Multi-source data-based urban deformation anomaly joint identification method and system

Abstract

The invention discloses an urban deformation anomaly joint identification method and system based on multi-source data, and relates to the technical field of pattern recognition, and the method comprises the steps: obtaining InSAR remote sensing data and ground monitoring data for a to-be-identified urban deformation region, constructing multi-source deformation observation data facing structure units, and carrying out the recognition of the multi-source deformation observation data; time sequence alignment and interpolation fusion are carried out, and a time sequence monitoring data set is constructed; a machine learning or deep learning model is adopted to construct a structure deformation anomaly detection model, and a historical data set is adopted for training; inputting the time sequence monitoring data set into the structure deformation anomaly detection model, and extracting to obtain multi-modal features; and performing abnormal deformation identification on the multi-modal features according to an abnormal discrimination logic triggered by multiple conditions, and performing visual display and early warning. Through the technical scheme of the invention, intelligent identification, high-precision analysis and dynamic early warning of the deformation abnormity of the urban structure are realized, and the identification precision and response efficiency of the deformation abnormity of the structure are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pattern recognition, in particular to a city deformation anomaly joint identification method based on multi-source data and a city deformation anomaly joint identification system based on multi-source data. BACKGROUND

[0002] With the acceleration of urbanization, a large number of old buildings, subway stations, bridge foundations and other infrastructure are in complex geological environment and high load state for a long time, and are affected by many factors such as foundation settlement, underground construction, rain erosion, earthquake activity, prone to structural deformation, and further induce safety hazards. How to carry out long-term, continuous and high-precision deformation monitoring and anomaly warning on urban infrastructure has become one of the core problems in urban operation safety management.

[0003] At present, widely used monitoring methods include remote sensing InSAR technology and various ground sensor technologies. InSAR has the advantages of large coverage, long monitoring period and high spatial resolution, and can obtain wide-area ground subsidence information; ground sensors such as GNSS, static leveling instrument, tiltmeter and crack meter can provide high-precision local point information. However, these data sources have significant differences in spatial resolution, time sampling frequency, data accuracy and monitoring range, and most urban monitoring systems have not yet realized effective fusion and joint analysis of these multi-source data.

[0004] The current common urban structure monitoring method is still mainly driven by single-source data, such as relying only on InSAR for macroscopic evaluation, or using settlement meters to monitor local changes, lacking a unified analysis mechanism for "global monitoring-local precision measurement-intelligent fusion". This mode is prone to missed detection, misjudgment or response lag when facing sudden deformation or slow trend evolution, and cannot meet the actual needs of urban structure management and risk identification. SUMMARY

[0005] In view of the above problems, the present application provides a city deformation anomaly joint identification method and system based on multi-source data, which realizes intelligent identification, high-precision analysis and dynamic early warning of city structure deformation anomaly through unified processing and time series fusion analysis of InSAR remote sensing data and ground monitoring data, and multi-modal feature extraction and composite anomaly discrimination strategy based on structure deformation anomaly detection model, and improves the identification accuracy and response efficiency of structure deformation anomaly.

[0006] To achieve the above purpose, the present application provides a city deformation anomaly joint identification method based on multi-source data, comprising: For the to-be-recognized urban deformation region, InSAR remote sensing data and ground monitoring data are acquired, coordinate conversion, time standardization processing and abnormality elimination are performed on different data sources, the InSAR remote sensing data and the ground monitoring data are associated to corresponding structure units based on a spatial index of urban structure, and a multi-source deformation observation data for the structure units is constructed; Time series alignment and interpolation fusion are performed on the multi-source deformation observation data, and a time series monitoring data set of each structure unit is constructed on a unified time axis; A structure deformation anomaly detection model is constructed by using a machine learning or deep learning model, and the structure deformation anomaly detection model is trained by using a historical data set; The time series monitoring data set is input into the structure deformation anomaly detection model, and multi-modal features for representing structure deformation trend change, deformation rate jump and prediction residual are extracted; According to a preset multi-condition triggered anomaly discrimination logic, the multi-modal features are subjected to abnormal deformation recognition; The recognized urban abnormal deformation is visualized and warned.

[0007] In the above technical solution, preferably, InSAR remote sensing data and ground monitoring data are acquired and unified preprocessing of multi-source asynchronous data is performed, and the specific process includes: InSAR remote sensing data and GNSS, static level, crack meter and tiltmeter ground monitoring data are acquired; For the InSAR remote sensing data and the ground monitoring data, a standardized space-time data input framework is established, coordinate conversion, time standardization processing and abnormality elimination are performed on data of different sampling frequencies and spatial accuracies, and remote sensing grid units and ground monitoring points are mapped and aggregated to corresponding structure units through spatial mapping relationship, and unified preprocessing is completed.

