pixel category probability distribution to obtain the environmental element identification result data of the target water area.

[0086] This paper employs a feature decomposition method based on Gaussian pyramid decomposition to process standardized image data sequences and obtain multi-scale feature representations. A set of scale factor sequences defines the proportion of image downscaling. Each image in the standardized image data sequence is used as the bottom layer of the pyramid, i.e., the original scale image. A Gaussian kernel with a preset standard deviation is used to perform convolutional smoothing on the image at the current scale to suppress noise. The smoothed image is then downsampled, taking one point every other pixel to obtain an image with half the scale, which becomes the next pyramid layer. The same Gaussian smoothing and downsampling operations are repeated on the latest pyramid layer image until the preset number of scale layers is reached. Finally, the original image is decomposed into a series of images with progressively decreasing resolution but progressively generalized content; this image sequence, from fine to coarse, is the multi-scale feature representation of the standardized image data sequence.

[0087] A cross-scale feature interaction method based on Laplacian pyramid reconstruction processes multi-scale feature representations to obtain a fused feature map. Starting from the top layer of the Gaussian pyramid (the coarsest scale), this image is used as the current working layer. This working layer image is upsampled by inserting a new pixel (linearly interpolated from adjacent pixels) between every two pixels, doubling its size. The image of the next layer in the Gaussian pyramid is subtracted from this upsampled image to obtain a Laplacian pyramid layer containing detailed differences between two adjacent scales. All Laplacian pyramid layers from different scales are upsampled to the size of the original image and summed pixel-by-pixel. Finally, this sum of details is added pixel-by-pixel to the original image at the bottom layer of the Gaussian pyramid. This synthesized image, fusing information from coarse to fine scales, is the fused feature map of the normalized image data sequence.

[0088] A class assignment mapping method based on a fully connected neural network classifier is used to process the fused feature map to obtain the pixel class probability distribution. Specifically, a label set containing all the environmental element categories to be identified in the target water area is predefined. A feature vector is formed by extracting each pixel location and all pixel values ​​within its neighborhood window from the fused feature map. This feature vector is then input into a pre-trained fully connected neural network classifier. The number of neurons in the input layer of this classifier equals the dimension of the feature vector, and the number of neurons in the output layer equals the total number of environmental element categories. After entering the network from the input layer, the feature vector undergoes linear transformations and nonlinear activation functions in several hidden layers, ultimately producing a set of values ​​in the output layer. These output values ​​are normalized so that their sum is a fixed value. After processing, each output value represents the probability that the pixel belongs to the corresponding environmental element category. This feature extraction and classifier processing process is repeated for each pixel in the fused feature map, ultimately resulting in a map with the same size as the fused feature map but with a set of class probability values ​​for each pixel location. This map represents the pixel class probability distribution of the fused feature map.

[0089] The pixel category probability distribution is processed using a spatial consistency optimization method based on conditional random fields to obtain environmental element identification data. Specifically, each pixel in the pixel category probability distribution is considered a node, and a connection relationship between each node and its neighboring nodes is defined to form a graph model. Two energy functions are defined. The first function measures the consistency between the category probability of a single node and the final assigned label, typically using the negative logarithm of the probability value corresponding to that label in the pixel category probability distribution. The second function measures the smoothness of label assignment between adjacent nodes; the energy is zero when adjacent nodes are assigned the same label, and a positive value inversely proportional to the similarity of their pixel features when they are assigned different labels. The graph cut algorithm minimizes the total energy of all nodes in the entire image. This algorithm iteratively finds a label assignment scheme that minimizes the total energy defined above. The final label assignment scheme is the environmental element category to which each pixel belongs, and the two-dimensional graph of this category identification is the environmental element identification data of the target water area.

[0090] This step involves performing Gaussian pyramid decomposition on a standardized image sequence to obtain multi-scale feature representations ranging from fine to coarse. A Laplacian pyramid reconstruction method is then applied to fuse detailed information from each scale, generating a fused feature map containing rich contextual features. A fully connected neural network classifier is used to process the fused feature map pixel-by-pixel to obtain the probability distribution of each location belonging to various environmental elements. Finally, a conditional random field model is used to optimize the spatial consistency of the probability distribution, resulting in accurate and coherent environmental element identification. The entire process effectively extracts and fuses multi-scale features while ensuring the spatial reasonableness and accuracy of the identification results.

[0091] S3. Perform dynamic evolution analysis on the environmental element identification result data to obtain element-level spatiotemporal evolution trajectory data of the target water area;

[0092] In this embodiment of the invention, the step of performing dynamic evolution analysis on the environmental element identification result data to obtain element-level spatiotemporal evolution trajectory data of the target water area includes:

[0093] Perform cross-time frame instance matching on the environmental element identification result data to obtain the associated element instance set of the environmental element identification result data;

[0094] Based on the set of associated element instances, the spatiotemporal location sequence of the element instances in the set of associated element instances is reconstructed to obtain the smooth spatiotemporal trajectory of the element instances.

[0095] The smoothed spatiotemporal trajectory is subjected to motion feature parameter quantization to obtain the temporal motion feature data of the element instance;

[0096] Evolutionary pattern analysis is performed on the temporal motion characteristic data to obtain element-level spatiotemporal evolution trajectory data of the target water area.

[0097] The temporal motion feature data is subjected to change trend feature extraction to obtain a multi-dimensional change trend description of the temporal motion feature data;

[0098] Based on the historical evolution pattern library of the target water area, the evolution pattern matching of the multi-dimensional change trend description is performed to obtain the evolution pattern matching result of the time-series motion feature data.

