Multi-sensor big data fusion dynamic intelligent detection method, device and system

Abstract

The application discloses a kind of multi-sensing big data fusion dynamic intelligent detection method, device and system.Therein, the method comprises: obtaining the positioning data and original graph of the underground engineering of target area;Using positioning data to generate and original graph pixel level matching space mask, according to the mask partition of space mask, original graph is partitioned and adaptively filtered and denoised, and target graph is obtained;Using target graph, ground micro-deformation data of fiber strain sensing module, micro-chip density detection data of monitoring unit in target area, four kinds of data such as stratum stress and water content data, determine the collapse hidden danger type of target area;In the case where the collapse hidden danger type of target area is empty, issue a repair instruction to indicate that the target area is filled with material by drilling.This application solves the technical problem that the detection accuracy of underground cavity and collapse hidden danger in related art is low.

Description

Technical Field

[0001] This application relates to the field of soil safety assessment, and more specifically, to a multi-sensor big data fusion dynamic intelligent detection method, device, and system. Background Technology

[0002] This section is intended to provide background or context for the content set forth in the claims or specification, and the content described herein is not acknowledged as prior art simply because it is included in this section.

[0003] With the continuous development of urban underground space, the construction and operation of underground projects (such as subway tunnels, integrated utility tunnels, underground parking lots, and commercial streets) and areas with dense aging pipeline networks (such as old urban areas and core business districts) are increasing. In these underground engineering environments, water supply, drainage, gas, electricity, and communication pipelines are often intertwined, with burial depths typically ranging from 0.5m to 6m and minimum spacing less than 0.3m. Affected by factors such as pipeline aging and leakage, construction disturbance, and dynamic changes in groundwater, these factors can easily induce loosening of the surrounding soil and the development of cavities, leading to geological disasters such as ground subsidence. These risks have become a key area of ​​concern for the safe operation and disaster prevention and mitigation of urban underground engineering projects.

[0004] Currently, the detection of underground cavities and potential collapse hazards along urban roads mainly relies on mobile surveys using vehicle-mounted ground-penetrating radar (GPR). GPR technology is widely used for the initial screening of underground cavities in urban roads due to its high detection efficiency and wide coverage. However, the complexity of urban underground spaces leads to relatively low accuracy in detecting underground cavities and potential collapse hazards.

[0005] There is currently no effective solution to the above problems. Summary of the Invention

[0006] This application provides a multi-sensor big data fusion dynamic intelligent detection method, device, and system to at least solve the technical problem of low detection accuracy of underground cavities and collapse hazards in related technologies.

[0007] According to one aspect of the embodiments of this application, a multi-sensor big data fusion dynamic intelligent detection method is provided, comprising: obtaining positioning data and original ground-penetrating radar maps obtained by an electromagnetic positioning module of a target area through a detection vehicle; generating a spatial mask that matches the original map at the pixel level using the positioning data; performing partitioned adaptive filtering and denoising on the original map according to the mask partitions of the spatial mask to obtain a target map, wherein the partitioned adaptive filtering and denoising is used to eliminate the influence of the reflected signals of the underground project on the identification of soil collapse hazards, and the spatial mask is used to characterize the underground project; determining the collapse hazard type of the target area using four types of data: the target map, ground micro-deformation data of fiber optic strain sensing modules in the target area, micro-core density detection data of monitoring units in the target area, and stratum stress and moisture content data; the monitoring units are installed in the target area by drilling at fixed points; and issuing a repair command when the collapse hazard type in the target area is a cavity, instructing to fill the cavity in the target area with material by drilling.

[0008] According to another aspect of the embodiments of this application, a multi-sensor big data fusion dynamic intelligent detection device is also provided, comprising: an acquisition unit, used to acquire positioning data obtained by an electromagnetic positioning module of the target area to locate underground engineering and the original map of ground penetrating radar through a detection vehicle; a denoising unit, used to generate a spatial mask that matches the original map at the pixel level using the positioning data, and to perform partitioned adaptive filtering denoising on the original map according to the mask partition of the spatial mask to obtain a target map, wherein the partitioned adaptive filtering denoising is used to eliminate the influence of the reflection signal of the underground engineering on the identification of soil collapse hazards, and the spatial mask is used to characterize the underground engineering; a processing unit, used to determine the collapse hazard type of the target area using four types of data: the target map, the ground micro-deformation data of the fiber optic strain sensing module, the micro-core density detection data of the monitoring unit, and the stratum stress and water content data, wherein the monitoring unit is installed in the target area by drilling a fixed point; and a repair unit, used to issue a repair command when the collapse hazard type of the target area is a cavity, to instruct the filling of the cavity in the target area with material by drilling.

[0009] According to another aspect of the embodiments of this application, a multi-sensor big data fusion dynamic intelligent detection system is also provided, including: a detection vehicle, used to acquire positioning data obtained by an electromagnetic positioning module for locating underground engineering and the original map of ground penetrating radar when passing through a target area; a computer device, used to generate a spatial mask that matches the original map at the pixel level using the positioning data, and to perform partitioned adaptive filtering and denoising on the original map according to the mask partition of the spatial mask to obtain a target map, wherein the partitioned adaptive filtering and denoising is used to eliminate the influence of the reflected signal of the underground engineering on the identification of soil collapse hazards, and the spatial mask is used to characterize the underground engineering; using four types of data, namely the target map, ground micro-deformation data of the fiber optic strain sensing module, micro-core density detection data of the monitoring unit, and stratum stress and water content data, to determine the type of collapse hazard in the target area, wherein the monitoring unit is installed in the target area by drilling a fixed point; and a repair device, used to fill the cavity in the target area with material through drilling when the collapse hazard type in the target area is a cavity.

[0010] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, the storage medium including a stored program that executes the above-described method when the program is run.

[0011] According to another aspect of the embodiments of this application, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor performs the above-described method through the computer program.

[0012] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of any of the embodiments of the methods described above.

[0013] In this embodiment, positioning data and the original ground-penetrating radar map are obtained by using an electromagnetic positioning module to locate underground engineering in the target area via a detection vehicle. A spatial mask matching the original map at the pixel level is generated using the positioning data. Based on the mask partitioning of the spatial mask, the original map is subjected to partitioned adaptive filtering and denoising to obtain the target map. Partitioned adaptive filtering and denoising is used to eliminate the influence of reflected signals from underground engineering on the identification of soil collapse hazards, and the spatial mask is used to characterize the underground engineering. Using four types of data—the target map, ground micro-deformation data from fiber optic strain sensing modules within the target area, micro-core density detection data from monitoring units within the target area, and ground stress and moisture content data—the type of collapse hazard in the target area is determined. The monitoring units are installed in the target area using a borehole-based fixed-point method. If the collapse hazard type in the target area is a cavity, a repair command is issued to instruct the filling of the cavity with material through drilling. This solves the technical problem of low detection accuracy of underground cavities and collapse hazards in related technologies. Attached Figure Description

[0014] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of a multi-sensor big data fusion dynamic intelligent detection system according to an embodiment of this application; Figure 2 This is a schematic diagram of the sensor layout of the vehicle-mounted mobile detection unit according to an embodiment of this application; Figure 3 This is a schematic diagram of the hierarchical nested structure of the monitoring unit according to an embodiment of this application; Figure 4 This is a flowchart of a multi-sensor big data fusion dynamic intelligent detection method according to an embodiment of this application; Figure 5 This is a schematic diagram of a partitioned adaptive filtering based on a pipeline space mask according to an embodiment of this application; Figure 6 This is an improvement D according to the embodiments of this application. A schematic diagram of the S-evidence theory fusion scheme; Figure 7 This is a schematic diagram of a closed-loop dynamic detection mechanism according to an embodiment of this application; Figure 8 This is a schematic diagram of a multi-sensor big data fusion dynamic intelligent detection device according to an embodiment of this application; Figure 9 This is a structural block diagram of a terminal according to an embodiment of this application. Detailed Implementation

[0015] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0016] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0017] After researching the relevant technologies, the inventors discovered that: 1) Mobile survey and detection based on vehicle-mounted ground-penetrating radar: For example, the method for detecting and identifying underground cavities in urban areas uses vehicle-mounted 3D ground-penetrating radar and GPS to collect data on main urban roads, analyze geological cavity values, and then assess the underground safety index and surrounding building load index of abnormal main roads to achieve graded detection planning. Road detection methods based on ground-penetrating radar preprocess and feature extraction (including amplitude, frequency, phase, phase axis continuity, and multiple reflection characteristics) of radar image data, and use a road collapse risk assessment model to determine the target risk level.