[0008] In the above technical solution, preferably, time series alignment and interpolation fusion are performed on the InSAR remote sensing data and the ground monitoring data, and a unified structure time series monitoring data set is constructed, and the specific process includes: Time series alignment and interpolation fusion are performed on the InSAR remote sensing data and the ground monitoring data, and a unified structure time series monitoring data set is constructed, and the specific process includes: A sliding time window is set on a unified time axis, and available observation values of each data source in each time window are used as candidate inputs of the time window; According to data freshness, observation accuracy and data source importance, weight coefficients are assigned to each data source, observation values from different data sources in the same time window are weighted and fused, and deformation representative values corresponding to the time window are obtained; The missing observation values in the time window are filled by interpolation or data-driven interpolation based on reliable data sources, so as to construct a continuous structure unit time series monitoring data set under a unified time step.

[0009] In the technical solution, the time series monitoring data set is input into the structure deformation anomaly detection model to extract multi-modal features, and the specific process includes: The time series monitoring data set of each structure unit is taken as model input in the form of a sliding time window with a preset length, the structure deformation anomaly detection model adopts a time series prediction model and / or a reconstruction model, and the structure deformation anomaly detection model outputs corresponding deformation prediction values and / or reconstruction values for each time window; The deformation prediction values and / or reconstruction values are compared with the actual observation values in the corresponding time window, and prediction residual features for representing model fitting deviation are calculated, and trend change features and rate jump features for representing slow trend, accelerated subsidence or sudden jump behavior are extracted based on the first-order difference and the second-order difference of the deformation variable in the time window; The prediction residual features, trend change features and rate jump features are combined into multi-modal time series features for describing the deformation evolution behavior of the structure unit; The structure deformation anomaly detection model is trained by the historical data set, the multi-modal time series features are pre-labeled in the historical data set, and the structure deformation anomaly detection model is trained to converge to a preset loss function.

[0010] In the technical solution, preferably, according to a preset multi-condition triggered anomaly discrimination logic, the multi-modal features are subjected to abnormal deformation identification, and the specific process includes: The trend change features, rate jump features and prediction residual features are respectively mapped to a standardized index space, and a basic threshold and an alarm threshold are respectively set for each type of feature; When any feature index exceeds the corresponding alarm threshold, the corresponding structure unit is determined to be significantly abnormal deformation; When multiple feature indexes simultaneously exceed their respective basic thresholds but do not reach the alarm thresholds, a comprehensive abnormal score is calculated according to the excess value and the duration of each feature index relative to the basic threshold, and when the comprehensive abnormal score exceeds a preset threshold, the corresponding structure unit is determined to be early abnormal or developing abnormal; According to the comprehensive abnormal score and the excess value and offset value of each feature index relative to the preset threshold, the deformation type of the region where the corresponding structure unit is located is determined.

[0011] In the technical solution, preferably, the identified urban abnormal deformation is visualized and displayed and early warning is given, and the specific process includes: According to the structure unit corresponding to the multi-modal feature of the abnormal deformation, the position of the abnormal structure unit and the abnormal level are spatially visualized and labeled on a GIS platform; According to the deformation type and abnormal level of the abnormal deformation, different marking styles or area filling methods are used for differentiated display on the GIS platform, and deformation time sequence and feature index evolution information are given in attribute information; The abnormal deformation result and the deformation type result are packaged into an early warning message as early warning information through a preset interface, and sent to a preset person and / or a city operation safety management platform.

[0012] The application also provides a city deformation anomaly joint identification system based on multi-source data, which applies the city deformation anomaly joint identification method based on multi-source data disclosed in any of the above technical solutions, comprising: A multi-source data acquisition module is configured to acquire InSAR remote sensing data and ground monitoring data for a city deformation area to be identified, perform coordinate conversion, time standardization processing and abnormality elimination on different data sources, and associate the InSAR remote sensing data and the ground monitoring data to corresponding structure units based on city structure space indexes to construct multi-source deformation observation data for structure units; A unified time sequence alignment module is configured to perform time sequence alignment and interpolation fusion on the multi-source deformation observation data to construct a time sequence monitoring data set of each structure unit on a unified time axis; A model construction and training module is configured to construct a structure deformation anomaly detection model using a machine learning or deep learning model, and train the structure deformation anomaly detection model using a historical data set; An abnormal feature extraction module is configured to input the time sequence monitoring data set into the structure deformation anomaly detection model to extract multi-modal features representing structure deformation trend changes, deformation rate jumps and prediction residuals; An abnormal deformation identification module is configured to identify abnormal deformation of the multi-modal features according to a preset multi-condition triggered abnormality discrimination logic; A deformation display and early warning module is configured to visually display and warn about the identified city abnormal deformation.