[0099] Based on the preset confidence judgment rules, the confidence of the evolution pattern matching results is evaluated to obtain the evolution pattern results of the temporal motion feature data;

[0100] The evolution model results are sequence-tuned to obtain element-level spatiotemporal evolution trajectory data of the target water area.

[0101] A cross-time-frame instance matching method based on the intersection-union ratio (IU) threshold is used to construct a set of associated element instances from environmental element identification data. Two consecutive frames of images from the time-sorted environmental element identification data are read sequentially. For each identified independent environmental element instance in the preceding frame, the overlap area between its pixel region and the pixel region of each environmental element instance in the following frame is calculated. The overlap area is divided by the total area after merging the two instance regions to obtain an IU value. If this value exceeds a preset IU threshold, the two element instances located in adjacent time frames are determined to be the same element at different times, and an association is established. All adjacent frame pairs are processed sequentially, and cross-frame association chains are merged. Finally, all instances determined to be the same element in the entire time series are grouped into a single set, and the entire set constitutes the associated element instance set.

[0102] A trajectory reconstruction method based on sliding window mean filtering is used to process the associated feature instance set to obtain a smoothed spatiotemporal trajectory. A set of feature instances is selected from the associated feature instance set. The center pixel coordinates of each instance in the corresponding time frame image are extracted. The pixel coordinates are converted into real geographic coordinates according to the image scale normalization information, and the timestamp corresponding to each coordinate point is recorded to form an original spatiotemporal location sequence. A sliding window of fixed time length is defined. Starting from the sequence start time, the arithmetic mean of the latitude and longitude coordinates of all locations within the window is calculated, and this average is used as the smoothed position of the window center time. The window is then slid forward by a fixed time step, and the above averaging operation is repeated until the window covers the end time of the sequence. All window center times and their corresponding smoothed position points are connected in chronological order to form a new position sequence, which is the smoothed spatiotemporal trajectory of the feature instance.

[0103] Motion feature parameters of a smoothed spatiotemporal trajectory are quantified using difference-based calculations to obtain temporal motion feature data. Specifically, each location point in the smoothed spatiotemporal trajectory, arranged chronologically, and its corresponding timestamp are obtained. The latitude and longitude coordinate differences between adjacent location points are calculated. The latitude coordinate difference is divided by the corresponding time interval to obtain the average eastward velocity component for that time interval. The latitude coordinate difference is divided by the corresponding time interval to obtain the average northward velocity component for that time interval. The square root of the sum of the squares of the eastward and northward velocity components is calculated to obtain the average resultant velocity for that time interval. The difference in resultant velocities between adjacent time intervals is calculated and divided by the time interval to obtain the average acceleration. The change in the azimuth angle of the line connecting adjacent location points is calculated and divided by the time interval to obtain the average turning angular velocity. These calculated instantaneous velocities, instantaneous accelerations, and instantaneous turning angular velocities are arranged and stored chronologically to constitute the temporal motion feature data of this element instance.

[0104] Evolutionary pattern analysis based on polynomial fitting is applied to temporal motion characteristic data to obtain element-level spatiotemporal evolution trajectory data. Specifically, a sequence of characteristic physical quantities, such as the resultant velocity sequence, is selected from the temporal motion characteristic data. The least squares method is used to fit this sequence into a polynomial function of a specified order. The coefficients of this polynomial function reflect the main trend pattern of the characteristic physical quantity's change over time. The same polynomial fitting is then applied to various characteristic physical quantity sequences in the temporal motion characteristic data. All fitted polynomial functions are used as mathematical descriptions of the motion evolution pattern of the element instance. This series of mathematical descriptions, along with the smoothed spatiotemporal trajectory of the element instance and the original set of associated instances, are encapsulated and integrated to ultimately form element-level spatiotemporal evolution trajectory data that can completely characterize the change law of the motion state of a single environmental element within its observation period.

[0105] Based on the moving average comparison-based trend feature extraction method, this method processes time-series motion feature data to obtain a multi-dimensional trend description. Specifically, it selects a feature physical quantity sequence from the time-series motion feature data and calculates two moving averages of different time lengths for this sequence: a short-term moving average and a long-term moving average. The numerical relationship between the short-term and long-term moving averages is compared at each time point. If the short-term moving average crosses below the long-term moving average and remains above it, the period is marked as an upward trend. If the short-term moving average crosses above the long-term moving average and remains below it, the period is marked as a downward trend. If the two moving averages are intertwined and the difference remains within a small range, the period is marked as an oscillating trend. This calculation and marking process is repeated for each feature physical quantity sequence in the time-series motion feature data, ultimately resulting in a set of trend labels describing the direction and stability of motion changes from different physical dimensions. This set constitutes the multi-dimensional trend description of the time-series motion feature data.

[0106] Edit distance-based pattern matching is performed on multi-dimensional trend descriptions to obtain evolutionary pattern matching results. Specifically, a historical evolutionary pattern library is pre-constructed, storing various standard change pattern sequences derived from historical data. Each standard change pattern sequence consists of a series of trend labels arranged chronologically. The edit distance between the sequence to be matched and each standard pattern sequence in the historical evolutionary pattern library is calculated. The standard pattern sequence or sequences with the smallest edit distance are selected as candidate matching patterns. These selected standard patterns and their corresponding edit distance information together constitute the evolutionary pattern matching results of the temporal motion feature data.