[0018] 2) Precise Monitoring Based on Borehole-Embedded Sensors: For identified high-risk areas, continuous monitoring using borehole-embedded sensors can acquire the micromechanical parameters of the soil. For example, in an intelligent monitoring and early warning system for road collapse in collapsible loess areas, sensors such as humidity meters, displacement gauges, pore water pressure gauges, and earth pressure gauges are pre-embedded in roads or key monitoring areas, combined with a 3S integrated system (remote sensing, geographic information system, global positioning system) and a ground radar system for dual-control monitoring and early warning. In embedded automated monitoring and early warning devices for road collapse, a combination of anchoring structures, probes or ropes, and stress detection components is used to monitor changes in prestress and provide early warning of loosening of the underlying soil layer.

[0019] 3) Multi-source data fusion and information management: With the development of IoT and AI technologies, multi-source data fusion and information management can be adopted. For example, in a road collapse risk management information system, an IoT monitoring subsystem, a health status subsystem, and a 3D GIS subsystem can be integrated to achieve integrated daily management, proactive early warning, and emergency response. In urban road collapse monitoring methods based on vehicle-road-air integration, multi-platform collaborative detection using satellite remote sensing, integrated detection vehicles, drones, and ground-penetrating radar is employed.

[0020] Analysis reveals the following clear technical shortcomings in the dynamic detection of ground subsidence in complex areas of urban underground engineering: 1) Insufficient interference removal capability and high false alarm rate: The signal processing of the ground-penetrating radar detection scheme does not take into account prior information such as the spatial location, type, burial depth, and material of the complex characteristics of the pipeline network for targeted processing. For example, in urban areas with complex pipeline networks, the strong reflections of metal pipelines, the weak reflections of non-metallic pipelines, and the reflection signals of soil defects (looseness, cavities) highly overlap in the time and frequency domains. General denoising methods cannot effectively distinguish between pipeline reflection signals and soil defect signals, resulting in a false alarm rate of more than 35% for defect identification in densely populated pipeline areas. In addition, it does not establish a secondary verification mechanism for interference removal effectiveness, making it difficult to guarantee the reliability of the processing results.

[0021] 2) Low accuracy of multi-source data fusion and weak ability to handle conflicting evidence: A fixed-weight weighted averaging method is used for multi-source data. For example, in urban pipeline network detection scenarios, macroscopic detection data from vehicle-mounted ground-penetrating radar and microscopic measured data from borehole sensors often conflict due to differences in detection scale, physical principles, and environmental interference (e.g., radar at the same location shows "loose" while a density sensor shows "dense"). When dealing with such highly conflicting evidence, the classic DS evidence theory is prone to distortion and even paradoxes in the fusion results. It has not improved the evidence theory to suit the characteristics of this scenario, nor has it introduced evidence credibility assessment and dynamic weight adjustment mechanisms, resulting in fusion results that fail to meet engineering requirements in terms of accuracy.

[0022] 3) The detection process is an open-loop model, lacking a data reverse correction mechanism: The above detection scheme is an open-loop, step-by-step process of "vehicle-mounted general survey → borehole detailed survey". The borehole measurement data is only used to verify or correct the judgment of currently identified hazards, and is not input back into the inversion model of the vehicle-mounted ground-penetrating radar. Therefore, dynamic calibration of radar detection parameters (such as dielectric constant and wave velocity) cannot be achieved. Consequently, hazards that are not identified due to interference in the initial detection are still difficult to effectively detect in subsequent re-detections, and the overall detection accuracy cannot be iteratively improved through the accumulation of measured data.

[0023] 4) Poor spatiotemporal synchronization of multi-sensor data and weak fusion foundation: In vehicle-mounted detection systems, pipeline positioning modules and ground-penetrating radar modules are mostly deployed independently and collected data in stages, lacking strict synchronization triggering and spatial registration mechanisms. Multi-source data from the same detection location exhibit spatial deviations (≥20cm) and temporal deviations (≥100ms), making it difficult to establish accurate pixel-level mapping relationships during subsequent data fusion. Furthermore, the borehole monitoring unit and the vehicle-mounted detection unit lack a unified timestamp synchronization benchmark, limiting the reliability of multi-source data fusion.

[0024] 5) The method for constructing the disease feature database is simplistic and the sample coverage is insufficient: Although the above techniques use feature extraction and machine learning for disease identification, they do not systematically describe the method for constructing a disease feature database for complex scenarios in pipeline areas. Obtaining actual samples is costly and time-consuming, and samples from a single source are insufficient to cover diverse scenarios with different geological conditions, pipeline types, disease types, and development stages, resulting in insufficient generalization ability of the classification model.

[0025] In summary, the aforementioned technologies suffer from several drawbacks in detecting subsidence in complex urban underground engineering projects, including insufficient ability to eliminate network interference, low accuracy of multi-source data fusion, open-loop detection process, poor spatiotemporal synchronization of data, and a single database construction method. These shortcomings are interconnected and mutually restrictive, resulting in low accuracy in identifying subsidence hazards in densely networked areas (measured false alarm rate > 35%), long detection cycles (typically requiring more than 7 days), and large positioning errors (≥ 1.2m). These limitations make it difficult to meet the engineering requirements for accurate, dynamic, and efficient detection of surface subsidence in complex urban underground pipeline areas. Therefore, there is an urgent need for a dynamic detection device and method that can systematically solve the above problems and achieve multi-sensor collaboration and closed-loop data iteration.

[0026] To address some or all of the shortcomings of the aforementioned technologies, this solution provides a multi-sensor fusion dynamic detection scheme for ground subsidence in urban complex pipeline areas. Through systematic technical means such as collaborative sensor deployment, zoned adaptive filtering, improved DS evidence theory fusion, and closed-loop iteration of vehicle-mounted and borehole data, it achieves effective elimination of pipeline interference, reliable fusion of conflicting evidence, and iterative improvement of detection accuracy, ultimately achieving efficient, accurate, and dynamic detection of subsidence hazards in complex pipeline areas. This solves the following key technical problems in the dynamic detection of ground subsidence in urban complex pipeline areas: 1) Solving the technical problem that ground-penetrating radar detection schemes, due to the lack of targeted signal processing combined with prior pipeline information, cannot effectively distinguish between pipeline reflection signals and soil defect signals, resulting in a high false alarm rate (generally higher than 35%) for defect identification in densely populated pipeline areas; 2) Solving the technical problem that multi-sensor fusion schemes, due to the use of fixed-weight classical DS evidence theory, cannot effectively handle conflicting evidence between vehicle-mounted macroscopic detection data and borehole microscopic measured data, leading to a high distortion rate in the fusion results. 3) Address the technical issue in the open-loop detection process of "vehicle-mounted general survey → borehole detailed survey" where borehole measurement data is only used to verify the survey results and is not used to back-correct the radar inversion model, resulting in the inability to iteratively improve the accuracy of the entire detection area. 4) Resolve the technical issue of poor spatiotemporal synchronization of data (spatial deviation ≥20cm, temporal deviation ≥100ms) caused by independent deployment and step-by-step acquisition of multiple sensors, leading to a weak foundation for multi-source data fusion. 5) Address the technical issue of a single method for constructing the disease feature database, insufficient sample coverage, and difficulty in supporting high-precision disease identification in diverse scenarios within complex pipeline networks.

[0027] According to one aspect of the embodiments of this application, a method embodiment of a multi-sensor big data fusion dynamic intelligent detection method is provided. Optionally, in this embodiment, the above-described multi-sensor big data fusion dynamic intelligent detection method can be applied to the following multi-sensor big data fusion dynamic intelligent detection system: The detection vehicle is used to acquire positioning data and raw ground-penetrating radar maps obtained by the electromagnetic positioning module when passing through the target area to locate underground engineering. Computer equipment (which can be integrated into the detection vehicle or an edge computing device) is used to generate a spatial mask that matches the original map pixel-level using the positioning data. Based on the mask partitions of the spatial mask, the original map is subjected to partitioned adaptive filtering and denoising to obtain the target map. The partitioned adaptive filtering and denoising is used to eliminate the influence of reflected signals from underground engineering on the identification of potential soil collapse hazards. The spatial mask is used to characterize the underground engineering. Using four types of data—the target map, ground micro-deformation data from the fiber optic strain sensing module, micro-core density detection data from the monitoring unit, and ground stress and moisture content data—the type of potential collapse hazard in the target area is determined. The monitoring unit is installed in the target area using a borehole-based fixed-point method. Repair equipment (which may be part of a detection vehicle or a separate device) is used to fill the cavities in the target area with material through drilling when the collapse hazard type in the target area is a cavity.