[0013] In the above technical solution, preferably, the unified time sequence alignment module is specifically configured to: A sliding time window is set on the unified time axis, and available observation values of each data source in each time window are used as candidate inputs of the time window; According to data freshness, observation accuracy and data source importance, weight coefficients are assigned to each data source, and observation values from different data sources in the same time window are weighted and fused to obtain a deformation representative value corresponding to the time window. The missing observation values in the time window are filled by interpolation or data-driven interpolation based on reliable data sources, so as to construct a continuous structure unit time series monitoring data set under a uniform time step.

[0014] In the technical solution, preferably, the abnormal feature extraction module is specifically configured to: The time series monitoring data set of each structure unit is taken as model input in the form of a sliding time window with a preset length, and the structure deformation anomaly detection model adopts a time series prediction model and / or a reconstruction model, and outputs a corresponding deformation prediction value and / or a reconstruction value for each time window; The deformation prediction value and / or the reconstruction value are compared with the actual observation value in the corresponding time window, and a prediction residual feature for representing model fitting deviation is calculated, and a trend change feature and a rate jump feature for representing a slow trend, accelerated subsidence or sudden jump behavior are extracted based on the first-order difference and the second-order difference of the deformation variable in the time window; The prediction residual feature, the trend change feature and the rate jump feature are combined into a multi-modal time series feature for describing the deformation evolution behavior of the structure unit; The structure deformation anomaly detection model is trained by the historical data set, the multi-modal time series feature is pre-labeled in the historical data set, and the structure deformation anomaly detection model is trained to converge to a preset loss function.

[0015] In the technical solution, preferably, the abnormal deformation identification module is specifically configured to: The trend change feature, the rate jump feature and the prediction residual feature are respectively mapped to a standardized index space, and a basic threshold and an alarm threshold are set for each type of feature; When any feature index exceeds the corresponding alarm threshold, the corresponding structure unit is determined to be significantly abnormal deformation; When multiple feature indexes simultaneously exceed their respective basic thresholds but do not reach the alarm threshold, a comprehensive abnormal score is calculated according to the excess value and the duration of each feature index relative to the basic threshold, and when the comprehensive abnormal score exceeds a preset threshold, the corresponding structure unit is determined to be early abnormal or developing abnormal; According to the comprehensive abnormal score and the excess value and the offset value of each feature index relative to the preset threshold, the deformation type of the region where the corresponding structure unit is located is determined.

[0016] Compared with the prior art, the beneficial effects of the present application are: through unified processing and time sequence fusion analysis of InSAR remote sensing data and ground monitoring data, through multi-modal feature extraction and composite anomaly discrimination strategy based on a structure deformation anomaly detection model, intelligent identification, high-precision analysis and dynamic early warning of urban structure deformation anomalies are realized, and the identification precision and response efficiency of structure deformation anomalies are improved. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A flowchart of a method for jointly identifying urban deformation anomalies based on multi-source data according to an embodiment of the present application is disclosed. Figure 2 A functional structure diagram of a system for jointly identifying urban deformation anomalies based on multi-source data according to an embodiment of the present application is disclosed. DETAILED DESCRIPTION

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in connection with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0019] The present application will be described in further detail below in connection with the drawings: As shown in the drawings, according to a method for jointly identifying urban deformation anomalies based on multi-source data provided by the present application, the method comprises: Figure 1 For a city deformation area to be identified, InSAR remote sensing data and ground monitoring data are obtained and unified preprocessed. The InSAR remote sensing data and the ground monitoring data are time series aligned and interpolated and fused, and a time series monitoring data set of unified structures is constructed. A machine learning or deep learning model is used to construct a structure deformation anomaly detection model, and the structure deformation anomaly detection model is trained using a historical data set. The time series monitoring data set is input into the structure deformation anomaly detection model, and multi-modal features are extracted. According to a pre-set multi-condition triggered anomaly discrimination logic, the multi-modal features are subjected to anomaly deformation identification. The identified urban anomaly deformation is visualized and displayed and early warning is performed. The identified urban anomaly deformation is visualized and displayed and early warning is performed.