[0107] The evolution pattern matching results are evaluated using a confidence threshold to obtain the evolution pattern results. Specifically, a confidence threshold rule is preset, which stipulates that when the edit distance corresponding to the best matching pattern in the evolution pattern matching results is less than an absolute threshold, the match is considered a high-confidence match. Simultaneously, it stipulates that when the difference between the edit distance of the best matching pattern and the edit distance of the second-best matching pattern is greater than a relative threshold, the match result is considered to have high discriminative confidence. The edit distance value of the best match in the current evolution pattern matching results is calculated. The difference between the edit distance of the best match and the edit distance of the second-best match is calculated. These two values ​​are compared with the preset absolute threshold and relative threshold, respectively. If both comparison conditions are met, the best matching pattern is determined to be the final confirmed evolution pattern. If the conditions are not met, the matching confidence is insufficient, and an uncertainty label is given. This determination, together with the matching pattern on which it is based, constitutes the evolution pattern result of the temporal motion feature data.

[0108] The evolution model results undergo temporal logic tuning to obtain element-level spatiotemporal evolution trajectory data. Specifically, the evolution model results are checked to determine if they are high-confidence definitive matches. For matches determined to be high-confidence, the corresponding evolution model label sequence is aligned and calibrated with the original temporal motion feature data on the timeline. The calibrated or padded evolution model label sequence is then fused and encapsulated with the smoothed spatiotemporal trajectory and temporal motion feature data generated in previous steps. During fusion, it is ensured that the evolution model labels, motion feature values, and spatial location points correspond completely and consistently in timestamps. The resulting multidimensional dataset, containing spatial trajectories, motion parameters, and qualitative evolution models, constitutes the element-level spatiotemporal evolution trajectory data of the target water area.

[0109] This step obtains a set of associated element instances by performing cross-frame instance matching on the environmental element identification results, ensuring continuous tracking of the same element over time. A sliding window mean filter is used to smooth the instance position sequence, resulting in a more reasonable spatiotemporal trajectory. Differential calculations are used to extract temporal motion feature data such as velocity and acceleration from the smoothed trajectory. Polynomial fitting and moving average comparisons are used to obtain a mathematical model and multi-dimensional qualitative description of the motion trend. Edit distance matching and confidence assessment are combined to identify the current motion evolution pattern from the historical pattern library. Finally, temporal logic tuning fuses the spatial trajectory motion features and evolution patterns to form complete element-level spatiotemporal evolution trajectory data. This process achieves a progressively deeper analysis from discrete identification to continuous trajectory and then to patterned description.

[0110] S4. The element-level spatiotemporal evolution trajectory data is networked to obtain dynamic relationship map data of the element-level spatiotemporal evolution trajectory data, and the abnormal topology structure of the dynamic relationship map data is extracted to obtain the environmental evolution characteristic data of the target water area.

[0111] In this embodiment of the invention, the step of constructing a network from the element-level spatiotemporal evolution trajectory data to obtain dynamic relationship graph data of the element-level spatiotemporal evolution trajectory data, and extracting abnormal topology from the dynamic relationship graph data to obtain environmental evolution characteristic data of the target water area, includes:

[0112] The feature-level spatiotemporal evolution trajectory data is converted into graph nodes to obtain the feature nodes of the feature-level spatiotemporal evolution trajectory data.

[0113] Based on the spatiotemporal proximity and motion correlation in the feature-level spatiotemporal evolution trajectory data, the feature nodes are associated with edges to obtain the basic association graph of the feature-level spatiotemporal evolution trajectory data.

[0114] The basic association graph is serialized into a time-series dynamic graph to obtain the dynamic relationship map data of the element-level spatiotemporal evolution trajectory data;

[0115] Based on the historical topological baseline of the target water area, topological deviation is extracted from the dynamic relationship map data to obtain the topological deviation of the dynamic relationship map data. The formula for calculating the topological deviation is as follows:

[0116] ;

[0117] In the formula, This indicates that the dynamic relationship graph data is in Topological deviation at time t, This indicates that the dynamic relationship graph is in Network density at any given time This represents the mean network density of the historical topology baseline. This represents the network density variance of the historical topology baseline. This indicates that the dynamic relationship graph is in The average clustering coefficient at time t, This represents the mean of the average clustering coefficients of the historical topological baseline. This represents the average variance of the clustering coefficients of the historical topological baseline;

[0118] Anomaly interval detection is performed on the topological deviation, and the intervals with deviations exceeding a preset threshold and the corresponding dynamic relationship graph data are structurally aggregated to obtain the environmental evolution characteristic data of the target water area.

[0119] Anomaly interval detection is performed on the topological deviation to obtain a sequence of candidate anomaly intervals whose deviation exceeds a preset threshold;

[0120] The candidate abnormal interval sequence is subjected to interval merging processing to obtain the abnormal interval of the topological deviation.

[0121] Based on the abnormal intervals, interval data extraction is performed on the dynamic relationship graph data to obtain the abnormal time period sub-graph set of the dynamic relationship graph data;

[0122] The abnormal time period sub-map set and the abnormal interval are structurally aggregated to obtain the environmental evolution characteristic data of the target water area.

[0123] Each independent spatiotemporal evolution element in the element-level spatiotemporal evolution trajectory data is treated as an independent graph node. Each element is assigned a unique node identifier containing its basic spatiotemporal attribute information. The attribute information of the node is entered and the node entity is created based on the spatiotemporal evolution trajectory of the element. The graph nodeization process of the element-level spatiotemporal evolution trajectory data is completed, and the element nodes of the element-level spatiotemporal evolution trajectory data are obtained.