[0028] See Figure 1 The detection devices involved in this application mainly include: 1) Vehicle-mounted mobile detection unit, such as Figure 2 As shown: This unit is mounted on a detection vehicle (such as a small electric detection vehicle). All sensors are integrated into a rigid mounting platform at the bottom of the detection vehicle, with the platform 10cm-15cm above the ground. The sensors are fixedly arranged in a spatially synchronous linear array along the central axis of the detection vehicle's direction of travel, as detailed below: Electromagnetic positioning modules for underground pipelines and other underground engineering facilities: Deployed at the center of the array, these modules employ dual-channel electromagnetic induction sensors with an operating frequency of 10kHz-1MHz, a planar positioning error ≤5cm, and a burial depth detection error ≤10cm. Their function is to provide prior information such as pipeline type, diameter, burial depth, direction, and geographical coordinates for subsequent radar signal processing. High-frequency impulse ground-penetrating radar modules: Symmetrically deployed on both sides of the array, two units in total, with a center frequency of 1.5GHz, a detection depth of 0-8m, a range resolution ≤2cm, and a sampling frequency of 100GHz. The distance between the electromagnetic positioning modules and the radar modules on both sides is fixed at 12cm. This distance corresponds to half the wavelength of the 1.5GHz radar in soil, ensuring that the spatial synchronization error between pipeline positioning data and radar data at the same underground location is ≤5cm. Data from two radars can be cross-validated to reduce the detection error of a single radar. A distributed fiber optic strain sensing module, consisting of two lines, is deployed between the electromagnetic and radar modules. It has a detection accuracy of 1με and a sampling frequency of 1kHz, and its function is to collect ground micro-deformation data in real time during the detection process to assist in judging the development trend of defects. A GPS / BeiDou positioning module is installed on a rigid position on the top of the detection vehicle, with a planar positioning accuracy of ≤0.5m and an elevation accuracy of ≤1cm. It synchronizes with all sensors and records the geographical coordinates of each collection point. The vehicle-mounted data acquisition terminal has a built-in timestamp synchronization module, supports 8-channel synchronous acquisition, has a storage capacity of 1TB, and uploads data to the backend unit in real time via a wireless transmission module after initial filtering.

[0029] 2) Borehole fixed-point monitoring unit, such as Figure 3As shown: This unit features a layered, nested structure with multiple sensors within the same borehole, suitable for small-diameter boreholes (≤110mm) in urban pipe network areas. The borehole placement is determined by intelligent optimization using a genetic algorithm based on the initial detection results from the vehicle-mounted mobile detection unit. The borehole depth is the maximum burial depth of the pipe network within the detection area +2m, ensuring coverage of potential hazards beneath the pipe network. Borehole spacing: 5m-8m in high-risk areas (pipe network spacing <1m, around old pipelines), and 10m-15m in general risk areas. Sensors within the same borehole are deployed layer by layer from top to bottom: Wellhead Protection and Wireless Transmission Module: Deployed at the borehole head, featuring a corrosion-resistant stainless steel protective shell, and an integrated 4G / 5G wireless transmission module and battery power supply module. Data upload intervals are configurable (10min-1h). Formation Stress Sensor: Deployed in the stratum corresponding to the average burial depth of the pipeline network, one or more units, with a detection accuracy of ±0.01MPa, used to monitor stress changes in the soil surrounding the pipeline network. Groundwater Level and Moisture Content Sensor: Deployed in the regional groundwater level fluctuation layer, one or more units, with a water level detection accuracy of ±1cm and a moisture content detection accuracy of ±0.5%, used to monitor the impact of hydrological conditions on soil stability. Micro-core Compaction Detection Module: Deployed at the bottom of the borehole (the hidden layer below the pipeline network), one or more sets, integrating a torque meter, thrust sensor, and displacement gauge. Soil compaction resolution thickness ≤20mm, detection accuracy ≥92%, used to obtain the micromechanical parameters of the soil. Corrosion-resistant rubber isolation chambers are installed between each sensor to avoid signal crosstalk within the borehole, ensuring independent and accurate data from each sensor.

[0030] 3) Back-end data fusion processing unit (i.e. computer equipment): It can be composed of an industrial computer, a multi-source data fusion algorithm module, a pipeline area disease feature database, and a three-dimensional visualization terminal. It is deployed in the urban underground engineering safety monitoring center and has the full-process functions of data reception, processing, fusion, identification, early warning and visualization.

[0031] Figure 4 This is a flowchart of a multi-sensor big data fusion dynamic intelligent detection method according to an embodiment of this application, such as... Figure 4 As shown, the method may include the following steps: Step S410: The positioning data and the original map of the ground penetrating radar obtained by the electromagnetic positioning module of the target area are acquired by the detection vehicle.

[0032] Step S420: Generate a spatial mask that matches the original map pixel by pixel using the positioning data. Perform partitioned adaptive filtering and denoising on the original map according to the mask partition of the spatial mask to obtain the target map. Partitioned adaptive filtering and denoising is used to eliminate the influence of the reflection signal of underground engineering on the identification of soil collapse hazards. The spatial mask is used to characterize underground engineering.

[0033] In the above scheme, based on the mask partitioning of the spatial mask, a suitable filtering and denoising scheme can be used to denoise the corresponding regions in the original map. Then, secondary verification and iterative optimization are performed using the residual energy percentage to obtain a target map where the residual energy percentage is lower than the first threshold. Different filtering schemes are used to perform filtering and denoising steps for different regions (which can be divided according to different materials, types, etc., mainly reflecting the strength of reflection, i.e., divided into several regions according to reflection intensity, such as three regions of strong, medium and low, two regions of strong and low, etc.): For the first region in the original map corresponding to the first type of underground engineering of the spatial mask, the first wavelet basis function is used to perform P-layer wavelet decomposition. The high-frequency coefficients obtained by decomposition are processed using an adaptive soft thresholding function. σ is the noise standard deviation of the high-frequency coefficient, N is the length of the high-frequency coefficient, h is the actual burial depth of the pipeline, h0 is the radar calibration reference burial depth, and P is a positive integer. For the second region in the original map corresponding to the second type of underground engineering in the spatial mask, Q-layer wavelet decomposition is performed using the second wavelet basis function. The high-frequency coefficients obtained from the decomposition are processed using an improved semi-soft thresholding function. εs is the dielectric constant of the pipeline medium, εg is the dielectric constant of the surrounding soil, Q is a positive integer, the second wavelet basis function is different from the first wavelet basis function, and the second type of underground engineering is different from the first type of underground engineering; for the third region in the original map corresponding to the non-underground engineering of the spatial mask, a general hard threshold function is used for basic noise reduction; Perform secondary verification and iterative optimization steps: overlay the filtered radar image with the spatial mask and calculate the residual interference signal energy ratio of the corresponding area of ​​the underground project; if the residual interference signal energy ratio is not lower than the first threshold, readjust the decomposition level of the wavelet basis function and the correction coefficient of the threshold function and then perform the filtering and denoising step; if the residual interference signal energy ratio is lower than the first threshold, use the filtered radar image as the target image.

[0034] During overlay, the filtered radar image R_filtered can be multiplied pixel-by-pixel with the binary space mask M_selected (1 for underground engineering areas, 0 for the rest) to obtain a matrix R_selected containing only signals from the underground engineering areas. The proportion η is calculated using the following formula: η= (∑∑ R_selected²) / (∑∑ R_original_selected²) × 100%, The numerator is the sum of squares of energy in the filtered underground engineering region, and the denominator is the sum of squares of energy in the corresponding region of the original spectrum.

[0035] When adjusting the adjustable coefficients in the threshold function, for the metal pipeline area (adaptive soft threshold): the adjustable coefficient is the burial depth enhancement factor k (initially taken as 0.1 in the formula). If the residual energy is too high, k is increased by a certain value or proportion (e.g., 0.1→0.2) to make the threshold increase more significantly with the burial depth, thus stronger suppression of strong reflections. For the non-metallic pipeline area (improved semi-soft threshold): the adjustable coefficient is the dielectric difference scaling factor γ (initially taken as 1). If the residual energy is too high, γ is increased by a certain value or proportion (e.g., 1→1.5) to amplify the influence of the difference in dielectric constant between the pipeline and the soil on the threshold, thus strengthening the suppression of weak reflections. Decomposition layer number: usually, the decomposition layer number is increased (e.g., 5 layers→6 layers) to separate the signal in finer frequency bands, but excessive smoothing of the effective signal should be avoided.