[0020] In this embodiment, through unified processing and time series fusion analysis of InSAR remote sensing data and ground monitoring data, through multi-modal feature extraction and composite anomaly discrimination strategy based on structural deformation anomaly detection model, intelligent identification, high-precision analysis and dynamic early warning of structural deformation anomalies such as urban houses, bridges and subway stations are realized, and the identification accuracy and response efficiency of structural deformation anomalies are improved.

[0021] Specifically, the method can fuse multi-source asynchronous monitoring data, adapt to data consistency processing under different time scales and spatial scales, and has anomaly identification capability to improve the intelligent and systematic level of urban infrastructure safety monitoring.

[0022] In the above embodiment, preferably, InSAR remote sensing data and ground monitoring data are acquired and unified preprocessing of multi-source asynchronous data is performed, and the specific process includes unified spatial reference and time reference processing of radar line-of-sight deformation variables of InSAR remote sensing data and point deformation observation values of ground monitoring data.

[0023] InSAR remote sensing data and GNSS, static level, crack meter and inclinometer ground monitoring data are acquired; By establishing a spatial index of urban structures, the InSAR grid cells are projected onto the structural units such as buildings, bridges and subway stations in the unified coordinate system, and the buffer zone or weighted average method is used to aggregate the multiple InSAR grid observation values covering the same structural unit into the remote sensing deformation time series of the structural unit; For InSAR remote sensing data and ground monitoring data, a standardized spatiotemporal data input framework is established to perform coordinate conversion, time standardization processing and anomaly rejection on data with different sampling frequencies and spatial accuracies. Specifically, the monitoring points of GNSS, static level, crack meter and inclinometer are mapped to the corresponding structural units, and through the spatial mapping relationship from points to structural units, a multi-source deformation observation time series is constructed for the structural units. For data with different sampling periods, a unified timestamp standard is used to align the original observation time to a unified time axis, completing the unified preprocessing and providing a basis for subsequent sliding window interpolation and dynamic weighted fusion.

[0024] Specifically, InSAR remote sensing data and ground monitoring data are uniformly preprocessed to realize fusion analysis of multi-source data, which can perform multi-source comparison analysis on deformation anomalies of urban structures and improve the identification accuracy of structural deformation anomalies.

[0025] In the above embodiment, preferably, the InSAR remote sensing data and ground monitoring data are time series aligned and interpolated and fused to construct a unified structural time series monitoring data set, and the specific process includes: The InSAR remote sensing data and the ground monitoring data are aligned in time sequence by using a sliding window interpolation and a dynamic weighting algorithm to construct a time sequence monitoring data set of a unified structure.

[0026] The unified time sequence alignment module introduces a sliding time window and a dynamic weight distribution mechanism according to the characteristics that the InSAR remote sensing data has a long time interval and the ground monitoring data has a fine time granularity. For the unified time axis of the target structure unit, the InSAR observation values and the ground monitoring observation values available in each time window are collected, the weight coefficients are calculated according to the data freshness, the observation accuracy and the importance of the data source, the observation values of each data source in the window are weighted and fused to obtain the deformation representative value corresponding to the time window, and the missing data points in the window are filled by interpolation or data-driven interpolation based on the reliable data source, so as to form a time sequence monitoring data set that is continuous, smooth, and takes into account the wide-area and local accuracy under a unified time step.

[0027] Specifically, the time sequence aligned data can generate a complete and continuous fused time sequence, and solve the problem of different synchronization and inconsistent density.

[0028] In the above embodiment, preferably, the time sequence monitoring data set is input into a structure deformation anomaly detection model to extract multi-modal features, and the specific process includes: The time sequence monitoring data set is input into the structure deformation anomaly detection model to extract multi-modal time sequence features of trend change, rate jump and prediction residual from the fused time sequence data. The structure deformation anomaly detection model is trained by using a historical data set, the multi-modal time sequence features are pre-labeled in the historical data set, and the structure deformation anomaly detection model is trained until a preset loss function converges.

[0029] Specifically, the structural deformation anomaly detection model adopts the form of a time series prediction model and / or a reconstruction model, and the time series monitoring data set of the unified structure is divided into input samples according to a preset length of a sliding time window. For each structure unit, the fused deformation time series of the unit is input in a given time window, and the model outputs the deformation prediction value at the next time or subsequent time and / or the reconstruction value of the current window. By comparing the model prediction value or the reconstruction value with the actual observation value, the prediction residual feature for characterizing the fitting deviation of the model can be calculated; at the same time, according to the deformation amount trend and the first-order and second-order difference results in the time window, the trend change and rate jump features for characterizing the slow subsidence, stage acceleration or sudden jump behavior are extracted. The above prediction residual feature, trend change and rate jump feature jointly constitute a multi-modal time series feature, which is used for subsequent anomaly discrimination logic. In the implementation process, the time series prediction model can adopt a deep learning structure based on a recurrent neural network or a convolutional neural network, or adopt a machine learning model such as a gradient boosting tree.

