An abnormal region detection method based on point position multi-time phase atlas and a storage medium

Abstract

The application discloses an abnormal area detection method based on a point position multi-time phase atlas, which comprises the following steps: collecting multi-time phase core data according to detection requirements, and generating a point position multi-time phase atlas through data registration; establishing a normal fluctuation model based on the point position multi-time phase atlas, and marking suspected abnormal point positions; calculating the abnormal degree of the point positions through an abnormal clustering algorithm, and identifying abnormal point position clusters; constructing an energy function, identifying the boundary of an abnormal area, and preliminarily identifying the range of the abnormal area; combining the abnormal point position clusters and the boundary of the abnormal area, constructing a diffusion area model to predict the diffusion trend of the abnormal area; according to the boundary and the diffusion trend, determining whether the point positions in the area meet abnormal characteristics, and confirming the final abnormal area; and integrating the detection results to generate an abnormal area report containing abnormal information, visual charts and suggested measures. The application aims to solve the problems that the traditional method is difficult to accurately capture the dynamic evolution of abnormalities, the adaptability of feature fusion is poor, and the diffusion trend prediction is inaccurate.

Description

Technical Field

[0001] This invention relates to the field of regional detection technology, and in particular to an abnormal region detection method and storage medium based on multi-temporal image atlases of points. Background Technology

[0002] In fields such as ecological environment monitoring and urban safety early warning, anomaly detection is a crucial link in ensuring system stability and security. Traditional methods often rely on single-temporal data or manual judgment, which has significant limitations. Single-temporal data struggles to capture the dynamic evolution of anomalies and easily misses gradual anomalies; manual judgment is heavily influenced by subjective factors, resulting in low efficiency and insufficient accuracy. With the development of technologies such as remote sensing and the Internet of Things, a large number of multi-temporal atlases of data points have been generated, providing a data foundation for dynamic anomaly detection. However, how to efficiently utilize these multi-temporal atlases to accurately extract anomaly features and detect anomaly areas remains a technical challenge. This application aims to address the problems of traditional methods' inability to accurately capture the dynamic evolution of anomalies, poor adaptability of fused features, and inaccurate prediction of diffusion trends. Summary of the Invention

[0003] This invention provides an anomaly region detection method based on a multi-temporal atlas of point locations, comprising:

[0004] Collect multi-temporal core data according to the detection requirements, and generate multi-temporal map sets of points through data registration;

[0005] A normal fluctuation model is established based on a multi-temporal atlas of points, and suspected abnormal points are marked.

[0006] Based on suspected abnormal locations, the degree of abnormality of the locations is calculated using an anomaly clustering algorithm to identify abnormal location clusters;

[0007] An energy function is constructed based on the cluster of abnormal points to identify the boundaries of abnormal regions and to preliminarily identify the range of abnormal regions.

[0008] By combining anomalous point clusters and anomalous region boundaries, a diffusion region model is constructed to predict the diffusion trend of anomalous regions.

[0009] Based on the regional boundaries and diffusion trends, determine whether the points within the region meet the abnormal characteristics, and confirm the final abnormal area.

[0010] Integrate the test results to generate an anomaly report containing abnormal information, visual charts, and recommended measures.

[0011] The above-described anomaly detection method based on a multi-temporal point map atlas includes: collecting multi-temporal core data according to detection requirements and generating a multi-temporal point map atlas through data registration; and including:

[0012] Collect multi-temporal core data according to detection requirements, and perform spatiotemporal alignment and standardization preprocessing for detection tasks;

[0013] Based on the preprocessed multi-temporal core data, a multi-temporal atlas of points is generated by fusing feature point matching and spatiotemporal registration algorithms.

[0014] The above-described method for anomaly region detection based on multi-temporal point atlases includes establishing a normal fluctuation model based on the multi-temporal point atlases and marking suspected anomaly points, including:

[0015] For each location, analyze its multi-temporal atlas sequence and construct a normal fluctuation model that reflects the normal change pattern and fluctuation range of the location itself;

[0016] The latest temporal data of each point is compared with its own dynamic benchmark model, and data points that deviate from the normal fluctuation range are automatically identified and marked as suspected abnormal points.

