Multi-scale deformation monitoring system and method based on InSAR and measurement robot collaborative operation

Abstract

The invention provides a multi-scale deformation monitoring system and method based on InSAR and measurement robot collaborative operation, and the system comprises a space sensing layer, a ground execution layer, and a cloud decision layer. The spatial sensing layer is used for receiving multiband SAR data, preprocessing the data and identifying an earth surface deformation abnormal area by using a time sequence InSAR processing technology; the ground execution layer dynamically deploys a measurement robot network according to the spatial distribution of the abnormal region; and the cloud decision-making layer obtains a deformation prediction model by adopting data fusion processing and an LSTM neural network, formulates an overall decision-making scheme according to fused data and a prediction result, and issues a decision-making instruction to the ground execution layer. According to the method, the hidden danger recognition precision is improved through cooperative operation of the InSAR and the measurement robot, and the cloud end integrates real-time monitoring data and a historical deformation sequence, and based on LSTM model prediction, the disaster response efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of engineering safety monitoring and geological disaster warning, and specifically to a multi-scale deformation monitoring system and method based on the collaborative operation of InSAR and a survey robot. Background Art

[0002] InSAR (Interferometric Synthetic Aperture Radar) is a technology for monitoring surface deformation using synthetic aperture radar (SAR) images. It detects minute changes on the earth's surface by comparing the phase differences of SAR images acquired at different times, with an accuracy that can reach the millimeter level. The advantages of InSAR are its wide coverage, enabling the monitoring of large-area surface deformation, and being unaffected by weather conditions as the radar can penetrate clouds. However, InSAR also has some limitations. For example, its temporal resolution is limited, depending on the satellite revisit period (such as 12 days for Sentinel-1), making it unable to capture sudden deformations; its spatial resolution is insufficient, with the strip mode resolution being greater than 3m, making it difficult to monitor local minute deformations (such as the displacement of bridge bearings); and the atmospheric error is significant, with the ionospheric and tropospheric delays causing deformation calculation errors of up to 2 - 5 cm.

[0003] A survey robot is an automated total station, usually used for engineering monitoring, such as the deformation monitoring of dams, bridges, buildings, etc. The survey robot can automatically track targets and obtain high-precision three-dimensional coordinate data in real time, with high temporal resolution and being suitable for real-time or near-real-time monitoring. However, its coverage is limited. Usually, it can only monitor specific points where retroreflective prisms are arranged. The effective monitoring radius of a single device is less than 800m, and dense point layout is required for wide-area monitoring, resulting in high costs; and it requires a line-of-sight condition and is greatly affected by environmental occlusion.

[0004] In the prior art, only the integrated monitoring method of InSAR and GNSS (Global Navigation Satellite System) has been proposed, but the GNSS station layout is fixed and the density is low, making it difficult to meet the local high-precision requirements. Summary of the Invention

[0005] Aiming at the deficiencies of the above prior art, the present invention provides a multi-scale deformation monitoring system and method based on the collaborative operation of InSAR and a survey robot. By constructing a spatio-temporal coupling model, a dynamic reference network, and a multi-source data fusion algorithm, it realizes the collaborative optimization of large-range surface deformation monitoring and millimeter-level local precise measurement, and solves problems such as the inability to have both scale and accuracy, significant interference from atmospheric errors, and insufficient real-time performance in traditional monitoring technologies.

[0006] The technical solution provided by the present invention: A multi-scale deformation monitoring system based on the collaborative operation of InSAR and total station, including a space perception layer, a ground execution layer, and a cloud decision-making layer.

[0007] The space perception layer is used to receive multi-band SAR data, perform data preprocessing, and generate regional deformation rate maps and identify surface deformation anomaly regions using time series InSAR processing technology.

[0008] The ground execution layer dynamically deploys a total station network according to the spatial distribution of the anomaly regions. Each intelligent total station is equipped with multiple sensors to obtain its own position, attitude, and surrounding environment information in real time. The total stations establish connections through a wireless communication module and adopt a self-organizing network protocol to automatically adjust the connection relationship according to signal strength and distance, forming a stable communication network.

[0009] The cloud decision-making layer uses a feature-level fusion method to extract the common features of satellite data and ground measurement data for fusion processing, then uses an LSTM neural network for short-term deformation trend prediction. Finally, according to the fusion data and prediction results, an overall decision-making plan is formulated and the decision instructions are sent to the ground execution layer. When the real-time monitored deformation in any region of the fusion deformation field or the trend of the LSTM model exceeds the threshold, the cloud decision-making layer triggers and pushes warning information.