[0124] Spatiotemporal proximity is determined based on the spatiotemporal coordinates of each element in the element-level spatiotemporal evolution trajectory data. When the spatial distance between the elements corresponding to two element nodes in the same time dimension is less than a preset spatial threshold, and the time interval in the same spatial dimension is less than a preset time threshold, the two are determined to have spatiotemporal proximity. At the same time, motion correlation is determined based on the motion direction and motion rate of each element. When the angle between the motion directions of the elements corresponding to two element nodes is less than a preset angle threshold, and the difference in motion rate is within a preset rate difference range, the two are determined to have motion correlation. For element nodes that have both spatiotemporal proximity and motion correlation, a connection edge is established between them. Each connection edge is assigned attribute information containing the criteria for proximity and correlation determination, thus completing the construction of the connection edge of the element nodes and obtaining the basic correlation graph of the element-level spatiotemporal evolution trajectory data. The basic correlation graph is processed by time slicing according to the time axis sequence. The time interval of the time slice is set according to the time acquisition frequency of the element-level spatiotemporal evolution trajectory data. Each subgraph corresponding to the time slice is assigned a corresponding timestamp information. All subgraphs with timestamps are serialized and arranged in the order of the time axis to form a continuous time-series dynamic graph sequence. This completes the time-series dynamic graph serialization of the basic correlation graph and obtains the dynamic relationship map data of the element-level spatiotemporal evolution trajectory data.

[0125] The historical topological baseline of the target water area is retrieved. This historical topological baseline is formed by long-term statistical processing of the topological features of the correlation graph constructed from spatiotemporal evolution trajectory data of the same type in the target water area. It includes benchmark values ​​of fixed topological features such as the number of topological nodes, the number of associated edges, and the topological connection density. The dynamic relationship graph data is then extracted. For each time window subgraph corresponding to a given time, the actual number of associated edges in the subgraph is compared to the maximum possible number of associated edges that the element nodes in the subgraph can form. The ratio of the actual number of associated edges to the maximum possible number of associated edges is then calculated to obtain the dynamic relationship graph. Network density at any time Extract the network density values ​​corresponding to all historical time window subgraphs in the historical topology baseline, and calculate the arithmetic mean of all extracted network density values ​​to obtain the mean network density of the historical topology baseline. Based on all extracted network density values ​​and Calculate the density value of each network in turn. The network density variance of the historical topology baseline is obtained by squared differences and then taking the arithmetic mean of all squared differences. .

[0126] Extracting dynamic relationship graph data For each time window subgraph corresponding to a given moment, the actual number of associated edges between neighboring nodes and the maximum possible number of associated edges that neighboring nodes can form for each feature node in the subgraph is counted. The ratio of this ratio to the clustering coefficient of a single node is then calculated. The arithmetic mean of the clustering coefficients of all feature nodes in the subgraph is then used to obtain the dynamic relationship graph. Average clustering coefficient at time 1 Extract the average clustering coefficient values ​​corresponding to all historical time window subgraphs in the historical topological baseline, and calculate the arithmetic mean of all extracted average clustering coefficient values ​​to obtain the mean average clustering coefficient value of the historical topological baseline. Based on all extracted average clustering coefficient values ​​and Calculate the average clustering coefficient value and its relationship with each clustering coefficient in turn. The average variance of the clustering coefficients of the historical topological baseline is obtained by squared differences and then taking the arithmetic mean of all squared differences. The calculated , , , , , Substitute the values ​​into the topology deviation calculation formula and complete the following steps sequentially. and , and Calculate the difference, then square the two difference results, and finally compare the squared results with... , Perform division calculations, then sum the results of the two divisions, and finally take the square root of the sum to obtain the dynamic relationship graph data. topological deviation at time This completes the extraction of topological deviation from dynamic relationship graph data.

[0127] A preset threshold is set for topological deviation, which is determined by the statistical data of historical topological deviation of the target water area. Each value of topological deviation is assigned a corresponding time dimension information. The values ​​of topological deviation are compared with the preset threshold point by point. The time intervals corresponding to the values ​​exceeding the preset threshold are extracted and arranged in chronological order. Each extracted time interval is assigned a unique interval identifier to complete the detection of abnormal intervals of topological deviation and obtain a sequence of candidate abnormal intervals with deviation exceeding the preset threshold.

[0128] The time range of each interval in the candidate abnormal interval sequence is verified one by one. When the time interval between two adjacent candidate abnormal intervals is less than a preset interval merging threshold, the time ranges of the two intervals are merged to form a continuous time interval. The merged interval is re-assigned an interval identifier. Intervals that do not meet the merging condition are kept in their original state. This process of merging the candidate abnormal interval sequence is completed, resulting in the abnormal intervals of the topological deviation. Based on the time range corresponding to the abnormal intervals of the topological deviation, all timestamped subgraphs are extracted from the time-series dynamic graph sequence of the dynamic relationship graph data. The extracted subgraphs are arranged in chronological order, and the extracted subgraph sets are assigned association identifiers corresponding to the abnormal intervals. This process of extracting interval data from the dynamic relationship graph data is completed, resulting in the abnormal time period subgraph sets of the dynamic relationship graph data. The abnormal time period sub-map set of dynamic relationship map data is associated and matched with the abnormal interval of topological deviation. The time range, deviation value and other attribute information of the abnormal interval are structured and integrated with the topological features, node edge attributes and other information of the abnormal time period sub-map set. The information is entered and stored according to the preset structured data format to form a unified dataset. The structured aggregation of the abnormal time period sub-map set and abnormal interval is completed to obtain the environmental evolution characteristic data of the target water area.