[0017] The above-described method for detecting abnormal regions based on multi-temporal image atlases includes, in which, based on suspected abnormal points, an anomaly clustering algorithm is used to calculate the degree of anomaly at each point and identify clusters of abnormal points, including:

[0018] Based on the feature data of suspected anomalies, an anomaly clustering algorithm is used to initially group them, and the local anomaly degree value of each point in its group is calculated.

[0019] Based on the local anomaly values ​​of each point, clusters with high significance are optimized and selected, and these are identified as anomalous point clusters with clear spatial aggregation characteristics.

[0020] The above-described anomaly region detection method based on multi-temporal image atlases includes, in which, an energy function is constructed based on anomalous point clusters to identify the boundaries of the anomalous region and to preliminarily identify the extent of the anomalous region, including:

[0021] By integrating the statistical characteristics and geospatial attributes of anomalous point clusters, an energy function for boundary identification is constructed.

[0022] By minimizing the energy function to solve for the optimal segmentation boundary, the precise range of the abnormal region can be preliminarily determined.

[0023] The above-described anomaly region detection method based on multi-temporal point atlases includes, in which anomaly point clusters and anomaly region boundaries are combined to construct a diffusion region model to predict the diffusion trend of anomaly regions, including:

[0024] Analyze the anomalous intensity gradient within the anomalous region and external environmental factors to determine the diffusion driving factors and parameters;

[0025] A diffusion model based on spatiotemporal geographic weighted regression is constructed to simulate the diffusion trend and direction of anomalous regions under the influence of driving factors.

[0026] The above-described anomaly region detection method based on multi-temporal image atlases includes, based on region boundaries and diffusion trends, determining whether points within the region meet anomaly characteristics and confirming the final anomaly region, including:

[0027] The spread trend is fed back to the anomaly detection module to enhance the verification of points on potential spread paths;

[0028] By combining the initial results, boundary and verification information, the final range and confidence level of the anomaly region are optimized and confirmed.

[0029] The beneficial effects achieved by this invention are as follows:

[0030] In-depth analysis of multi-temporal image sets of locations enables precise capture of the dynamic evolution of anomalies over time, providing more comprehensive temporal information support for anomaly detection. Dynamically assigning feature weights based on domain characteristics allows the anomaly detection model to adapt to different domain scenarios, significantly improving the accuracy and reliability of detection results. Regarding anomaly diffusion trend prediction, the diffusion model, fully considering the spatiotemporal nonlinear relationships of factors, accurately simulates the anomaly diffusion pattern, providing a scientific basis for anomaly prevention and control, and effectively assisting various fields in responding to anomaly issues in a timely and efficient manner. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0032] Figure 1 This is a flowchart of an abnormal region detection method based on a multi-temporal image atlas provided in this application embodiment. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] like Figure 1 As shown in the figure, this application provides an anomaly region detection method based on a multi-temporal atlas of points, including:

[0035] S1: Collect multi-temporal core data according to the detection requirements, and generate multi-temporal map sets of points through data registration;

[0036] The process involves collecting multi-temporal core data based on testing requirements and generating multi-temporal atlases of locations through data registration, including the following sub-steps:

[0037] S11: Collect multi-temporal core data according to detection requirements, and perform spatiotemporal alignment and standardization preprocessing for detection tasks;

[0038] Based on the preset detection requirements, the collection objects and time frequencies of multi-temporal core data are determined, and raw data covering different time dimensions of the target area are collected. Then, the spatiotemporal alignment module is activated to establish a unified coordinate system based on fixed geographic landmarks in the region. Data from different time phases are mapped to this coordinate system to eliminate spatial location deviations, and the timestamps of each data are extracted to calibrate time scale differences. Next, standardized preprocessing is performed. First, the data format is uniformly converted, and then the feature normalization method is used to eliminate the influence of different data sources and remove random noise in the data, ensuring that the preprocessed data meets the requirements of subsequent feature point matching and spatiotemporal registration in terms of spatiotemporal consistency and data quality.

[0039] S12: Based on the preprocessed multi-temporal core data, a multi-temporal map set of points is generated by fusion through feature point matching and spatiotemporal registration algorithms.