[0036] When determining preliminary detection results (cavities, etc.) based on target maps, features such as hyperbolic morphology, strong amplitude phase reversal, and lower shadow areas are extracted from the filtered maps. Cavities / looseness are identified by threshold segmentation (amplitude exceeding the background mean by 3 times the standard deviation) or by pre-trained classifiers (SVM / CNN). The location, burial depth, type, and confidence level of suspected hidden dangers are output to guide the layout of detailed borehole investigations.

[0037] In the above scheme, after generating a spatial mask that matches the original map pixel by pixel using the positioning data, and performing partitioned adaptive filtering and noise reduction on the original map according to the mask partition of the spatial mask to obtain the target map, the preliminary detection results are determined based on the target map. The preliminary detection results are used to indicate whether there are suspected collapse hazards and their locations. After drilling holes at the suspected collapse hazard locations based on the preliminary detection results, monitoring units are deployed. The monitoring units include a communication module, a stratum stress sensor, a groundwater level and water content sensor, and a micro-core density detection module.

[0038] Step S430: Using four types of data, namely, the target map, the ground micro-deformation data of the fiber optic strain sensing module in the target area, the micro-core density detection data of the monitoring unit in the target area, and the formation stress and water content data, the type of collapse hazard in the target area is determined. The monitoring unit is installed in the target area by drilling at fixed points. In the above scheme: the dynamic consistency coefficient between the i-th type of data and other types of data can be determined based on the Pearson correlation coefficient rij between any two types of data, including target map, ground micro-deformation data, microcore compaction detection data, and formation stress and water content data. Where n=4, and the value range of Di is [0,1]; Based on the dynamic consistency coefficient Di and static credibility Ci of the i-th data, the comprehensive weight ωi of the i-th data is determined, ωi = αCi + βDi, where α is the static weight coefficient, β is the dynamic weight coefficient, and α+β=1; The comprehensive weight ωi of the i-th data is normalized to obtain the normalized weight ωi'; The basic probability allocation mi(A) of the i-th data is weighted and corrected using the normalized weight ωi' of the i-th data to obtain the corrected probability allocation m'i(A): m'i(A) = ωi' mi(A); Based on the DS combination rule, the corrected probability allocations of the target map, ground micro-deformation data, micro-core compaction detection data and stratum stress and water content data are fused to obtain the comprehensive soil characteristic parameters, and then the collapse hazard type of the target area is determined. Specifically, the comprehensive soil characteristic parameters can be input into the pre-trained model to determine the collapse hazard type of the target area.

[0039] The specific meaning of the basic probability assignment mi(A): mi(A) represents the probability of the i-th type of data (such as ground-penetrating radar images, density data, etc.) on the recognition framework. Trust level of a subset A; identification framework ={dense, slightly loose, moderately loose, severely loose, void}, encompassing all possible soil states; Subset A: can be a single state (e.g., "void") or a combination of multiple states (e.g., "loose or void"), used to express uncertainty; Physical representation: mi(A) represents the probability mass of considering the true soil state as belonging to A when judged individually based on the i-th type of data. For example, radar data assigns the probability of judging "void" as m1({void}) = 0.6, while also considering a probability of "moderately loose" as 0.1, with the remaining 0.3 allocated to the entire set. (Indicates uncertainty).

[0040] When using classic DS combination rules to merge objects: The objects of fusion: the probability assignments (m1', m2', m3', m4') of four types of data after weighted adjustment, where each mi' is defined in... A basic probability assignment function; fusion method: orthogonal summation according to Dempster's combination rule. For two pieces of evidence m1' and m2', the fusion formula is: , In the denominator Using the conflict coefficient, the four types of evidence are successively merged in pairs to obtain a total probability allocation mtotal; The fusion result is a fused basic probability assignment `mtotal`, which integrates the confidence levels of all four types of evidence for each subset of soil states. Specifically, it takes the form: `mtotal({dense})`, `mtotal({slightly loose})`, `mtotal({void})`, `mtotal({...}`. }); where, the state corresponding to the maximum probability in the single-element subset (such as "void") of mtotal is taken as the final "soil comprehensive feature parameter" (i.e., the soil state category determined by fusion). This category can be directly input into the pre-trained model (such as SVM classifier) ​​below for final hazard type confirmation, or directly output as the identification result.

[0041] In step S440, if the collapse hazard type in the target area is a cavity, a repair instruction is issued to instruct that material be filled into the cavity in the target area by drilling.

[0042] Urban road subsidence detection requires combining vehicle-mounted mobile detection with borehole-based fixed-point monitoring, and attempts to introduce multi-source data fusion. However, the aforementioned methods are mostly simple combinations or step-by-step implementations of different detection techniques, and have not yet proposed a complete solution with synergistic effects for the systemic technical challenges of complex urban underground engineering scenarios (such as complex pipeline networks). The technical solution in this application, through systematic technological innovations such as collaborative sensor deployment, zoned adaptive filtering, improved DS evidence theory fusion, and vehicle-to-borehole closed-loop iteration, achieves efficient, accurate, and dynamic detection of urban complex ground subsidence hazards, achieving significant technological progress and engineering application value. It can solve the technical problem of low detection accuracy of underground cavities and subsidence hazards in related technologies.

[0043] This application adopts an integrated closed-loop system architecture of "vehicle-mounted mobile survey + borehole fixed-point detailed survey + background data fusion", and solves the above-mentioned technical problems through the following core technological innovations: 1) Spatial Synchronization Sensor Co-deployment: The vehicle-mounted unit adopts a spatially synchronized linear array deployment. The distance between the electromagnetic positioning module and the ground penetrating radar module for underground engineering such as pipelines is fixed at 1 / 2 medium wavelength (12cm) to ensure that the spatial synchronization error is ≤5cm. All sensors have built-in GPS / BeiDou timestamp synchronization modules with a time synchronization accuracy of ≤1ms, providing a reliable foundation for multi-source data fusion. Optionally, for ground penetrating radars with different center frequencies, the distance between the electromagnetic positioning module and the radar module can be adjusted according to the medium wavelength of the radar in the soil, generally between 1 / 2 medium wavelength and 1 medium wavelength, to maintain spatial synchronization accuracy. 2) Zonal Adaptive Filtering Based on Pipeline Spatial Mask: Utilizing prior information on underground engineering such as pipelines obtained by the pipeline electromagnetic positioning module, a spatial mask matrix matching the radar map at the pixel level is generated. Matching adaptive denoising schemes are adopted for different regions. For example, different wavelet bases, decomposition levels, and threshold functions are used for adaptive denoising of metal pipelines, non-metal pipelines, and non-pipeline areas, respectively, and a secondary verification iterative optimization mechanism is introduced. Optionally, for denoising of metal pipeline areas, in addition to the db4 wavelet base, the db6, db8, and other Daubechies series wavelet bases can also be selected, and the decomposition level can be selected between 4 and 6 levels, adapted according to the actual interference intensity. 3) Improved DS Evidence Theory Based on Optimization of Evidence Credibility and Consistency: Introducing static credibility (based on sensor calibration accuracy) and dynamic consistency coefficient (based on Pearson correlation coefficient between data), constructing a comprehensive weight (ω=0.6C+0.4D), and then weighting and correcting the basic probability allocation of evidence before fusion to solve the problem of distortion in the fusion of conflicting evidence; Optionally, the static weight coefficient α and the dynamic weight coefficient β can be adjusted according to the data conflict probability of the specific detection scenario. Generally, the value range of α is 0.5-0.8, the value range of β is 0.2-0.5, and α+β=1; 4) Vehicle-mounted-borehole data closed-loop dynamic detection mechanism: The soil parameters (density, water content, dielectric constant) measured in the borehole are used to correct the inversion model of the vehicle-mounted ground penetrating radar, thereby achieving iterative improvement of the detection accuracy across the entire area; periodic re-measurement is carried out in general areas, and continuous monitoring is carried out in high-risk areas; 5) Disease feature database based on "actual measurement + simulation + amplification": The database is constructed by combining actual measurement data (12,000 sets from more than 60 cities), numerical simulation (8,000 sets from GPRMAX forward modeling), and generative adversarial network amplification (30,000 sets in total). A mapping relationship between 12-dimensional feature parameters and disease types is established to support high-precision training and self-learning updates of the SVM classifier. Optionally, in addition to using the SVM classifier, random forest, BP neural network, or deep learning model (such as CNN) can also be used for disease identification, which requires the database of this application for corresponding training.