[0129] This topology deviation calculation formula standardizes and merges the deviations of two topology indicators, network density and average clustering coefficient, eliminating the influence of differences in the dimensions of the two indicators. This provides a unified numerical benchmark for calculating the comprehensive deviation of the dynamic relationship graph topology from the historical topology baseline. The formula amplifies the deviation of the dynamic relationship graph from the historical topology baseline through difference and square calculations. The deviation of the topological index from the historical topological baseline at any given time allows even minor topological changes to be clearly reflected numerically, improving the sensitivity of topological anomaly identification. The calculation formula uses a square root operation to restore the fused sum of squares to the original dimensional deviation value, thus ensuring the calculated topological deviation... It can intuitively reflect the dynamic relationship map data in The degree of topological anomaly at any given time facilitates the subsequent setting of a uniform threshold for anomaly detection. This calculation formula extracts the topological deviation from dynamic relational graph data, and the obtained topological deviation... This provides a quantifiable criterion for subsequent anomaly detection of dynamic relationship graph data, giving clear numerical basis for the extraction of anomaly intervals. The calculation formula quantifies the degree of topological deviation in the dynamic relationship graph through numerical calculation, standardizing the process of extracting topological deviation from dynamic relationship graph data and ensuring the reproducibility and accuracy of the subsequent anomaly topology extraction process.

[0130] The beneficial effects are as follows: By constructing a network and extracting abnormal topology from element-level spatiotemporal evolution trajectory data step by step, a refined characterization and dynamic tracking of the spatiotemporal evolution elements of the target water area are achieved. Relying on clear judgment thresholds and standardized processing procedures, the construction of dynamic relationship graph data and the extraction of abnormal topology are reproducible. Simultaneously, through interval detection, merging, and structured aggregation, abnormal spatiotemporal intervals and corresponding topological features in the environmental evolution process of the target water area are accurately located. The resulting environmental evolution feature data can completely and accurately reflect the environmental evolution patterns and abnormal characteristics of the target water area, providing accurate and comprehensive data source support for environmental analysis and judgment of the target water area. In this embodiment of the invention, topological deviation is used to quantify the comprehensive deviation of the overall topological structure of the dynamic relationship graph from its historical normal state at a certain moment. In the calculation formula, the mean network density represents the average level of the tightness of connections between nodes in the dynamic relationship graph under the historical normal state. The network density variance characterizes the dispersion of network density fluctuations around its mean under the historical normal state. The mean average clustering coefficient represents the average level of local clustering of nodes in the dynamic relationship graph under the historical normal state. The variance of the average clustering coefficient characterizes the dispersion of the average clustering coefficient around its mean under historical normal conditions. The final value obtained is the topological deviation of the dynamic relationship graph data at time t. The larger this topological deviation value, the more significant the deviation of the dynamic relationship graph at time t from the historical normal state in terms of both global network connectivity density and local clustering characteristics.

[0131] Anomaly detection based on threshold comparison is performed on topology deviation to obtain a candidate anomaly interval sequence. A fixed topology deviation threshold is set. Each value in the topology deviation sequence is read in chronological order. Each topology deviation value is compared with the preset threshold. If the topology deviation value is greater than the threshold, that moment is marked as an anomaly. All moments marked as anomalies are extracted in chronological order. The continuity of these anomalies on the time axis is checked. A series of temporally continuous and uninterrupted anomalies are merged into a time interval. All such time intervals obtained after traversing the entire topology deviation sequence are sorted according to their starting time to form a sequence. This sequence is the candidate anomaly interval sequence for topology deviation.

[0132] Based on the minimum time interval determination, the candidate abnormal interval sequence is processed by interval merging to obtain abnormal intervals with a set minimum time interval threshold. Starting from the first interval in the candidate abnormal interval sequence, the time interval between the current interval and the next adjacent interval in the sequence is checked. If the time interval is less than or equal to the minimum time interval threshold, the two intervals are considered sufficiently close in time. These two intervals are merged into a new, larger interval. The merged new interval replaces the positions of the original two intervals and continues to be compared with subsequent adjacent intervals and perform possible merging operations. If the time interval is greater than the minimum time interval threshold, the current interval is retained, and the next interval is used as the new current interval for subsequent comparisons. After traversing and processing the entire candidate abnormal interval sequence, all intervals that remain after the necessary merging operations constitute the abnormal intervals of the topological deviation.

[0133] Based on the matched timestamps, interval data extraction is performed on the abnormal intervals to obtain the abnormal time period sub-graph set and dynamic relationship graph data. This data is a graph sequence organized in chronological order, where each graph is identified by a timestamp. An abnormal interval is read, containing a start time and an end time. All graphs whose timestamps fall between this start and end time are searched in the dynamic relationship graph data. These found graphs are extracted in chronological order and assembled into a new graph sequence. The above search and extraction operations are repeated for each abnormal interval. The new graph sequences corresponding to all abnormal intervals are collected to form a set. Each element in this set is a graph sequence, representing the complete evolution of the dynamic relationship graph within an abnormal time period. This set is the abnormal time period sub-graph set of the dynamic relationship graph data.

[0134] A structured data description framework is defined by performing structured aggregation on the anomalous time period sub-atlas and anomalous intervals to obtain environmental evolution characteristic data. This framework assigns an independent record to each anomalous event. Each record contains several fixed fields, each storing different data. Based on the framework, a structured record is generated for each anomalous event. All structured records of anomalous events are arranged in chronological order to form a dataset. This dataset is the environmental evolution characteristic data of the target water area.

[0135] This step transforms each environmental element into a node by extracting trajectory centroids and motion feature summaries, constructing a basic network graph expressing the spatiotemporal and motion relationships between elements. A sliding time window is used to serialize this static graph into a dynamically evolving relationship graph, thereby capturing the temporal changes in topology. The topological deviation of the dynamic graph is calculated based on historical baselines to quantify its anomaly level. Significant anomalous periods are identified through threshold detection and interval merging, and corresponding subgraph sequences are extracted. Finally, key information such as anomalous periods and their complete dynamic graph data is aggregated into a structured environmental evolution feature dataset. This process achieves structured analysis and archiving from element trajectories to network relationships and anomalous event features.