[0040] The high-precision spatiotemporal registration process begins by receiving standardized preprocessed multi-temporal core data, which employs a hierarchical feature matching strategy. First, significant key points are extracted from the data volume of each temporal phase, generating high-dimensional feature descriptors. Then, coarse matching is performed, and the optimal spatial transformation model is iteratively calculated to eliminate mismatched point pairs, achieving pixel-level spatial alignment. Further fine-tuning compensation is applied to minor displacements between consecutive temporal phases. After registration, the multi-temporal data is fused and recombined according to geographic locations, generating a data cube stacked along the time dimension for each unique geographic coordinate, thus constructing a structurally sound and spatiotemporally consistent multi-temporal atlas. Finally, a quality assessment loop is executed, automatically evaluating registration accuracy by calculating cross-correlation coefficients and root mean square error between sequences, and retrospectively reprocessing local areas that fail to meet preset tolerance thresholds.

[0041] S2: Establish a normal fluctuation model based on multi-temporal atlas of points and mark suspected abnormal points;

[0042] The process of establishing a normal fluctuation model based on a multi-temporal atlas of data points and marking suspected abnormal data points includes the following sub-steps:

[0043] S21: Analyze the multi-temporal atlas sequence of each point and construct a normal fluctuation model that reflects the normal change pattern and fluctuation range of the point itself;

[0044] Historical observation sequence data, standardized and spatiotemporally aligned, is extracted from multi-temporal image sets of various locations. A normal fluctuation model is constructed using this historical observation sequence data as input. This model employs an encoder-decoder structure: the encoder consists of two layers of unidirectional gated recurrent units (GRUs) to deeply capture complex time-dependent patterns and periodic regularities in the historical sequences of the locations; the decoder is implemented by fully connected layers, responsible for mapping the learned temporal features to predictions of future temporal values. The model's input is a one-dimensional sequence of observations of a location over multiple past temporal phases, and the output is a prediction of its next temporal value. The system uses historical sequence data validated to be in a normal state to train the model in a supervised manner. The Adam optimizer minimizes the mean squared error loss between the predicted and true values. Appropriate learning rates and batch sizes are set during the process, and iterative optimization continues until the model converges, enabling it to accurately predict the "baseline" state of the location. After training, a dynamic confidence interval is adaptively calculated based on the predicted residual distribution of the model across all training samples. This interval can sensitively reflect the statistical characteristics and uncertainties of the normal fluctuations at that point.

[0045] S22: Compare the latest temporal data of each point with its own dynamic benchmark model, and automatically identify and mark data points that deviate from the normal fluctuation range as suspected abnormal points.

[0046] Based on the latest temporal data and normal fluctuation models and dynamic confidence intervals for each location, the latest temporal data undergoes standardization processing consistent with historical sequences. The continuous historical observation sequences preceding the latest temporal data for that location are input into the model. A two-layer unidirectional GRU module reproduces the time-dependent pattern, decodes the output, and predicts the latest temporal data. The residual between the actual observed value and the model's predicted value for the latest temporal data is calculated. The system compares this residual with the dynamic confidence interval. Simultaneously, considering domain characteristics (e.g., ecological domains focus on long-term trend deviations, while industrial domains focus on instantaneous mutations), a deviation sensitivity coefficient is set. If the residual exceeds the interval and still meets the anomaly judgment criteria after coefficient weighting, the location is automatically marked as a suspected anomaly, and characteristic information such as residual magnitude and deviation direction is recorded.

[0047] S3: Based on suspected abnormal locations, calculate the degree of abnormality of the locations using an anomaly clustering algorithm, and identify clusters of abnormal locations;

[0048] The process of calculating the degree of anomalousness of suspected anomalous locations using an anomaly clustering algorithm and identifying anomalous location clusters includes the following sub-steps:

[0049] S31: Based on the feature data of suspected anomalies, use an anomaly clustering algorithm to initially group them, and calculate the local anomaly degree value of each point in its group;

[0050] The system acquires feature data of suspected anomalies, including spatial coordinates, anomaly residual magnitude, deviation direction, and temporal features. After standardizing each feature, a multi-dimensional feature space is constructed, and adaptive weights are assigned based on the importance of features in the domain's anomaly patterns. Simultaneously, the joint density distribution of points in the physical and attribute spaces is calculated. By dynamically optimizing bandwidth parameters to adapt to differences in point set density across different regions, cluster boundaries are automatically determined, identifying potential anomaly clusters exhibiting aggregation in both spatiotemporal and attribute dimensions, thus completing preliminary grouping. For each group, the weighted Mahalanobis distance between a point and the group center across multiple feature dimensions is calculated. This distance is compared with the average dispersion of points within the group to obtain the local anomaly degree value of each point within its group. This value accurately characterizes its relative prominence within the local cluster.