[0010] Further, the space perception layer includes an SAR data receiving module, a data preprocessing module, and a time series InSAR processing module. The SAR data receiving module is used to receive multi-band SAR data from an SAR satellite constellation. The data preprocessing module is used to denoise the received SAR data, and then perform data correction to eliminate errors caused by satellite attitude and orbit deviation factors, making the data consistent with the actual geographical coordinates. The time series InSAR processing module selects multiple master images to perform interferometric processing on multi-temporal SAR data of the same region, obtains the overall deformation information of the region, generates a regional deformation rate map, intuitively shows the deformation speed differences in different regions, and identifies surface deformation anomaly regions.

[0011] Further, the space perception layer further includes an SBAS / PSI processing module. Through time series interferometric measurement technology, stable scatterers are selected as monitoring points, and minute deformation information is extracted through multi-temporal data for large-scale overall monitoring.

[0012] Further, the sensors equipped on the intelligent total station include a laser rangefinder and an inertial measurement unit. Using an advanced navigation algorithm, combined with map data and sensor information, the total station autonomously plans the optimal travel path, adjusts the navigation strategy in real time, and through dynamic networking, the total stations share measurement data and position information in real time to collaboratively complete complex tasks.

[0013] Furthermore, the cloud decision-making layer includes a data fusion processing module, an LSTM neural network deformation prediction module, a data transmission and system coordination processing module, and an early warning module.

[0014] The data fusion processing module adopts a feature-level fusion method to extract the common features of satellite data and ground measurement data, fuse and process the deformation trends, invert the local atmospheric delay amount using the coordinates of the total station, generate a regional atmospheric phase map through spatial interpolation, correct the InSAR atmospheric phase, and use a weighted average algorithm to assign different weights according to the reliability and importance of the data sources to achieve data fusion.

[0015] The LSTM neural network deformation prediction module includes an input layer, a hidden layer, a network training module, and an output layer. Among them, the LSTM layer and the Dropout layer are designed as stacked hidden layers. The input layer will complete the preprocessing of the fused deformation monitoring data, including the division and standardization of samples to meet the data requirements of the LSTM layer. The hidden layer learns the non-linear correlation between settlement time series through LSTM unit cells, and uses the Dropout layer to randomly deactivate a certain proportion of neurons. The network training module uses the ADAM optimizer to optimize the network weights. The output layer includes the trained network and its predicted values.

[0016] The data transmission and system coordination processing module receives the satellite data processed by the space perception layer and the measurement data of the ground execution layer, formulates an overall decision-making plan based on the fused data and prediction results, and issues decision-making instructions to the ground execution layer.

[0017] The early warning module is used to trigger and push early warning information when the real-time monitored deformation in any area of the fused deformation field or the trend of the LSTM model exceeds the threshold.

[0018] Another technical solution provided by the present invention: A multi-scale deformation monitoring method based on the collaborative operation of InSAR and total station, including the following steps:

[0019] (1) Receive multi-band SAR data of the SAR satellite constellation, denoise and correct the data, and then use the time series InSAR technology to obtain the overall deformation information of the region and identify the surface deformation abnormal area.

[0020] (2) Total station dynamic networking

[0021] (2-1) According to the spatial distribution of the abnormal area, dynamically deploy the total station network, and the node spacing D is adaptively adjusted according to the deformation gradient.

[0022] D = 500 / (1 + 0.2 × G)

[0023] where G is the deformation rate, in mm / month.

[0024] (2-2) Deliver the surveying robot to the target position, deploy the triangular support and complete the initial positioning. The surveying robot achieves millimeter-level positioning through multi-sensor fusion, and the dynamic positioning error model is:

[0025] ΔP(t) = αgv(t) + βga(t) + γgw(t) + ε

[0026] where α, β, and γ are sensor weight coefficients, v(t) is the velocity, a(t) is the acceleration, w(t) is the angular velocity, and ε is the noise error.

[0027] (2-3) After positioning, conduct network communication tests, lay prisms at key points or directly automatically track and aim at the target for real-time three-dimensional coordinate monitoring;

[0028] (3) Data fusion and collaborative monitoring

[0029] Unify the spatio-temporal reference of InSAR and the surveying robot, correct the InSAR atmospheric delay error using the surveying robot data, estimate the atmospheric phase distribution of the entire area using spatial interpolation method with the robot points, and use the weighted average algorithm for the overlapping points, and assign different weights according to the reliability and importance of the data sources for monitoring data fusion;

[0030] (4) Cloud decision-making and task scheduling

[0031] Transmit the satellite data and ground survey data to the cloud server. After the cloud decision-making layer fuses and analyzes the data, use the LSTM neural network for short-term deformation trend prediction, select the root mean square error, mean absolute error, and mean prediction accuracy as the accuracy evaluation indicators, and formulate an overall decision-making plan according to the fusion data and prediction results. When the real-time monitored deformation in any area of the fusion deformation field or the LSTM model trend exceeds the threshold, the cloud triggers and pushes warning information.