[0136] S5. Based on a preset environmental risk assessment knowledge base, perform dynamic risk assessment on the environmental evolution characteristic data to obtain comprehensive assessment data of the environmental evolution characteristic data.

[0137] In this embodiment of the invention, the step of performing dynamic risk assessment on the environmental evolution characteristic data based on a preset environmental risk assessment knowledge base to obtain comprehensive assessment data of the environmental evolution characteristic data includes:

[0138] Based on the scenario pattern set in the preset environmental risk assessment knowledge base, the environmental evolution feature data is pattern matched to obtain the initial scenario matching result of the environmental evolution feature data.

[0139] Based on the initial scenario matching results, the environmental risk assessment knowledge base is subjected to associated data extraction to obtain the potential risk field data of the environmental evolution characteristic data.

[0140] A multidimensional risk index assessment is performed on the potential risk field data to obtain the fusion risk index of the environmental evolution characteristic data;

[0141] Based on the threshold ranges and level determination rules in the environmental risk assessment knowledge base, the fusion risk index is compared with the threshold to obtain the risk level of the fusion risk index, and the data is integrated to obtain the comprehensive assessment data of the environmental evolution characteristic data.

[0142] Based on a pre-set set of scenario patterns in the environmental risk assessment knowledge base, environmental evolution characteristic data is compared one by one with each scenario pattern in the knowledge base. The system identifies which pre-set scenario pattern the data is most similar to in terms of change trends, key indicator combinations, and spatiotemporal distribution, thus completing pattern matching and obtaining initial scenario matching results. This matching process is accomplished by calculating the similarity between the characteristic data and each pattern, selecting the scenario pattern with the highest similarity as the matching result.

[0143] Based on the scenario patterns identified in the initial scenario matching results, all data pre-associated with these patterns are extracted from the environmental risk assessment knowledge base. This associated data includes historical cases, types of potential risks, affected areas, and related physicochemical parameter sets. Together, these constitute potential risk field data for the current environmental evolution characteristics, providing comprehensive background information for subsequent risk assessments.

[0144] A multidimensional risk index assessment is performed on the extracted potential risk field data. This assessment is carried out from multiple fixed dimensions such as hazard, probability, vulnerability, and exposure. Specific indicators under each dimension are quantitatively scored, and the scores of all dimensions are merged and calculated according to pre-set rules to finally generate a single, comprehensive fusion risk index to characterize the overall risk level.

[0145] Based on the predefined threshold ranges and level determination rules corresponding to different risk levels in the environmental risk assessment knowledge base, the fused risk index obtained in the previous step is compared with these thresholds. Depending on the specific threshold range the index falls into, its risk level is determined, such as low, medium, or high risk. Finally, environmental evolution characteristic data, initial scenario matching results, potential risk field data, fused risk index, and their determined risk levels are integrated and structured to form the final comprehensive assessment data report.

[0146] This step, by matching environmental evolution data with knowledge base scenarios, can quickly identify the corresponding initial risk types. Based on the matching results, relevant potential risk field data is extracted, providing comprehensive and accurate background information for the assessment. Next, a multi-dimensional fusion assessment of the risk fields is performed, ultimately generating a quantitative fusion risk index to comprehensively characterize the risk level. Finally, this index is compared with a preset threshold to automatically determine the risk level and integrate all intermediate results, thereby outputting a well-structured and clearly defined comprehensive assessment data, effectively improving the efficiency and systematic nature of environmental risk assessment.

[0147] S6. The comprehensive assessment data is structured and packaged to obtain a comprehensive monitoring report of the target water area.

[0148] In this embodiment of the invention, the step of structurally encapsulating the comprehensive assessment data to obtain a comprehensive monitoring report for the target water area includes:

[0149] The core conclusions are extracted from the comprehensive evaluation data to obtain the report summary data of the comprehensive evaluation data;

[0150] Evidence chains are established by linking the report summary data, the environmental element identification result data, the element-level spatiotemporal evolution trajectory data, and the environmental evolution characteristic data to obtain a traceable data association network for the comprehensive evaluation data.

[0151] Visualize and render the traceable data association network to obtain a draft evaluation report of the comprehensive evaluation data;

[0152] Based on a preset report template, the draft assessment report is structured and organized to obtain a comprehensive monitoring report for the target water area.

[0153] The core judgments extracted from the comprehensive assessment data mainly include the final determined risk level, key fusion risk index values, and the most important potential risk types. These extracted core information collectively constitute the report summary data, which is a concise summary of the lengthy assessment results.

[0154] The generated report summary data is linked to the environmental element identification results data, detailed spatiotemporal evolution trajectory data of each environmental element, and original environmental evolution characteristic data in the previous stages to form an evidence chain. The method is to establish a two-way connection relationship between these data entities, indicating who derived whom and who supported whose conclusions, thereby weaving a complete traceable data association network that shows the data source and reasoning path.

[0155] The completed traceable data association network is visualized and rendered. Specifically, nodes represent various data entities, and lines represent the relationships between them. Different colors and shapes are configured for nodes and lines according to data types and risk levels. Finally, an intuitive graphic is generated. This graphic, which contains all the core data and their relationships, is the draft of the assessment report.

[0156] Based on a pre-set fixed report template, the draft assessment report is structured and organized. The process involves filling in the corresponding chapter positions in the template with visualization graphics, report summary data, and other necessary background information, such as the overview, data analysis, and conclusions. After systematic typesetting and integration, a standard document with a standardized format and complete content is finally generated. This document is the comprehensive monitoring report of the target water area.