[0051] The formula for calculating the degree of local anomaly is as follows:

[0052]

[0053] This represents the weighted Mahalanobis distance between the i-th point and the k-th cluster center across multiple feature dimensions; It is the standard deviation of the weighted Mahalanobis distance between all points in the k-th cluster and the cluster center; This represents the mean local anomaly level of all points within the k-th cluster; The estimated local anomaly value for the i-th point is a preliminary anomaly quantification result calculated based on its own characteristics and its relationship with surrounding points. The statistical dispersion parameter of the kth cluster is determined by the distribution characteristics of the local outlier estimates within the cluster and is used to adjust the sensitivity of the error function to the degree of deviation. It is the magnitude of the feature gradient vector at the i-th point, reflecting the degree of drastic change in the local feature space; The angle between the feature directions of the i-th point and the j-th point in the same cluster is obtained by the dot product of the feature vectors. It is used to measure the consistency of the directions of the two points in the feature space. The smaller the angle, the more similar the trend of feature change. It is the spatial attenuation coefficient, whose value is set according to the characteristics of the domain to control the attenuation rate of the influence weight of the surrounding points on the target point as the spatial distance increases. The larger the value, the faster the attenuation. Indicates the first The attribute weight of each point; Representing the The spatial coordinate vector of a point; Then it is the first Spatial coordinate vectors of points.

[0054] S32: Based on the local anomaly values ​​of each point, optimize and select clusters with high significance, and identify them as anomalous point clusters with clear spatial aggregation characteristics.

[0055] Based on the preliminary clusters and the local anomaly scores of each point, the comprehensive characteristics of each cluster are first calculated: including the mean and standard deviation of the local anomaly scores within the cluster, the compactness of the spatial distribution of the point set, and the gradient difference between the cluster boundary and the normal region. Combining domain characteristics (e.g., the ecological domain emphasizes the spatial continuity of clusters, while the industrial domain focuses on the concentration of anomaly intensity), dynamic weights are assigned to each characteristic to calculate the comprehensive score of each cluster. High-scoring clusters are selected using an adaptive threshold (calibrated based on the score distribution of historically valid anomaly clusters). Simultaneously, a secondary verification is performed on edge clusters with scores close to the threshold, merging spatially adjacent and similar small clusters to eliminate fragmentation. Finally, the clusters that satisfy the characteristics of clear spatial aggregation, significant anomaly intensity, and high internal consistency are identified as anomaly point clusters.

[0056] S4: Construct an energy function based on the cluster of abnormal points to identify the boundaries of the abnormal region and preliminarily identify the range of the abnormal region;

[0057] The process of constructing an energy function based on anomaly clusters, identifying the boundaries of anomaly regions, and initially determining the extent of anomaly regions includes the following sub-steps:

[0058] S41: Integrate the statistical characteristics and geospatial attributes of anomalous point clusters to construct an energy function for boundary identification;

[0059] Based on the cluster data of anomaly points, statistical features (including mean anomaly intensity, density distribution gradient, and spatial decay law of local anomaly severity) and geospatial attributes (covering terrain slope, land cover type coding, spatial topological relationships, etc.) within the clusters are extracted. Normalization is performed on the statistical features, and the geographic attributes are transformed into quantitative indicators. Dynamic weights are assigned to the two types of features based on domain characteristics, and a two-factor energy function is constructed: the internal term is calculated using a spatial interpolation model of anomaly intensity, causing the function value to increase with distance from high anomaly density areas; the external term introduces a geographic constraint penalty mechanism, applying an increment to potential segments crossing key land cover boundaries, while embedding a spatial smoothing coefficient to avoid unreasonable sawtooth fluctuations at the boundaries.