[0044] As an optional embodiment, the technical solution of this application is further described in detail below with reference to specific implementation methods: 1) Radar signal partitioning adaptive filtering method based on pipeline network spatial mask refer to Figure 5 This method addresses the limitation of the aforementioned technologies in specifically eliminating interference from underground engineering structures such as pipelines. It utilizes prior information from the electromagnetic positioning module to achieve precise and adaptive elimination of such interference. Taking pipelines as an example, the specific steps are as follows: Step 1.1: Pixel-level network spatial mask matrix generation.

[0045] Based on the pipeline network type, diameter, burial depth, direction, and geographic coordinates obtained by the underground pipeline electromagnetic positioning module, and combined with the pixel resolution, detection depth, and geographic coordinate mapping relationship of the radar image, a pipeline network spatial mask matrix M is generated that perfectly matches the pixel-level of the ground-penetrating radar 2D image. The size of the mask matrix M is completely consistent with the radar image, and the matrix element assignment rule is as follows: the pixel area corresponding to the horizontal projection of the pipeline is assigned a value of 1, and the remaining non-pipeline areas are assigned a value of 0. Furthermore, according to the pipeline type, the mask matrix is ​​divided into a metal pipeline sub-matrix M1 and a non-metallic pipeline sub-matrix M2.

[0046] Step 1.2: Partition-based adaptive wavelet denoising.

[0047] The radar spectrum is partitioned based on the mask matrix, and differentiated wavelet denoising strategies are adopted for the interference characteristics of different regions: For the M1 region (strong reflection peak interference, corresponding to the first region mentioned above) corresponding to the metal pipeline: a 5-level wavelet decomposition using the Daubechies 4 (db4) wavelet basis is selected, and the high-frequency coefficients obtained from the decomposition are processed using an adaptive soft thresholding function. The formula for calculating the threshold T is: , Where σ is the noise standard deviation of the high-frequency coefficient, N is the length of the high-frequency coefficient, h is the actual burial depth of the pipeline, and h0 is the radar calibration reference burial depth (2m). This formula, by incorporating the pipeline burial depth h, allows the threshold to adapt to the signal attenuation characteristics of the pipeline; the greater the burial depth, the larger the threshold adjustment coefficient, thus more effectively suppressing deep-layer strong reflection interference. Wavelet reconstruction is then performed on the high-frequency and low-frequency coefficients after threshold processing to obtain the radar signal after removing strong interference.

[0048] For the M2 region (weak reflection interference, corresponding to the second region mentioned above) corresponding to non-metallic pipelines: a three-level wavelet decomposition is performed using the Symlets4 (sym4) wavelet basis, and the high-frequency coefficients obtained from the decomposition are processed using an improved semi-soft thresholding function. The formula for calculating the threshold T' is: , Where εs is the dielectric constant of the pipeline medium, and εg is the dielectric constant of the surrounding soil. This formula uses the difference in dielectric constant between the pipeline and the soil to quantify the intensity of weak reflections; the greater the difference, the higher the threshold, thus achieving targeted rejection. The coefficients after thresholding are reconstructed to obtain the radar signal after weak interference is removed.

[0049] For non-pipeline areas (areas where M=0, corresponding to the third area mentioned above): a general hard threshold function is used for basic noise reduction to preserve the effective reflection signals of soil defects.

[0050] Step 1.3: Secondary verification and iterative optimization.

[0051] The filtered radar image is superimposed on the pipeline network spatial mask matrix to calculate the proportion of residual interference signal energy in the corresponding pipeline area. If the proportion of residual energy exceeds a preset threshold of 5%, the wavelet decomposition level and the correction coefficient of the threshold calculation formula are readjusted until the proportion of residual energy is below 5%, thus completing the pipeline network interference removal.

[0052] 2) An improved DS evidence theory fusion method based on the optimization of evidence credibility and consistency refer to Figure 6 This method addresses the problem of distorted results in classical DS evidence theory when dealing with highly conflicting evidence. It introduces static credibility and dynamic consistency coefficients of evidence, and performs weighted correction on the basic probability allocation (BPA) of evidence before fusion.

[0053] Step 2.1: Determine the source of evidence and the identification framework.

[0054] The evidence sources consist of four types of simultaneously collected valid data: ground-penetrating radar data after interference removal; microcore density detection data; formation stress and water content data; and ground micro-deformation data. The identification framework Θ = {dense, slightly loose, moderately loose, severely loose, void} corresponds to five risk levels of potential collapse hazards.

[0055] Step 2.2: Calculation of static credibility of evidence.

[0056] Based on the laboratory calibration accuracy and historical engineering measurement accuracy of each sensor, the static confidence level Ci for each evidence source is determined as follows: C1 (radar data) = 0.88, C2 (density data) = 0.92, C3 (stress moisture content data) = 0.85, and C4 (micro-deformation data) = 0.82. Static confidence level is an inherent property of the sensor and can be calibrated quarterly based on subsequent engineering verification results.

[0057] Step 2.3: Calculation of the dynamic consistency coefficient of evidence.

[0058] For the multi-source data at the current moment, calculate the Pearson correlation coefficient rij between any two evidence sources, and obtain the average correlation coefficient of evidence i with all other evidence, which is the dynamic consistency coefficient Di of the evidence. The calculation formula is: , Where n=4 is the number of evidence sources, and Di ranges from [0,1]. The larger Di is, the higher the consistency of the evidence with other evidence, and the higher its credibility at that moment.

[0059] Step 2.4: Calculation of Evidence Overall Weight and BPA Correction.

[0060] The formula for calculating the overall weight ωi of each source of evidence is: ωi = α·Ci + β·Di, Where α is the static weighting coefficient (value 0.6), β is the dynamic weighting coefficient (value 0.4), and α + β = 1. This weighting strategy is determined based on the historical probability statistics of conflicts between vehicle-mounted data and borehole data in the urban pipeline network area, which preserves the inherent reliability of the sensors while dynamically responding to conflicts between data.

[0061] The combined weights of all evidence sources are normalized to obtain normalized weights ωi'. These normalized weights are then used to weight and correct the basic probability assignment mi(A) for each piece of evidence. m'i(A) = ωi' · mi(A), A Θ, Step 2.5: Improve the DS combination rule fusion.

[0062] The BPA of all the revised evidence is fused using the classic DS combination rule to obtain the final fusion result, which determines the disease type and risk level of the soil.

[0063] 3) Construction and interaction method of pipeline network disease characteristic database Database Construction: The database was constructed using a combination of measured data, numerical simulation, and sample augmentation. For measured data, 12,000 valid measured samples were collected from complex pipe network areas in over 60 cities across China. These samples included pipe network interference signals of different types, burial depths, and pipe diameters, as well as radar signals, density data, and stress data related to defects such as cavities, looseness, and voids. For numerical simulation, GPRMAX software was used to establish forward models for different geological conditions (sand, clay, soft soil), different pipe network parameters, and different defect types, generating 8,000 simulated samples to supplement scenarios not covered by the measured data. For sample augmentation, Generative Adversarial Networks (GANs) were used to augment the samples, ultimately forming a standardized database containing 30,000 valid samples.

[0064] Feature extraction and mapping: For all samples in the database, wavelet packet transform is used to extract the time-frequency domain features of the signal, including 12-dimensional feature parameters such as energy entropy, peak coefficient, variance, and skewness, and a one-to-one mapping relationship between "feature parameters - disease type - risk level" is established.

[0065] Interaction logic with the algorithm: During the training phase, the SVM classifier is trained under supervision using samples from the database to obtain an optimized classification model with a classification accuracy ≥95%. During the identification phase, the feature parameters obtained from the fusion of multi-source data are input into the trained SVM classifier. The classifier then uses the mapping relationships in the database to identify the disease type and risk level. During the self-update phase, newly identified and field-verified valid samples are added to the database. The SVM classification model is retrained quarterly to achieve self-learning updates for both the database and the classification model.

[0066] 4) Vehicle-mounted-borehole data closed-loop dynamic detection mechanism This mechanism achieves deep coupling between vehicle-mounted and borehole data. The specific process is as follows: Step 4.1: The vehicle-mounted mobile detection unit completes the initial detection of the entire area. After interference elimination and preliminary identification, the suspected disease area is delineated.

[0067] Step 4.2: Based on the location, size, and depth of the suspected disease area, the back-end unit uses a genetic algorithm to intelligently optimize the layout, number, and depth of boreholes to ensure that the boreholes are located in the core area of ​​the suspected disease area, thereby enabling targeted and detailed investigation.