[0032] Furthermore, the data processing process using the time-series InSAR technology in step (1) is as follows:

[0033] For N + 1 SAR images, arrange them in chronological order as t0, t1, t2........t N , and the corresponding time-series phase values are φ(t0), φ(t1), φ(t2)..........φ(t N ), where it is assumed that the deformation phase of φ(t0) is 0;

[0034] (1-1) Select one of them as the common master image, and register the other N images to this common master image, so that all SAR images have the same spatial coordinates;

[0035] (1-2) Set shorter time baselines and spatial baseline thresholds for interference combination to obtain M multi-looked differential interferograms, where Based on the master and slave image time points of each interferogram, establish an interferogram pair list. The first column is the acquisition time IE of the respective master images of the M differential interferograms j , and the second column is the acquisition time IS of the corresponding slave images j , and the corresponding unwrapped interference phase is δφ j , (j = 1, 2... M);

[0036] (1-3) For the j-th interferogram obtained from the master image t A and the slave image t B , taking the pixel with azimuth coordinate x and range coordinate r as an example, establish an interference phase model as shown in Equation (1):

[0037]

[0038] Among them, d(t A , x, r), d(t B , x, r) are the deformations at the pixel (x, r) corresponding to the t A image and the t B image respectively; B ⊥,j is the vertical baseline of the j-th interferogram; Δh is the elevation residual at the pixel (x, r) of the j-th interferogram; R is the distance from the pixel (x, r) of the j-th interferogram to the radar sensor; θ is the incident angle corresponding to the pixel (x, r) of the j-th interferogram; φ atm (t A , x, r), φ atm (t B , x, r) are the atmospheric delay phases at the pixel (x, r) corresponding to the t A image and the t B image respectively; Δn j is the noise phase of the j-th interferogram; λ represents the radar wavelength;

[0039] Without considering the atmospheric and noise phases, the model is simplified to the following formula:

[0040]

[0041] Convert the phase in Equation (2) to the average phase rate v between two image pairs, as shown in Equation (3):

[0042]

[0043] Model the average deformation rate between image pairs, then Equation (3) is transformed into:

[0044]

[0045] That is, it is simplified to:

[0046]

[0047] Apply the SVD decomposition to matrix B to obtain the minimum norm solution of the velocity vector v. According to the settlement velocity in each time interval, integrate the velocities in each time period in the time domain to obtain the deformation amounts in each time period. Based on the above large-scale time-series InSAR monitoring results, identify the surface deformation anomaly areas.

[0048] Furthermore, the specific process of step (3) is as follows:

[0049] (3-1) Unify the spatio-temporal reference of InSAR and the total station, ensure the time synchronization of InSAR and total station data, and convert the three-dimensional deformation data of the total station to the InSAR line-of-sight direction:

[0050] d LOS = d E cosθsinα + d N cosθcosα + d U sinθ (8)

[0051] where θ is the radar incident angle, α is the satellite heading angle, d E , d N , d U are the deformation amounts in the east, north, and vertical directions;

[0052] (3-2) Use the total station data to correct the InSAR atmospheric delay error. At the total station monitoring point, the atmospheric delay phase difference is:

[0053] Δφ atm = φ InSAR - φ robot (9)

[0054] where φ InSAR represents the original interferometric phase of InSAR, and φ robot is the LOS phase converted from the actually measured deformation of the total station;

[0055] Use the Δφ atm at the total station points to estimate the atmospheric phase distribution over the entire area through Kriging interpolation:

[0056]

[0057] where the weight λ i is optimized by the semi-variogram to ensure unbiasedness and minimum variance;

[0058] Deduct the atmospheric delay phase difference from the original InSAR interferometric phase:

[0059] φ corrected = φ InSAR - Δφ atm (x,y) (11)

[0060] The total station uploads the coordinate data (X, Y, Z) every 10 minutes, and the cloud platform updates the GCP database in real time for InSAR phase resolution iterative optimization until the atmospheric residual phase ≤ 0.5 rad;

[0061] (3 - 3) Perform weighted averaging on the deformation data of the coincident points to obtain the fused deformation value ΔH f ,

[0062] ΔH f = W InSAR gΔH InSAR + W robot gΔH robot (12)

[0063] where, ΔH InSAR and ΔH robot are the deformation values of InSAR and the total station respectively, and the weight coefficients are calculated by the following formula:

[0064]

[0065] where, σ InSAR and σ robot are the standard deviations of the measurement errors of InSAR and the total station respectively.

[0066] Furthermore, the formulas for the root mean square error, mean absolute error, and mean prediction accuracy in step (4) are as follows:

[0067]

[0068]

[0069]

[0070] In the formula, represents the predicted value of the i-th high coherence point, Y i represents the monitored value of this point, and Pnum represents the total number of high coherence points in the settlement area.