[0060] The energy function formula is:

[0061]

[0062] Represented by boundary The energy function is used to quantify the rationality of the boundary as a dividing line for anomaly regions; the smaller the value, the better the boundary. and These are the dynamic weights for internal and external terms, balancing the influence of statistical characteristics and geographical attributes; This represents the anomalous region enclosed by boundary B; It is the Laplace operator of abnormal intensity, reflecting the severity of local anomalies; It is a value indicating the degree of local anomaly; Indicates the gradient of anomaly intensity; Represents the boundary normal vector; It is the geographical constraint strength coefficient; This is the feature constraint indicator function; The gradient represents the geographic constraint field (such as terrain slope), and is used to describe the spatial variation of geographic attributes;

[0063] S42: Solve for the optimal segmentation boundary by minimizing the energy function to initially determine the precise range of the abnormal region.

[0064] An initial boundary is generated using high-anomaly-density areas within a cluster as the core. This boundary covers all anomaly points and reserves a reasonable expansion buffer zone. Subsequently, a simulated annealing strategy is employed: the boundary morphology is adjusted based on the energy function gradient, and the probability acceptance mechanism of simulated annealing avoids getting trapped in local optima. An adaptive step size (dynamically scaled according to the rate of change of the energy function) is set to improve convergence efficiency. During iterative optimization, spatial smoothing constraints are introduced in real-time, and the boundary contour is processed using Gaussian filtering to eliminate high-frequency jagged fluctuations. Geographical constraints are simultaneously verified, penalizing optimization directions that touch key feature boundaries (such as roads and rivers) to ensure the boundary conforms to geographical rationality. When the energy value change over multiple iterations falls below the neighborhood threshold (set according to anomaly detection accuracy requirements) and the boundary morphology tends to stabilize, the solution is terminated, and the optimal segmentation boundary is output, delineating the initial anomaly region.

[0065] S5: Combine the clusters of abnormal points with the boundaries of abnormal regions to construct a diffusion region model to predict the diffusion trend of abnormal regions;

[0066] The process of constructing a diffusion region model by combining anomalous point clusters and anomalous region boundaries to predict the diffusion trend of anomalous regions includes the following sub-steps:

[0067] S51: Analyze the anomalous intensity gradient within the anomalous region and external environmental factors to determine the diffusion driving factors and parameters;

[0068] Based on the boundary and internal anomaly intensity data of the anomalous region, the discrete intensity values ​​of anomaly points are transformed into a continuous distribution field. The gradient field of the internal anomaly intensity is calculated, and the gradient direction and amplitude features are extracted to characterize the intrinsic dynamics of diffusion. Simultaneously, external environmental factors affecting diffusion are collected, including geographical attributes such as topographic relief, land cover type, and fluid movement direction, as well as dynamic factors in the time dimension (such as wind speed and flow rate changes). These factors are standardized and quantified according to domain characteristics (e.g., in hydrology, the weight of water flow direction is emphasized, while in urban areas, the correlation with road networks is strengthened). The contribution of each factor to diffusion is determined by the correlation between each factor and historical diffusion cases. A dynamic weight matrix is ​​constructed, and finally, a diffusion-driving parameter system is integrated, including internal gradient parameters, external environmental factor coefficients, and spatiotemporal attenuation factors.

[0069] The formula for calculating the factor diffusion contribution is as follows:

[0070]

[0071] Indicates the first Class factor at time Contribution to anomalous spread is used to quantify the extent to which the factor drives anomalous spread at a specific moment. This represents the initial time, which is the starting point for time measurement of the diffusion process. The contribution of factors to diffusion is analyzed from this moment. Indicates the first The flux divergence of class factors, where It is the first The flux vector of the class factor reflects the spatial strength of the factor-driven diffusion over time. The changes. It is the first The time decay coefficient of the class factor, with time change.