[0068] Step 4.3: The measured data of soil density, moisture content and dielectric constant obtained by the borehole fixed-point monitoring unit are input in reverse into the dielectric constant inversion model of the vehicle-mounted ground penetrating radar. The parameters of the model are dynamically corrected to improve the radar's accuracy in identifying defects in the subsequent detection area.

[0069] Step 4.4: For general detection areas, vehicle-mounted units are used for retesting every 7-15 days. The retesting data is processed using the corrected inversion model to achieve iterative improvement in the accuracy of detection across the entire area. For high-risk areas, borehole units are used for 24-hour continuous monitoring to update the soil parameter change curves in real time and capture early signals of potential hazards.

[0070] Based on the above core content, refer to Figure 7 The detection method of this application includes the following steps: 1) Preprocessing and parameter calibration of the detection area: Collect pipeline network completion data of the detection area and divide the general detection area into high-risk areas; perform time stamp synchronization calibration on all sensors with a synchronization accuracy of ≤1ms; perform sensor accuracy calibration in standard stratum areas without pipeline network interference and determine the reference parameters of each sensor.

[0071] The following provides a specific parameter configuration for an optional embodiment: Detection Area: Core business district of the old city, covering an area of ​​2 km², with pipeline spacing of 0.3m-2m and burial depth of 0.8m-5m. Vehicle-mounted unit parameters: Array spacing 12cm, ground-penetrating radar center frequency 1.5GHz, electromagnetic positioning module operating frequency 10kHz-1MHz, data acquisition frequency 10Hz, vehicle speed 5km / h. Drilling unit parameters: Drill holes are laid out at 8m intervals in high-risk areas, with a diameter of 110mm and a burial depth of 7m (maximum burial depth of pipelines is 5m+2m), and data upload frequency 30min / time. Database parameters: 30,000 standardized samples, SVM classifier pre-training accuracy of 95.2%. Filtering algorithm parameters: 5-layer decomposition using db4 wavelet basis for metal pipelines, 3-layer decomposition using sym4 wavelet basis for non-metallic pipelines, residual interference energy threshold 5%. Fusion algorithm parameters: static weight coefficient α=0.6, dynamic weight coefficient β=0.4, static confidence C=[0.88,0.92,0.85,0.82].

[0072] 2) Vehicle-mounted mobile full-domain synchronous detection: The detection vehicle equipped with the vehicle-mounted unit travels along the preset route at a speed of 5-10km / h. All sensors are triggered synchronously to collect radar maps, pipeline positioning data, ground micro-deformation data and geographic coordinate data, which are then uploaded to the back-end unit in real time.

[0073] 3) Adaptive elimination of pipeline interference: The back-end unit uses the above-mentioned "partition adaptive filtering method based on pipeline spatial mask" to process radar data, eliminate pipeline interference signals, and obtain effective radar map of defects.

[0074] 4) High-risk area borehole point monitoring: Based on the preliminary detection results, high-risk areas are delineated, boreholes are laid out and borehole point monitoring units are installed to obtain detailed measured data on soil density, stress, moisture content and water level, and upload them to the background unit in real time.

[0075] 5) Spatiotemporal registration and fusion of multi-source data: Based on timestamps and geographic coordinates, vehicle data and borehole data are spatiotemporally registered at the pixel level to generate a spatiotemporally synchronized multi-source dataset; the above-mentioned "improved DS evidence theory fusion method" is used to fuse the multi-source data to obtain the comprehensive characteristic parameters of the soil.

[0076] 6) Disease identification and risk classification: Input the fused feature parameters into the trained SVM classifier, call the pipeline network disease feature database, automatically identify the type, location, burial depth and scale of potential collapse hazards, and classify the risk level.

[0077] 7) Closed-loop dynamic monitoring and early warning: The above-mentioned "vehicle-mounted-borehole closed-loop dynamic detection mechanism" is used to periodically re-measure and continuously monitor the entire area, iteratively improving the detection accuracy. When a high-risk hazard of level III or above is identified, or when the rate of change of soil parameters exceeds the preset threshold (density decrease rate ≥ 2% / day), a real-time early warning is automatically triggered.

[0078] 8) Detection result output: Automatically generate detection reports, including information such as pipeline distribution, hidden danger parameters, risk level, and development trend, and generate a three-dimensional visualization model of underground diseases in the detection area.

[0079] Compared with related technologies, this application has the following advantages: 1) Significantly Improved Removal of Underground Interference from Pipelines: The proposed zonal adaptive filtering method based on pipeline spatial masks utilizes prior information obtained from pipeline electromagnetic positioning modules to generate a mask matrix. Different wavelet bases, decomposition levels, and threshold functions are applied to metal pipelines, non-metal pipelines, and non-pipeline areas, respectively, and a secondary verification and iterative optimization mechanism is introduced. Engineering field tests have verified that, compared with existing general wavelet denoising methods, the removal rate of pipeline interference signals is improved by 32%, and the false alarm rate for defect identification is reduced from 38% to 7%, effectively solving the core technical problem of high false alarm rates caused by severe signal interference in complex pipeline areas.

[0080] 2) Significantly Improved Accuracy of Conflicting Evidence Fusion: The improved DS evidence theory fusion method in this application introduces static credibility (based on sensor calibration accuracy) and dynamic consistency coefficient (based on Pearson correlation coefficient) of evidence. A comprehensive weight is constructed to weight and correct the evidence BPA before fusion, solving the problem of result distortion in fixed-weight fusion methods when dealing with conflicts between vehicle-mounted and borehole data. Experimental results show that the accuracy of the fusion result is improved by 28% for multi-source data conflict scenarios, providing a reliable decision-making basis for subsequent disease identification.

[0081] 3) Iterative improvement in detection accuracy: The vehicle-mounted-borehole closed-loop dynamic detection mechanism of this application is the first to realize the function of reverse correction of the vehicle-mounted ground-penetrating radar inversion model using borehole measured data, upgrading the two types of data from "simple superposition" to "deep coupling". Compared with the existing open-loop detection method, the detection efficiency is improved by 3 times, the defect location error is reduced from 1.2m to 0.3m, and the overall detection accuracy can be continuously optimized with the increase of the number of re-measurements.

[0082] 4) Significantly improved spatiotemporal synchronization of data: The spatially synchronized linear array deployment structure of this application, with the electromagnetic positioning module and radar module spaced at a fixed distance of 1 / 2 medium wavelength, ensures that the spatial synchronization error of multi-source data at the same detection location is ≤5cm; all sensors have built-in GPS / BeiDou timestamp synchronization modules, with a time synchronization accuracy of ≤1ms. Compared with the existing technology where the spatial deviation is ≥20cm and the time deviation is ≥100ms, the spatiotemporal synchronization of this application provides a reliable foundation for subsequent data fusion.

[0083] 5) High accuracy and low false alarm rate in disease identification: The "measured + simulated + amplified" database construction method of this application covers 12,000 sets of measured data, 8,000 sets of numerical simulation data, and 30,000 sets of amplified samples from more than 60 cities, significantly improving the generalization ability of the classifier. Verified in actual engineering projects, the accuracy rate of identifying subsidence hazards in densely piped areas (pipeline spacing ≤ 1m) reaches 94%, effectively capturing early loosening hazards (a warning is triggered when the density decreases by ≥ 2% / day), providing precise technical support for subsidence prevention.

[0084] 6) High detection efficiency and low overall cost: Using the technical solution of this application, it only takes 2 days to complete the full-area detection of a 2km² complex pipeline network, while the traditional solution takes more than 7 days; the number of boreholes is intelligently optimized through a genetic algorithm, reducing it by about 30% compared to the traditional uniform borehole layout method; at the same time, due to the significant reduction in false alarm rate, the workload of on-site verification and invalid handling is significantly reduced. Overall, the detection cost of this application is reduced by about 40%-50% compared to existing technologies.

[0085] In summary, this application has achieved efficient, accurate, and dynamic detection of ground subsidence hazards in complex urban pipe network areas through systematic technological innovations such as collaborative sensor deployment, zoned adaptive filtering, improved DS evidence theory fusion, and vehicle-to-borehole closed-loop iteration, resulting in significant technological progress and engineering application value.