[0157] This step extracts core conclusions to form a concise report summary, facilitating a quick grasp of the key assessment points. Next, the summary is linked with multi-source process data to create a traceable data association network, ensuring the rigor and verifiability of the assessment conclusions. This network is then visualized to generate an intuitive draft assessment report, transforming complex data relationships into easily understandable graphics. Finally, the content is structured according to a template to generate a comprehensive monitoring report of the target water area that is formatted correctly, logically clear, and complete, significantly improving the report's professionalism, readability, and decision support efficiency.

[0158] like Figure 2 The diagram shown is a functional block diagram of an image recognition-based aquatic environment monitoring system provided in an embodiment of the present invention.

[0159] The image recognition-based aquatic environment monitoring system 100 described in this invention can be installed in an electronic device. Depending on the functions implemented, the image recognition-based aquatic environment monitoring system 100 may include a data normalization module 101, an element identification module 102, an evolution analysis module 103, a relationship mining module 104, a risk assessment module 105, and a report generation module 106. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function, and are stored in the memory of the electronic device.

[0160] In this embodiment, the functions of each module / unit are as follows:

[0161] The data normalization module 101 is used to normalize the multi-source image dataset of the target water area to obtain a standardized image data sequence of the target water area.

[0162] The element recognition module 102 is used to perform feature deconstruction on the standardized image data sequence to obtain multi-scale feature data of the standardized image data sequence, and to perform semantic category mapping on the multi-scale feature data to obtain environmental element recognition result data of the target water area.

[0163] The evolution analysis module 103 is used to perform dynamic evolution analysis on the environmental element identification result data to obtain element-level spatiotemporal evolution trajectory data of the target water area.

[0164] The relationship mining module 104 is used to construct a network from the element-level spatiotemporal evolution trajectory data to obtain dynamic relationship graph data of the element-level spatiotemporal evolution trajectory data, and to extract abnormal topology from the dynamic relationship graph data to obtain environmental evolution feature data of the target water area.

[0165] The risk assessment module 105 is used to perform dynamic risk assessment on the environmental evolution characteristic data based on a preset environmental risk assessment knowledge base, and obtain comprehensive assessment data of the environmental evolution characteristic data.

[0166] The report generation module 106 is used to encapsulate the comprehensive assessment data in a structured manner to obtain a comprehensive monitoring report of the target water area.

[0167] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0168] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0169] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0170] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0171] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0172] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for monitoring aquatic environments based on image recognition, characterized in that, The method includes: S1. Perform data normalization on the multi-source image dataset of the target water area to obtain a standardized image data sequence of the target water area; S2. Perform feature deconstruction on the standardized image data sequence to obtain multi-scale feature data of the standardized image data sequence, and perform semantic category mapping on the multi-scale feature data to obtain environmental element identification result data of the target water area; S3. Perform dynamic evolution analysis on the environmental element identification result data to obtain element-level spatiotemporal evolution trajectory data of the target water area; S4. The element-level spatiotemporal evolution trajectory data is networked to obtain dynamic relationship map data of the element-level spatiotemporal evolution trajectory data, and the abnormal topology structure of the dynamic relationship map data is extracted to obtain the environmental evolution characteristic data of the target water area. S5. Based on a preset environmental risk assessment knowledge base, perform dynamic risk assessment on the environmental evolution characteristic data to obtain comprehensive assessment data of the environmental evolution characteristic data. S6. The comprehensive assessment data is structured and packaged to obtain a comprehensive monitoring report of the target water area.

2. The aquatic environment monitoring method based on image recognition as described in claim 1, characterized in that, The process of data normalizing the multi-source image dataset of the target water area to obtain a standardized image data sequence of the target water area includes: Target regions are extracted from the multi-source image dataset of the target water area to obtain the effective image dataset of the target water area; The effective image dataset is optimized for image sharpness to obtain an enhanced image dataset of the target water area; Based on a predefined spatial reference datum, the enhanced image dataset is scale-normalized to obtain a set of corrected image data for the target water area; The corrected image dataset is normalized to obtain a standardized image data sequence of the target water area.

3. The aquatic environment monitoring method based on image recognition as described in claim 1, characterized in that, The step of deconstructing the standardized image data sequence to obtain multi-scale feature data of the standardized image data sequence, and performing semantic category mapping on the multi-scale feature data to obtain environmental element identification result data of the target water area, includes: The standardized image data sequence is deconstructed to obtain a multi-scale feature representation of the standardized image data sequence; Cross-scale feature interaction is performed on the multi-scale feature representation to obtain the fused feature map of the standardized image data sequence; Based on the fused feature map, the pixel features in the fused feature map are subjected to category assignment mapping to obtain the pixel category probability distribution of the fused feature map; Spatial consistency optimization is performed on the pixel category probability distribution to obtain the environmental element identification result data of the target water area.

4. The aquatic environment monitoring method based on image recognition as described in claim 1, characterized in that, The dynamic evolution analysis of the environmental element identification results data to obtain the element-level spatiotemporal evolution trajectory data of the target water area includes: Perform cross-time frame instance matching on the environmental element identification result data to obtain the associated element instance set of the environmental element identification result data; Based on the set of associated element instances, the spatiotemporal location sequence of the element instances in the set of associated element instances is reconstructed to obtain the smooth spatiotemporal trajectory of the element instances. The smoothed spatiotemporal trajectory is subjected to motion feature parameter quantization to obtain the temporal motion feature data of the element instance; Evolutionary pattern analysis is performed on the temporal motion characteristic data to obtain element-level spatiotemporal evolution trajectory data of the target water area.