[0072] Calculate from time arrive The cumulative effect of the attenuation coefficient reflects how the factor's contribution to diffusion decreases over time. Represents all categorical factors Calculate from the initial time At the time Flux divergence modulus The integral of is taken at its maximum value; its function is to normalize the integral result of the numerator. Indicates the first The spatial influence coefficient of a factor class measures the strength of its spatial contribution to anomalous diffusion. Different factor classes... Corresponding to different degrees of spatial impact; It is the first Spatial distribution vector of class factors It is the velocity vector of the abnormal diffusion, the dot product. This indicates the spatial distribution and diffusion rate of the factor over time. The strength of synergistic effect;

[0073] Indicates the first The spatial distribution standard deviation of a factor class describes the degree of dispersion of its spatial distribution and reflects the influence of the spatial uniformity of the factor's distribution on its diffusion contribution. Indicates the first Class factor at time The actual spatial coverage range reflects the space occupied by the factor at a specific moment, affecting its contribution range to diffusion. Indicates the first The maximum spatial coverage of a class factor is the upper limit of its spatial coverage capability, used to define the actual spatial coverage. Normalization and other related processes were performed to determine its proportion of contribution to diffusion at maximum capacity.

[0074] S52: Construct a diffusion model based on spatiotemporal geographic weighted regression to simulate the diffusion trend and direction of anomalous areas under the influence of driving factors.

[0075] Based on the boundary data of anomalous regions using diffusion driving parameters, a diffusion model is constructed. This model comprises four interconnected modules: a data preprocessing module discretizes the anomalous region into uniformly sized grid cells and synchronously standardizes the driving parameters to eliminate dimensional differences; a spatiotemporal weighting module dynamically adjusts the calculation logic based on the spatiotemporal location of the grid cells to adapt to different point densities; a parameter solving module receives the weighted data and iteratively solves the spatiotemporal weighting coefficients of the driving parameters; and a trend simulation module integrates the aforementioned results to perform diffusion calculations. During training, historical diffusion case data is first divided into training and validation sets. Core parameters are determined according to domain characteristics, and iterative optimization is performed with the goal of minimizing the validation set residuals until the deviation between the model's predicted and actual values ​​stabilizes. The input consists of the grid spatiotemporal coordinates and domain-specific driving parameters, while the output includes the anomalous diffusion range, intensity distribution, and direction vector at different time points.

[0076] S6: Based on the regional boundaries and diffusion trends, re-examine whether the points within the region meet the abnormal characteristics, and confirm the final abnormal region;

[0077] The process of determining whether points within the region meet the abnormal characteristics based on the regional boundaries and diffusion trends, and confirming the final abnormal region, includes the following sub-steps:

[0078] S61: Feed the diffusion trend back to the anomaly detection module to enhance the verification of points on potential diffusion paths;

[0079] The system receives the predicted diffusion trend data of the abnormal area and spatially overlays it with the initial cluster of abnormal points. Based on the diffusion direction probability field, a strip-shaped area to be reviewed is dynamically delineated outward from the boundary of the identified abnormal area along the direction of maximum probability. For all points within this area, an enhanced review process is initiated: its core operation is to dynamically adjust the anomaly judgment rules in step S22. The system shrinks the confidence interval of its personalized dynamic benchmark model according to the diffusion gradient strength of the point; the higher the gradient value, the greater the interval shrinkage ratio, making the judgment criteria more stringent. Simultaneously, data from multiple time phases prior to the latest time phase of these points are called to form a short time series, which is input into their respective benchmark models to calculate the trend of their recent historical prediction residuals. Finally, a logical AND operation is performed on the original residual of the latest time phase of the point, the judgment result after shrinkage threshold adjustment, and the direction of its recent residual trend. Only when multiple conditions are met is the point finally marked as an anomaly confirmed after enhanced review.

[0080] S62: Based on the initial results, boundary and verification information, optimize and confirm the final range and confidence level of the anomaly region.

[0081] Based on the initial anomaly region boundary, diffusion trend data, and enhanced verification results (including the correlation score between newly added suspected anomalies and diffusion), spatial consistency checks are performed on the three types of data: buffer analysis is used to verify the spatial correlation between newly added points and the initial boundary, and the diffusion trend vector is used to evaluate whether it conforms to the predicted path, eliminating noise points with significant deviations. Then, based on domain characteristics, dynamic weights are assigned to the initial boundary, diffusion trend, and verification results to optimize the region range, incorporating verified newly added points and correcting unreasonable boundaries. Subsequently, by calculating the anomaly density, verification consistency index, and consistency with the diffusion trend within the region, the region is divided into high, medium, and low confidence levels, outputting the final anomaly region range and corresponding confidence levels.

[0082] S7: Integrate the detection results and generate an anomaly area report containing anomaly information, visualization charts, and recommended measures.