[0086] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0087] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0088] According to another aspect of the embodiments of this application, a multi-sensor big data fusion dynamic intelligent detection device for implementing the above-described multi-sensor big data fusion dynamic intelligent detection method is also provided. Figure 8 This is a schematic diagram of a multi-sensor big data fusion dynamic intelligent detection device according to an embodiment of this application, such as... Figure 8 As shown, the device may include: The acquisition unit 81 is used to acquire the positioning data and the original map of the ground penetrating radar obtained by the electromagnetic positioning module of the target area through the detection vehicle to locate the underground project. The denoising unit 82 is used to generate a spatial mask that matches the original map pixel by pixel using the positioning data, and to perform partitioned adaptive filtering denoising on the original map according to the mask partition of the spatial mask to obtain the target map. The partitioned adaptive filtering denoising is used to eliminate the influence of the reflection signal of the underground project on the identification of soil collapse hazards, and the spatial mask is used to characterize the underground project. The processing unit 83 is used to determine the type of collapse hazard in the target area by using four types of data: the target map, the ground micro-deformation data of the fiber optic strain sensing module, the micro-core density detection data of the monitoring unit, and the formation stress and water content data. The monitoring unit is installed in the target area by drilling. Repair unit 84 is used to issue a repair instruction when the collapse hazard type in the target area is a cavity, instructing to fill the cavity in the target area with material by drilling.

[0089] Optionally, the denoising unit is further configured to: denoise the corresponding region in the original map using an appropriate filtering and denoising scheme according to the mask partition of the spatial mask, and perform secondary verification and iterative optimization through the residual energy ratio to obtain the target map with a residual energy ratio lower than a first threshold.

[0090] Optionally, the denoising unit is further configured to: perform a filtering and denoising step: for the first region in the original map corresponding to the first type of underground engineering of the spatial mask, perform P-layer wavelet decomposition using the first wavelet basis function, and process the high-frequency coefficients obtained by decomposition using an adaptive soft thresholding function, wherein the adaptive soft thresholding function... σ is the noise standard deviation of the high-frequency coefficient, N is the length of the high-frequency coefficient, h is the actual burial depth of the pipeline, h0 is the radar calibration reference burial depth, and P is a positive integer. For the second region in the original map corresponding to the second type of underground engineering in the spatial mask, Q-layer wavelet decomposition is performed using the second wavelet basis function. The high-frequency coefficients obtained from the decomposition are processed using an improved semi-soft thresholding function. εs is the dielectric constant of the pipeline medium, εg is the dielectric constant of the surrounding soil, Q is a positive integer, the second wavelet basis function is different from the first wavelet basis function, and the second type of underground engineering is different from the first type of underground engineering. For the third region in the original map corresponding to the non-underground engineering of the spatial mask, a general hard threshold function is used for basic denoising. Secondary verification and iterative optimization steps are performed: the filtered radar map is superimposed on the spatial mask, and the residual interference signal energy ratio of the region corresponding to the underground engineering is calculated. If the residual interference signal energy ratio is not lower than the first threshold, the decomposition level of the wavelet basis function and the correction coefficient of the threshold function are readjusted and the filtering and denoising steps are performed. If the residual interference signal energy ratio is lower than the first threshold, the filtered radar map is used as the target map.

[0091] Optionally, the processing unit is further configured to: after generating a spatial mask that matches the original map pixel-level using the positioning data, performing partitioned adaptive filtering and noise reduction on the original map according to the mask partitions of the spatial mask to obtain the target map, determine preliminary detection results based on the target map, wherein the preliminary detection results are used to indicate whether there is a suspected collapse hazard and its location; and after drilling at the suspected collapse hazard location based on the preliminary detection results, deploy the monitoring unit, wherein the monitoring unit includes a communication module, a formation stress sensor, a groundwater level and water content sensor, and a micro-core density detection module.

[0092] Optionally, the processing unit is further configured to: determine the dynamic consistency coefficient Di between the i-th type of data and other types of data based on the Pearson correlation coefficient rij between any two types of data among the target map, the ground micro-deformation data, the micro-core compactness detection data, and the formation stress and water content data. Where n=4, and the value range of Di is [0,1]; Based on the dynamic consistency coefficient Di and static reliability Ci of the i-th type of data, the comprehensive weight ωi of the i-th type of data is determined, ωi = αCi + βDi, where α is the static weight coefficient, β is the dynamic weight coefficient, and α+β=1; The comprehensive weight ωi of the i-th type of data is normalized to obtain the normalized weight ωi'; The basic probability allocation mi(A) of the i-th type of data is weighted and corrected using the normalized weight ωi' of the i-th type of data to obtain the corrected probability allocation m'i(A): m'i(A) = ωi' mi(A); Based on the DS combination rule, the corrected probability allocations of the target map, the ground micro-deformation data, the micro-core compaction detection data, and the stratum stress and water content data are fused to obtain the comprehensive soil characteristic parameters, thereby determining the collapse hazard type of the target area.

[0093] Optionally, the processing unit is further configured to: fuse the corrected probability allocations of the target map, the ground micro-deformation data, the micro-core compaction detection data, and the stratum stress and water content data based on the DS combination rule to obtain the comprehensive soil characteristic parameters; input the comprehensive soil characteristic parameters into the pre-trained model to determine the collapse hazard type of the target area.

[0094] It should be noted that the examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above embodiments. It should also be noted that the above modules, as part of a device, can operate in environments such as... Figure 1 The hardware environment shown can be implemented through software or hardware, and the hardware environment includes the network environment.

[0095] According to another aspect of the embodiments of this application, a server or terminal for implementing the above-described multi-sensor big data fusion dynamic intelligent detection method is also provided.

[0096] Figure 9 This is a structural block diagram of a terminal according to an embodiment of this application, such as... Figure 9 As shown, the terminal may include: one or more (only one is shown in the figure) processors 901, memory 903, and transmission devices 905, such as... Figure 9 As shown, the terminal may also include input / output devices 907.

[0097] The memory 903 can be used to store software programs and modules, such as the program instructions / modules corresponding to the multi-sensor big data fusion dynamic intelligent detection method and device in this embodiment. The processor 901 executes various functional applications and data processing by running the software programs and modules stored in the memory 903, thereby realizing the aforementioned multi-sensor big data fusion dynamic intelligent detection method. The memory 903 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 903 may further include memory remotely located relative to the processor 901, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0098] The aforementioned transmission device 905 is used to receive or send data via a network, and can also be used for data transfer between a processor and memory. Specific examples of the network described above may include wired networks and wireless networks. In one example, the transmission device 905 includes a Network Interface Controller (NIC), which can be connected to other network devices and a router via a network cable to communicate with the Internet or a local area network. In another example, the transmission device 905 is a radio frequency (RF) module used for wireless communication with the Internet.

[0099] Specifically, memory 903 is used to store application programs.

[0100] The processor 901 can invoke the application program stored in the memory 903 via the transmission device 905 to perform the following steps: The positioning data and the original map of the ground penetrating radar are obtained by using the electromagnetic positioning module of the detection vehicle to locate the target area of ​​the underground project. A spatial mask that matches the original map pixel-level is generated using the positioning data. The original map is then subjected to partitioned adaptive filtering and denoising based on the mask partitions of the spatial mask to obtain the target map. The partitioned adaptive filtering and denoising is used to eliminate the influence of the reflection signals of underground engineering on the identification of potential soil collapse hazards. The spatial mask is used to characterize the underground engineering. Using four types of data—the target map, ground micro-deformation data from the fiber optic strain sensing module within the target area, micro-core density detection data from the monitoring unit within the target area, and formation stress and water content data—the type of subsidence hazard in the target area is determined. The monitoring unit is installed in the target area using a borehole-based fixed-point method. If the collapse hazard type in the target area is a cavity, a repair instruction is issued to instruct that material be filled into the cavity in the target area by drilling.

[0101] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0102] Those skilled in the art will understand that Figure 9 The structure shown is for illustrative purposes only. The terminal can be a smartphone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile internet device (MID), a PAD, or other terminal devices. Figure 9 This does not limit the structure of the aforementioned electronic device. For example, the terminal may also include components that are more... Figure 9 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 9 The different configurations shown.

[0103] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0104] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to execute program code for a multi-sensor big data fusion dynamic intelligent detection method.

[0105] Optionally, in this embodiment, the storage medium may be located on at least one of the network devices in the network shown in the above embodiment.

[0106] Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps: The positioning data and the original map of the ground penetrating radar are obtained by using the electromagnetic positioning module of the detection vehicle to locate the target area of ​​the underground project. A spatial mask that matches the original map pixel-level is generated using the positioning data. The original map is then subjected to partitioned adaptive filtering and denoising based on the mask partitions of the spatial mask to obtain the target map. The partitioned adaptive filtering and denoising is used to eliminate the influence of the reflection signals of underground engineering on the identification of potential soil collapse hazards. The spatial mask is used to characterize the underground engineering. Using four types of data—the target map, ground micro-deformation data from the fiber optic strain sensing module within the target area, micro-core density detection data from the monitoring unit within the target area, and formation stress and water content data—the type of subsidence hazard in the target area is determined. The monitoring unit is installed in the target area using a borehole-based fixed-point method. If the collapse hazard type in the target area is a cavity, a repair instruction is issued to instruct that material be filled into the cavity in the target area by drilling.