5. The aquatic environment monitoring method based on image recognition as described in claim 1, characterized in that, The step of performing evolution pattern analysis on the temporal motion feature data to obtain element-level spatiotemporal evolution trajectory data of the target water area includes: The temporal motion feature data is subjected to change trend feature extraction to obtain a multi-dimensional change trend description of the temporal motion feature data; Based on the historical evolution pattern library of the target water area, the evolution pattern matching of the multi-dimensional change trend description is performed to obtain the evolution pattern matching result of the time-series motion feature data. Based on the preset confidence judgment rules, the confidence of the evolution pattern matching results is evaluated to obtain the evolution pattern results of the temporal motion feature data; The evolution model results are sequence-tuned to obtain element-level spatiotemporal evolution trajectory data of the target water area.

6. The aquatic environment monitoring method based on image recognition as described in claim 1, characterized in that, The process involves constructing a network from the element-level spatiotemporal evolution trajectory data to obtain dynamic relationship graph data of the element-level spatiotemporal evolution trajectory data, and extracting abnormal topology from the dynamic relationship graph data to obtain environmental evolution characteristic data of the target water area, including: The element-level spatiotemporal evolution trajectory data is converted into graph nodes to obtain the element nodes of the element-level spatiotemporal evolution trajectory data. Based on the spatiotemporal proximity and motion correlation in the feature-level spatiotemporal evolution trajectory data, the feature nodes are associated with edges to obtain the basic association graph of the feature-level spatiotemporal evolution trajectory data. The basic association graph is serialized into a time-series dynamic graph to obtain the dynamic relationship map data of the element-level spatiotemporal evolution trajectory data; Based on the historical topological baseline of the target water area, the topological deviation of the dynamic relationship map data is extracted to obtain the topological deviation of the dynamic relationship map data. Anomaly interval detection is performed on the topological deviation, and the intervals with deviations exceeding a preset threshold and the corresponding dynamic relationship graph data are structurally aggregated to obtain the environmental evolution characteristic data of the target water area.

7. The aquatic environment monitoring method based on image recognition as described in claim 6, characterized in that, The process involves detecting abnormal intervals in the topological deviation and structurally aggregating the intervals where the deviation exceeds a preset threshold and the corresponding dynamic relationship graph data to obtain environmental evolution characteristic data of the target water area, including: Anomaly interval detection is performed on the topological deviation to obtain a sequence of candidate anomaly intervals whose deviation exceeds a preset threshold; The candidate abnormal interval sequence is subjected to interval merging processing to obtain the abnormal interval of the topological deviation. Based on the abnormal intervals, interval data extraction is performed on the dynamic relationship graph data to obtain the abnormal time period sub-graph set of the dynamic relationship graph data; The abnormal time period sub-map set and the abnormal interval are structurally aggregated to obtain the environmental evolution characteristic data of the target water area.

8. The aquatic environment monitoring method based on image recognition as described in claim 1, characterized in that, The method, based on a pre-set environmental risk assessment knowledge base, performs dynamic risk assessment on the environmental evolution characteristic data to obtain comprehensive assessment data of the environmental evolution characteristic data, including: Based on the scenario pattern set in the preset environmental risk assessment knowledge base, the environmental evolution feature data is pattern matched to obtain the initial scenario matching result of the environmental evolution feature data. Based on the initial scenario matching results, the environmental risk assessment knowledge base is subjected to associated data extraction to obtain the potential risk field data of the environmental evolution characteristic data; A multidimensional risk index assessment is performed on the potential risk field data to obtain the fusion risk index of the environmental evolution characteristic data; Based on the threshold ranges and level determination rules in the environmental risk assessment knowledge base, the fusion risk index is compared with the threshold to obtain the risk level of the fusion risk index, and the data is integrated to obtain the comprehensive assessment data of the environmental evolution characteristic data.

9. The aquatic environment monitoring method based on image recognition as described in claim 1, characterized in that, The process of structuring and encapsulating the comprehensive assessment data to obtain a comprehensive monitoring report for the target water area includes: The core conclusions are extracted from the comprehensive evaluation data to obtain the report summary data of the comprehensive evaluation data; Evidence chains are established by linking the report summary data, the environmental element identification result data, the element-level spatiotemporal evolution trajectory data, and the environmental evolution characteristic data to obtain a traceable data association network for the comprehensive evaluation data. Visualize and render the traceable data association network to obtain a draft evaluation report of the comprehensive evaluation data; Based on a preset report template, the draft assessment report is structured and organized to obtain a comprehensive monitoring report for the target water area.

10. A water environment monitoring system based on image recognition, characterized in that, The system for implementing the image recognition-based aquatic environment monitoring method of claim 1 includes: The data normalization module is used to normalize the multi-source image dataset of the target water area to obtain a standardized image data sequence of the target water area; The feature recognition module is used to deconstruct the standardized image data sequence to obtain multi-scale feature data of the standardized image data sequence, and to perform semantic category mapping on the multi-scale feature data to obtain environmental feature recognition result data of the target water area. The evolution analysis module is used to perform dynamic evolution analysis on the environmental element identification result data to obtain element-level spatiotemporal evolution trajectory data of the target water area; The relationship mining module is used to construct a network from the element-level spatiotemporal evolution trajectory data to obtain dynamic relationship graph data of the element-level spatiotemporal evolution trajectory data, and to extract abnormal topology from the dynamic relationship graph data to obtain environmental evolution feature data of the target water area. The risk assessment module is used to perform dynamic risk assessment on the environmental evolution characteristic data based on a preset environmental risk assessment knowledge base, and obtain comprehensive assessment data of the environmental evolution characteristic data. The report generation module is used to encapsulate the comprehensive assessment data in a structured manner to obtain a comprehensive monitoring report of the target water area.

Images