[0083] The final anomaly area range and confidence level are integrated, along with extracted attribute features of anomaly points (such as anomaly type and intensity), diffusion trend parameters (such as direction and rate), and verification results. This data is then structured to form a basic anomaly information database. Based on domain characteristics, GIS spatial rendering technology is used to generate layered color maps of the anomaly areas. Time-series visualization tools are used to plot multi-temporal change curves for key points, and diffusion trend vectors are annotated using vector graphics. Simultaneously, based on anomaly intensity, confidence level, and diffusion risk, a domain decision rule base (such as emergency investigation strategies for high-confidence areas and preventative monitoring plans associated with diffusion paths) is invoked to generate tiered recommended measures. The basic anomaly information, multi-dimensional visualization charts, and targeted recommendations are integrated and packaged to form a standardized anomaly area report.

[0084] Corresponding to the above embodiments, the present invention provides a computer storage medium, including: at least one memory and at least one processor;

[0085] The memory is used to store one or more program instructions;

[0086] A processor is used to run one or more program instructions to execute an anomaly region detection method based on a point-based multi-temporal map atlas.

[0087] Corresponding to the above embodiments, this embodiment of the invention provides a computer-readable storage medium containing one or more program instructions, which are executed by a processor to provide an anomaly region detection method based on a point-based multi-temporal map atlas.

[0088] The embodiments disclosed in this invention provide a computer-readable storage medium storing computer program instructions. When the computer program instructions are executed on a computer, the computer performs the above-described method for detecting abnormal regions based on a multi-temporal atlas of points.

[0089] In this embodiment of the invention, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0090] The various methods, steps, and logic diagrams disclosed in the embodiments of this invention can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above methods.

[0091] The storage medium can be memory, such as volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.

[0092] Among them, non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.

[0093] Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (Synchlink DRAM, SLDRAM), and direct memory bus RAM (DRRAM).

[0094] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0095] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using a combination of hardware and software. When applied as software, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. Storage media can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0096] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. An anomaly region detection method based on multi-temporal point image atlases, characterized in that, include: Collect multi-temporal core data according to the detection requirements, and generate multi-temporal map sets of points through data registration; A normal fluctuation model is established based on a multi-temporal atlas of points, and suspected abnormal points are marked. Based on suspected abnormal locations, the degree of abnormality of the locations is calculated using an anomaly clustering algorithm to identify abnormal location clusters; An energy function is constructed based on the cluster of abnormal points to identify the boundaries of abnormal regions and to preliminarily identify the range of abnormal regions. By combining anomalous point clusters and anomalous region boundaries, a diffusion region model is constructed to predict the diffusion trend of anomalous regions. Specifically, based on the boundary and internal anomalous intensity data of the anomalous area, the discrete anomalous point intensity values ​​are transformed into a continuous distribution field, the gradient field of the anomalous intensity within the area is calculated, and the gradient direction and amplitude features are extracted to characterize the intrinsic dynamics of diffusion; at the same time, external environmental factors affecting diffusion are collected, including geographical attributes and dynamic factors in the time dimension, and each factor is standardized and quantified according to the characteristics of the domain. The contribution of each factor to diffusion is determined by the correlation between each factor and historical diffusion cases. A dynamic weight matrix is ​​constructed and finally integrated into a diffusion driving parameter system that includes internal gradient parameters, external environmental factor coefficients, and spatiotemporal decay factors. The formula for calculating the factor diffusion contribution is as follows: Indicates the first Class factor at time The contribution to the spread of anomalies is used to quantify the extent to which the factor drives the spread of anomalies at a specific moment. This represents the initial time, which is the starting point for time measurement of the diffusion process. The contribution of factors to diffusion is analyzed from this moment. Indicates the first The flux divergence of class factors, where It is the first The flux vector of the class factor reflects the spatial strength of the factor-driven diffusion over time. Changes; It is the first The time decay coefficient of the class factor, with time change; Calculate from time arrive The cumulative effect of the attenuation coefficient reflects how the factor's contribution to diffusion decreases over time. Represents all categorical factors Calculate from the initial time At the time Flux divergence modulus The integral of is taken at its maximum value; its function is to normalize the integral result of the numerator. Indicates the first The spatial influence coefficient of a factor class measures the strength of its spatial contribution to anomalous diffusion. Different factor classes... Corresponding to different degrees of spatial impact; It is the first Spatial distribution vector of class factors It is the velocity vector of the abnormal diffusion, the dot product. This indicates the spatial distribution and diffusion rate of the factor over time. The strength of synergistic effect; Indicates the first The spatial distribution standard deviation of a factor is used to describe the degree of dispersion of the spatial distribution of the factor and reflects the influence of the spatial uniformity of the factor's distribution on the diffusion contribution. Indicates the first Class factor at time The actual spatial coverage range reflects the space occupied by the factor at a specific moment, affecting its contribution range to diffusion; Indicates the first The maximum spatial coverage of a class factor is the upper limit of its spatial coverage capability, used to define the actual spatial coverage. Normalization was performed to determine its proportion of diffusion contribution at maximum capacity; Based on the regional boundaries and diffusion trends, determine whether the points within the region meet the abnormal characteristics, and confirm the final abnormal area. Integrate the test results to generate an anomaly report containing abnormal information, visual charts, and recommended measures.