[0107] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0108] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0109] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0110] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0111] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0112] In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between units or modules, and may be electrical or other forms.

[0113] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0114] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0115] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A dynamic intelligent detection method based on multi-sensor big data fusion, characterized in that, include: The positioning data and the original map of the ground penetrating radar are obtained by using the electromagnetic positioning module of the detection vehicle to locate the target area of ​​the underground project. A spatial mask that matches the original map pixel-level is generated using the positioning data. The original map is then subjected to partitioned adaptive filtering and denoising based on the mask partitions of the spatial mask to obtain the target map. The partitioned adaptive filtering and denoising is used to eliminate the influence of the reflection signals of underground engineering on the identification of potential soil collapse hazards. The spatial mask is used to characterize the underground engineering. Using four types of data—the target map, ground micro-deformation data from the fiber optic strain sensing module within the target area, micro-core density detection data from the monitoring unit within the target area, and formation stress and water content data—the type of subsidence hazard in the target area is determined. The monitoring unit is installed in the target area using a borehole-based fixed-point method. If the collapse hazard type in the target area is a cavity, a repair instruction is issued to instruct that material be filled into the cavity in the target area by drilling.

2. The method according to claim 1, characterized in that, Based on the mask partitioning of the spatial mask, the original map is subjected to partitioned adaptive filtering and denoising to obtain the target map, including: Based on the mask partitions of the spatial mask, the corresponding regions in the original map are denoised using an appropriate filtering and denoising scheme, and then further verified and iteratively optimized using the residual energy ratio to obtain the target map with a residual energy ratio lower than a first threshold.

3. The method according to claim 2, characterized in that, Based on the mask partitioning of the spatial mask, the corresponding regions in the original map are denoised using an appropriate filtering and denoising scheme. Secondary verification and iterative optimization are then performed using the residual energy percentage to obtain the target map where the residual energy percentage is lower than a first threshold, including: The filtering and denoising steps are as follows: For the first region in the original map corresponding to the first type of underground engineering in the spatial mask, P-layer wavelet decomposition is performed using the first wavelet basis function. The high-frequency coefficients obtained from the decomposition are then processed using an adaptive soft thresholding function. σ is the noise standard deviation of the high-frequency coefficient, N is the length of the high-frequency coefficient, h is the actual burial depth of the pipeline, h0 is the radar calibration reference burial depth, and P is a positive integer. For the second region in the original map corresponding to the second type of underground engineering in the spatial mask, Q-layer wavelet decomposition is performed using the second wavelet basis function. The high-frequency coefficients obtained from the decomposition are processed using an improved semi-soft thresholding function. εs is the dielectric constant of the pipeline medium, εg is the dielectric constant of the surrounding soil, Q is a positive integer, the second wavelet basis function is different from the first wavelet basis function, and the second type of underground engineering is different from the first type of underground engineering; for the third region in the original map corresponding to the non-underground engineering of the spatial mask, a general hard threshold function is used for basic noise reduction; Perform secondary verification and iterative optimization steps: superimpose the filtered radar image onto the spatial mask, and calculate the residual interference signal energy ratio of the corresponding area of ​​the underground project; if the residual interference signal energy ratio is not lower than the first threshold, readjust the decomposition level of the wavelet basis function and the correction coefficient of the threshold function, and then perform the filtering and denoising step; if the residual interference signal energy ratio is lower than the first threshold, use the filtered radar image as the target image.

4. The method according to claim 1, characterized in that, After generating a spatial mask that matches the original map pixel-level using the positioning data, and performing partitioned adaptive filtering and denoising on the original map according to the mask partitions of the spatial mask to obtain the target map, the method further includes: Preliminary detection results are determined based on the target map, wherein the preliminary detection results are used to indicate whether there are any suspected collapse hazards and their locations; Based on the preliminary detection results, the monitoring unit was deployed after drilling at the suspected collapse hazard location. The monitoring unit includes a communication module, a ground stress sensor, a groundwater level and water content sensor, and a micro-core density detection module.

5. The method according to any one of claims 1 to 4, characterized in that, Using four types of data—the target map, ground micro-deformation data from fiber optic strain sensing modules within the target area, micro-core density detection data from monitoring units within the target area, and formation stress and water content data—the type of subsidence hazard in the target area is determined, including: Based on the Pearson correlation coefficient between any two types of data from the target map, the ground microdeformation data, the microcore compaction detection data, and the formation stress and water content data, Determine the first Dynamic consistency coefficient between class data and other classes of data , Where n=4, The value range is [0,1]; According to the first Dynamic consistency coefficient of class data And the static confidence level Ci, determine the comprehensive weight ωi of the i-th type of data, ωi = αCi + βDi, where α is the static weight coefficient, β is the dynamic weight coefficient, and α+β=1; The comprehensive weight ωi of the i-th type of data is normalized to obtain the normalized weight ωi'; The basic probability assignment mi(A) of the i-th class of data is weighted and corrected using the normalized weight ωi' of the i-th class of data to obtain the corrected probability assignment m'i(A): m'i(A) = ωi' mi(A); Based on the DS combination rule, the corrected probability allocations of the target map, the ground micro-deformation data, the micro-core compaction detection data, and the stratum stress and water content data are fused to obtain the comprehensive soil characteristic parameters, thereby determining the type of collapse hazard in the target area.

6. The method according to claim 5, characterized in that, Based on the DS combination rule, the corrected probability allocations of the target map, the ground micro-deformation data, the micro-core compaction detection data, and the formation stress and water content data are fused to obtain comprehensive soil characteristic parameters, thereby determining the collapse hazard type of the target area, including: Based on the DS combination rule, the corrected probability allocations of the target map, the ground micro-deformation data, the micro-core compaction detection data, and the formation stress and water content data are fused to obtain the comprehensive characteristic parameters of the soil. The comprehensive characteristic parameters of the soil are input into a pre-trained model to determine the type of potential collapse in the target area.

7. A multi-sensor big data fusion dynamic intelligent detection device, characterized in that, include: The acquisition unit is used to acquire positioning data and the original map of the ground penetrating radar obtained by the electromagnetic positioning module of the target area through the detection vehicle to locate the underground project; The denoising unit is used to generate a spatial mask that matches the original map pixel by pixel using the positioning data, and to perform partitioned adaptive filtering denoising on the original map according to the mask partition of the spatial mask to obtain the target map. The partitioned adaptive filtering denoising is used to eliminate the influence of the reflection signal of the underground project on the identification of soil collapse hazards, and the spatial mask is used to characterize the underground project. The processing unit is used to determine the type of subsidence hazard in the target area by using four types of data: the target map, the ground micro-deformation data of the fiber optic strain sensing module, the micro-core density detection data of the monitoring unit, and the formation stress and water content data. The monitoring unit is installed in the target area by drilling. The repair unit is used to issue a repair instruction when the collapse hazard type in the target area is a cavity, instructing the filling of the cavity in the target area with material by drilling.

8. A multi-sensor big data fusion dynamic intelligent detection system, characterized in that, include: The detection vehicle is used to acquire positioning data and raw ground-penetrating radar maps obtained by the electromagnetic positioning module when passing through the target area to locate underground engineering. A computer device is used to generate a spatial mask that matches the original map pixel-level using the positioning data. Based on the mask partitions of the spatial mask, the original map is subjected to partitioned adaptive filtering and denoising to obtain a target map. The partitioned adaptive filtering and denoising is used to eliminate the influence of reflected signals from underground engineering on the identification of potential soil collapse hazards. The spatial mask is used to characterize the underground engineering. Using four types of data—the target map, ground micro-deformation data from the fiber optic strain sensing module, micro-core density detection data from the monitoring unit, and ground stress and moisture content data—the type of potential collapse hazard in the target area is determined. The monitoring unit is installed in the target area using a borehole-based fixed-point method. Repair equipment is used to fill the cavities in the target area with material by drilling when the collapse hazard type in the target area is a cavity.

9. A computer-readable storage medium, characterized in that, The storage medium includes a stored program, wherein the program executes the method described in any one of claims 1 to 6 when it is run.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the method described in any one of claims 1 to 6 via the computer program.

Images