[0030] In this embodiment, the structural deformation anomaly detection model based on a machine learning or deep learning model can identify and extract the trend change, rate jump and prediction residual features in the data in the input time series monitoring data set through training of the historical data set. Through the pattern recognition ability of the machine learning or deep learning model, the recognition and extraction accuracy of the multi-modal time series features is improved.

[0031] In the above embodiment, preferably, according to the preset multi-condition triggered anomaly discrimination logic, the multi-modal features are subjected to abnormal deformation recognition, and the specific process includes: The anomaly discrimination conditions based on trend mutation, threshold overrun and residual deviation are used to recognize the abnormal deformation of the multi-modal features; If the trend mutation, threshold overrun or residual deviation of the multi-modal features exceeds the preset threshold, it is determined that there is abnormal deformation in the region where the corresponding features are located; According to the exceeding value and the offset value of the multi-modal features relative to the preset threshold, the deformation type of the region where the corresponding features are located is determined.

[0032] Specifically, the abnormal deformation identification module maps the trend change feature, the rate jump feature and the prediction residual feature to a standardized index space, sets a basic threshold and an alarm threshold for each type of feature, and constructs a multi-condition composite triggering mechanism based on a logical combination rule. On the one hand, when any feature index exceeds the corresponding alarm threshold, it is determined that there is a significant abnormality, so as to avoid missing the sudden deformation. On the other hand, when multiple feature indexes exceed their respective basic thresholds but do not reach the alarm thresholds, the system calculates a comprehensive abnormality score according to the threshold exceeding amplitude and the duration, and determines that the structure unit is an early abnormality or a developing abnormality when the comprehensive abnormality score exceeds a set threshold. Through the above combination of single-index strong abnormality triggering and multi-index weak abnormality accumulation triggering, the sensitivity of deformation abnormality identification is taken into account, and the robustness and reliability of the results are improved.

[0033] In the implementation process, the multi-condition composite abnormality discrimination method is used to identify the abnormal deformation of the extracted multi-modal features, which can comprehensively consider various factors and improve the accuracy and robustness of the abnormal deformation identification results. At the same time, according to the abnormal amount or offset value exceeding the threshold, i.e. the abnormality degree, the type of deformation abnormality can be determined.

[0034] In the above embodiment, preferably, the identified urban abnormal deformation is visualized and warned, and the specific process includes: According to the region where the multi-modal feature with abnormal deformation is located, the visualization labeling is performed on the GIS platform; According to the deformation type of the abnormal deformation, different ways are used for labeling on the GIS platform; The abnormal deformation result and the deformation type result are sent to the preset personnel as warning information through the preset interface.

[0035] In this embodiment, the deformation display and warning module spatially associates the abnormality identification result with the urban basic geographic information data, superimposes the abnormality level, abnormality type and evolution process in the time dimension of the structure unit on the GIS map service in the form of point markers, area coloring or dynamic time axis, etc. For the structure unit exceeding the high-level warning threshold, the structure unit is highlighted or marked with special symbols on the map interface, and the deformation time sequence curve, feature index evolution curve, comprehensive abnormality score and recommended on-site review suggestion are given in the attribute panel. At the same time, the system pushes the warning message including the structure location information, abnormality level, occurrence time and brief description to the city operation safety management platform through the interface, to support cross-platform linkage response.

[0036] According to the city deformation anomaly joint identification method based on multi-source data disclosed in the above-embodiment, the functions are integrated to construct a complete system architecture including data acquisition, processing, analysis, early warning and display. At the same time, through the GIS platform, the region with abnormal deformation in the city deformation region to be identified is visually labeled. For different types of abnormal deformation, different labeling methods can be used, such as different color labeling, different identification labeling, different brightness labeling, etc. In addition, the deformation state, risk level and abnormal evolution process of the structure can also be dynamically displayed on the GIS platform, which is convenient for supervisors to carry out research and response.