2. The method for detecting abnormal regions based on multi-temporal point atlases according to claim 1, characterized in that, Based on the detection requirements, multi-temporal core data are collected, and multi-temporal atlases of points are generated through data registration, including: Collect multi-temporal core data according to detection requirements, and perform spatiotemporal alignment and standardization preprocessing for detection tasks; Based on the preprocessed multi-temporal core data, a multi-temporal atlas of points is generated by fusing feature point matching and spatiotemporal registration algorithms.

3. The method for detecting abnormal regions based on multi-temporal point maps according to claim 1, characterized in that, A normal fluctuation model is established based on a multi-temporal atlas of data points, and suspected abnormal data points are marked, including: For each location, analyze its multi-temporal atlas sequence and construct a normal fluctuation model that reflects the normal change pattern and fluctuation range of the location itself; The latest temporal data of each point is compared with its own dynamic benchmark model, and data points that deviate from the normal fluctuation range are automatically identified and marked as suspected abnormal points.

4. The method for detecting abnormal regions based on multi-temporal point maps according to claim 1, characterized in that, Based on suspected anomalous locations, anomaly clustering algorithms are used to calculate the degree of anomalousness of these locations and identify anomalous location clusters, including: Based on the feature data of suspected anomalies, an anomaly clustering algorithm is used to initially group them, and the local anomaly degree value of each point in its group is calculated. Based on the local anomaly values ​​of each point, clusters with high significance are optimized and selected, and these are identified as anomalous point clusters with clear spatial aggregation characteristics.

5. The method for detecting abnormal regions based on multi-temporal point atlases according to claim 1, characterized in that, An energy function is constructed based on the clusters of anomalous points to identify the boundaries of anomalous regions and to preliminarily determine the extent of the anomalous regions, including: By integrating the statistical characteristics and geospatial attributes of anomalous point clusters, an energy function for boundary identification is constructed. By minimizing the energy function to solve for the optimal segmentation boundary, the precise range of the abnormal region can be preliminarily determined.

6. The method for detecting abnormal regions based on multi-temporal point maps according to claim 1, characterized in that, By combining anomalous point clusters and anomalous region boundaries, a diffusion region model is constructed to predict the diffusion trend of anomalous regions, including: Analyze the anomalous intensity gradient within the anomalous region and external environmental factors to determine the diffusion driving factors and parameters; A diffusion model based on spatiotemporal geographic weighted regression is constructed to simulate the diffusion trend and direction of anomalous regions under the influence of driving factors.

7. The method for detecting abnormal regions based on multi-temporal point atlases according to claim 1, characterized in that, Based on the regional boundaries and diffusion trends, determine whether the points within the region meet the abnormal characteristics, and confirm the final abnormal area, including: The spread trend is fed back to the anomaly detection module to enhance the verification of points on potential spread paths; By combining the initial results, boundary and verification information, the final range and confidence level of the anomaly region are optimized and confirmed.

8. A computer-readable storage medium, characterized in that, It includes one or more program instructions, which are executed by a processor as described in any one of claims 1-7, for an anomaly region detection method based on a point-based multi-temporal map atlas.

Images