[0037] As shown in Figure 2 The present application also provides a city deformation anomaly joint identification system based on multi-source data, which applies the city deformation anomaly joint identification method based on multi-source data disclosed in any one of the above-embodiments, and comprises: A multi-source data acquisition module 1 is configured to acquire InSAR remote sensing data and ground monitoring data for a city deformation region to be identified and perform unified preprocessing on multi-source asynchronous data. A unified time sequence alignment module 2 is configured to perform time sequence alignment and interpolation fusion on the InSAR remote sensing data and the ground monitoring data, and construct a unified time sequence monitoring data set of the structure. A model construction and training module 3 is configured to construct a structure deformation anomaly detection model by using a machine learning or deep learning model, and train the structure deformation anomaly detection model by using a historical data set. An abnormal feature extraction module 4 is configured to input the time sequence monitoring data set into the structure deformation anomaly detection model, and extract multi-modal features. An abnormal deformation identification module 5 is configured to perform abnormal deformation identification on the multi-modal features according to a preset multi-condition triggered abnormality discrimination logic. A deformation display and early warning module 6 is configured to visually display and early warn the identified city abnormal deformation.

[0038] The city deformation anomaly joint identification system based on multi-source data comprises a multi-source data acquisition module, a unified time sequence alignment module, a model construction training module, an anomaly feature extraction module, an anomaly deformation identification module, and a deformation display and early warning module. The modules are sequentially connected by data flow to form a complete data link of data acquisition-fusion processing-anomaly detection-visualization and early warning. The multi-source data acquisition module is configured to acquire multi-source asynchronous deformation observation data from an InSAR monitoring platform and a ground monitoring network, and output data streams after unified preprocessing. The unified time sequence alignment module is configured to perform space-time integration alignment and interpolation fusion on the multi-source data, and output a time sequence monitoring data set for a unified structure unit. The model construction training module is configured to train a structure deformation anomaly detection model based on a historical time sequence monitoring data set and labeled samples. The anomaly feature extraction module is configured to automatically extract multi-modal time sequence features such as trend changes, rate jumps, and prediction residuals from the time sequence monitoring data set based on the trained structure deformation anomaly detection model. The anomaly deformation identification module is configured to perform a composite anomaly discrimination logic to give a deformation anomaly region and an anomaly type. The deformation display and early warning module is responsible for spatial visualization of the anomaly identification result on a GIS platform and pushing early warning information to relevant persons in charge.

[0039] In the above embodiment, preferably, the unified time sequence alignment module 2 is specifically configured to: align the InSAR remote sensing data and the ground monitoring data in time sequence by using a sliding window interpolation and a dynamic weighting algorithm, and construct a time sequence monitoring data set of a unified structure.

[0040] In the above embodiment, preferably, the anomaly feature extraction module 4 is specifically configured to: input the time sequence monitoring data set into the structure deformation anomaly detection model to extract multi-modal time sequence features such as trend changes, rate jumps, and prediction residuals from the fused time sequence data; the structure deformation anomaly detection model is trained by a historical data set, the multi-modal time sequence features are pre-labeled in the historical data set, and the structure deformation anomaly detection model is trained to converge to a preset loss function.

[0041] In the above embodiment, preferably, the anomaly deformation identification module 5 is specifically configured to: perform anomaly deformation identification on the multi-modal features based on anomaly discrimination conditions of trend mutation, threshold overrun, and residual deviation; if the trend mutation, threshold overrun, or residual deviation of the multi-modal features exceeds a preset threshold, it is determined that there is an anomaly deformation in a region where the corresponding features are located; determine the deformation type of the region where the corresponding features are located according to the exceeding value and the deviation value of the multi-modal features relative to the preset threshold.

[0042] The functions to be achieved by each module of the city deformation anomaly joint recognition system based on multi-source data disclosed in the above embodiments correspond to the respective steps of the city deformation anomaly joint recognition method based on multi-source data disclosed in the above embodiments respectively, and in the implementation process, the above embodiments are referred to for operation, and details are not repeated here.

[0043] The city deformation anomaly joint recognition method and system based on multi-source data disclosed in the above embodiments not only solve the inconsistency problems of multi-source monitoring data in time, accuracy and spatial distribution, but also break through the bottleneck of traditional anomaly recognition methods in accuracy, adaptability and responsiveness, and have good algorithm innovation and engineering implementability. The system architecture is clear and complete, and is suitable for embedded deployment of the city operation safety monitoring platform, and has wide popularization and application value and industrial transformation potential.

[0044] The application can be widely applied to city old community house safety monitoring, subway structure settlement evaluation, underground space operation and maintenance management, city physical examination and risk census, and provides key technical support for building an all-weather, multi-scale and intelligent city infrastructure safety monitoring system.

[0045] The above is only the preferred embodiment of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for joint identification of urban deformation anomalies based on multi-source data, characterized in that, include: For the urban deformation area to be identified, InSAR remote sensing data and ground monitoring data are acquired. Coordinate transformation, time standardization and anomaly removal are performed on different data sources. Based on the urban structure spatial index, the InSAR remote sensing data and ground monitoring data are associated with the corresponding structural units to construct multi-source deformation observation data for structural units. The multi-source deformation observation data are aligned and interpolated over time to construct a time-series monitoring dataset for each structural unit on a unified time axis. A structural deformation anomaly detection model is constructed using machine learning or deep learning models, and the structural deformation anomaly detection model is trained using historical datasets; The time-series monitoring dataset is input into the structural deformation anomaly detection model to extract multimodal features that characterize structural deformation trend changes, deformation rate jumps, and prediction residuals. Based on the preset multi-condition triggered anomaly discrimination logic, abnormal deformation identification is performed on the multimodal features; The identified urban anomalies are visualized and given early warning.

2. The method for joint identification of urban deformation anomalies based on multi-source data according to claim 1, characterized in that, The process of acquiring InSAR remote sensing data and ground monitoring data and performing unified preprocessing of multi-source asynchronous data includes: Acquire InSAR remote sensing data as well as ground monitoring data from GNSS, hydrostatic level, crack gauge and inclinometer; For the InSAR remote sensing data and the ground monitoring data, a standardized spatiotemporal data input framework is established. Coordinate transformation, time standardization and anomaly removal are performed on data with different sampling frequencies and spatial precisions. The remote sensing grid units and ground monitoring points are mapped and aggregated to the corresponding structural units through spatial mapping relationships to complete unified preprocessing.

3. The method for joint identification of urban deformation anomalies based on multi-source data according to claim 2, characterized in that, The InSAR remote sensing data and the ground monitoring data are time-series aligned and interpolated to construct a unified time-series monitoring dataset for structures. The specific process includes: Set up a sliding time window on a unified time axis, and use the available observations of each data source within each time window as candidate inputs for that time window; Weight coefficients are assigned to each data source based on data freshness, observation accuracy, and data source importance. Observations from different data sources within the same time window are weighted and fused to obtain the representative deformation value corresponding to the time window. For missing observations within the time window, interpolation or data-driven interpolation based on reliable data sources is used to fill in the missing data, thereby constructing a continuous time-series monitoring dataset of structural units under a unified time step.

4. The method for joint identification of urban deformation anomalies based on multi-source data according to claim 3, characterized in that, The time-series monitoring dataset is input into the structural deformation anomaly detection model to extract multimodal features. The specific process includes: The time-series monitoring dataset of each structural unit is used as the model input in the form of a sliding time window of a preset length. The structural deformation anomaly detection model adopts a time series prediction model and / or reconstruction model, and outputs the corresponding deformation prediction value and / or reconstruction value for each time window. The predicted deformation value and / or reconstructed value are compared with the actual observed value within the corresponding time window to calculate the predicted residual feature used to characterize the model fitting deviation. Based on the first and second differences of the deformation within the time window, trend change features and rate jump features used to characterize the gradual trend, accelerated sinking or sudden jump behavior are extracted. The predicted residual features, trend change features, and rate jump features are combined into multimodal temporal features to describe the deformation evolution behavior of structural units; The structural deformation anomaly detection model is trained using the historical dataset, which contains pre-labeled multimodal temporal features, and the model is trained until the preset loss function converges.

5. The method for joint identification of urban deformation anomalies based on multi-source data according to claim 4, characterized in that, Based on a preset multi-condition triggered anomaly detection logic, abnormal deformation identification is performed on the multimodal features. The specific process includes: The trend change characteristics, rate jump characteristics, and prediction residual characteristics are mapped to the standardized index space, and a basic threshold and an alarm threshold are set for each type of characteristic. When any characteristic indicator exceeds the corresponding alarm threshold, the corresponding structural unit will be judged as having significant abnormal deformation. When multiple feature indicators exceed their respective basic thresholds but do not reach the alarm threshold, a comprehensive anomaly score is calculated based on the excess value and duration of each feature indicator relative to the basic threshold. When the comprehensive anomaly score exceeds a preset threshold, the corresponding structural unit is determined to be an early anomaly or a developing anomaly. Based on the comprehensive anomaly score and the excess and offset values ​​of each feature index relative to the preset threshold, the deformation type of the region where the corresponding structural unit is located is determined.

6. The method for joint identification of urban deformation anomalies based on multi-source data according to claim 5, characterized in that, The process of visualizing and issuing early warnings for identified urban anomalies includes: Based on the structural units corresponding to the multimodal features with abnormal deformation, the locations and anomaly levels of the abnormal structural units are spatially visualized and labeled on the GIS platform. Based on the deformation type and anomaly level of the abnormal deformation, different marking styles or area coloring methods are used to distinguish and display them on the GIS platform, and deformation time series and characteristic index evolution information are given in the attribute information. The abnormal deformation results and deformation type results are packaged into an early warning message through a preset interface and sent to preset personnel and / or the city operation safety management platform.

7. A joint identification system for urban deformation anomalies based on multi-source data, characterized in that, The method for joint identification of urban deformation anomalies based on multi-source data as described in any one of claims 1 to 6 includes: The multi-source data acquisition module is used to acquire InSAR remote sensing data and ground monitoring data for the urban deformation area to be identified, perform coordinate transformation, time standardization processing and anomaly removal on different data sources, and associate the InSAR remote sensing data and ground monitoring data with the corresponding structural units based on the urban structure spatial index to construct multi-source deformation observation data for structural units. A unified time-series alignment module is used to perform time-series alignment and interpolation fusion on the multi-source deformation observation data, and to construct a time-series monitoring dataset for each structural unit on a unified time axis. The model building and training module is used to build a structural deformation anomaly detection model using machine learning or deep learning models, and to train the structural deformation anomaly detection model using historical datasets. An anomaly feature extraction module is used to input the time-series monitoring dataset into the structural deformation anomaly detection model and extract multimodal features to characterize structural deformation trend changes, deformation rate jumps, and prediction residuals. An abnormal deformation identification module is used to identify abnormal deformations of the multimodal features according to a preset multi-condition triggered abnormal discrimination logic. The deformation display and early warning module is used to visualize and issue early warnings for identified abnormal urban deformations.

8. The urban deformation anomaly joint identification system based on multi-source data according to claim 7, characterized in that, The unified timing alignment module is specifically used for: Set up a sliding time window on a unified time axis, and use the available observations of each data source within each time window as candidate inputs for that time window; Weight coefficients are assigned to each data source based on data freshness, observation accuracy, and data source importance. Observations from different data sources within the same time window are weighted and fused to obtain the representative deformation value corresponding to the time window. For missing observations within the time window, interpolation or data-driven interpolation based on reliable data sources is used to fill in the missing data, thereby constructing a continuous time-series monitoring dataset of structural units under a unified time step.

9. The urban deformation anomaly joint identification system based on multi-source data according to claim 7, characterized in that, The anomaly feature extraction module is specifically used for: The time-series monitoring dataset of each structural unit is used as the model input in the form of a sliding time window of a preset length. The structural deformation anomaly detection model adopts a time series prediction model and / or reconstruction model, and outputs the corresponding deformation prediction value and / or reconstruction value for each time window. The predicted deformation value and / or reconstructed value are compared with the actual observed value within the corresponding time window to calculate the predicted residual feature used to characterize the model fitting deviation. Based on the first and second differences of the deformation within the time window, trend change features and rate jump features used to characterize the gradual trend, accelerated sinking or sudden jump behavior are extracted. The predicted residual features, trend change features, and rate jump features are combined into multimodal temporal features to describe the deformation evolution behavior of structural units; The structural deformation anomaly detection model is trained using the historical dataset, which contains pre-labeled multimodal temporal features, and the model is trained until the preset loss function converges.

10. The urban deformation anomaly joint identification system based on multi-source data according to claim 7, characterized in that, The abnormal deformation identification module is specifically used for: The trend change characteristics, rate jump characteristics, and prediction residual characteristics are mapped to the standardized index space, and a basic threshold and an alarm threshold are set for each type of characteristic. When any characteristic indicator exceeds the corresponding alarm threshold, the corresponding structural unit will be judged as having significant abnormal deformation. When multiple feature indicators exceed their respective basic thresholds but do not reach the alarm threshold, a comprehensive anomaly score is calculated based on the excess value and duration of each feature indicator relative to the basic threshold. When the comprehensive anomaly score exceeds a preset threshold, the corresponding structural unit is determined to be an early anomaly or a developing anomaly. Based on the comprehensive anomaly score and the excess and offset values ​​of each feature index relative to the preset threshold, the deformation type of the region where the corresponding structural unit is located is determined.

Images