[0071] Furthermore, the warning information includes

[0072] Level 1 warning: Notify by email;

[0073] Level 2 warning: Start encrypted observation by the total station;

[0074] Level 3 early warning: The linkage emergency management system generates a reinforcement plan;

[0075] If the fusion data detects sudden deformation, the robot sampling frequency is automatically increased.

[0076] Advantages of the present invention compared with the prior art:

[0077] (1) InSAR realizes surface deformation monitoring at the square kilometer level through multi-band satellites (X / C / L), quickly locking abnormal areas (such as 10 km 2 subsidence area); the measurement robot cluster then performs millimeter-level encrypted monitoring on the ground (100 points / km 2 ), forming a "surface-point" linkage, improving the accuracy of hidden danger identification, and solving the problem that it is difficult for manpower to cover high-risk areas.

[0078] (2) The measurement robot realizes second-level data transmission, making up for the InSAR processing delay (originally 7 - 15 days); the cloud integrates real-time monitoring data (100Hz sampling) and historical deformation sequences (such as 5-year data), predicts the 48-hour trend based on the LSTM model, improves the disaster response efficiency, and breaks through the spatio-temporal limitations of traditional monitoring. Description of the Drawings

[0079] Figure 1 It is the system architecture diagram of the present invention.

[0080] Figure 2 It is the data fusion flow chart of InSAR and measurement robot of the present invention.

[0081] Figure 3 It is the effect diagram of integrated monitoring of mine slopes. Detailed Embodiment

[0082] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0083] The system architecture of the present invention mainly includes three-layer structure design: spatial perception layer, ground execution layer, and cloud decision-making layer.

[0084] Content 1: Spatial perception layer

[0085] 1.1 Multi-band SAR data reception

[0086] X - band data reception has a high frequency, short wavelength, and a resolution that can reach the sub - meter level. It can clearly distinguish small ground buildings and facilities and is suitable for urban fine monitoring. C - band data reception has a moderate wavelength and strong ability to penetrate clouds, rain, snow, and fog. It can obtain data all - weather and has unique advantages in monitoring the topography and landforms of vegetated areas. L - band data reception has a long wavelength and a wide coverage range. It has a significant effect on monitoring large - scale surface deformations and is suitable for regional deformation census.

[0087] 1.2 Data pre - processing

[0088] Denoising processing is carried out on the received SAR data. An adaptive filtering algorithm is used to reduce noise interference, improve data quality, and ensure the accuracy of subsequent processing. Data correction is performed, including geometric correction and radiometric correction, to eliminate errors caused by factors such as satellite attitude and orbital deviation, making the data consistent with the actual geographical coordinates and enhancing data usability.

[0089] 1.3 Temporal InSAR processing technology

[0090] The SBAS technology selects multiple master images and performs interferometric processing on multi - temporal SAR data of the same area to obtain the overall deformation information of the area. It is suitable for monitoring large - scale and slowly deforming areas. This technology can generate a regional deformation rate map, intuitively showing the differences in deformation speeds in different regions, providing an important basis for regional stability assessment, such as monitoring crustal deformation in earthquake - prone areas.

[0091] The PSI technology focuses on high - precision single - point deformation monitoring. Stable scatterers are selected as monitoring points, such as building corners and bridge supports. Micro - deformation information is extracted from multi - temporal data. It can achieve deformation monitoring accuracy at the millimeter - level or even sub - millimeter - level and can be used for the health monitoring of critical infrastructure, such as monitoring the small displacement of a dam body to early - warn potential risks. The SBAS / PSI processing module uses temporal interferometry technology, selects stable scatterers as monitoring points, and extracts micro - deformation information from multi - temporal data for large - scale overall monitoring.

[0092] Content Two: Ground execution layer

[0093] 2.1 Robot functions and performance

[0094] Each intelligent measurement robot is equipped with multiple sensors, such as a laser rangefinder and an inertial measurement unit, which can real - time obtain its own position, attitude, and surrounding environment information, providing data support for autonomous navigation. It has high - precision measurement capabilities. The laser ranging accuracy can reach the millimeter - level, and it can accurately measure parameters such as the size and distance of ground objects, meeting the requirements of high - precision measurement.

[0095] 2.2 Autonomous navigation ability

[0096] Using advanced navigation algorithms, combined with map data and sensor information, the robot can autonomously plan the optimal travel path, avoid complex terrains and obstacles, and improve the measurement efficiency. In a dynamic environment, such as a construction site or a natural disaster site, the robot can adjust the navigation strategy in real time to ensure the smooth completion of tasks, with strong adaptability.

[0097] 2.3 Networking Mechanism and Cooperative Operation

[0098] The robots establish connections through wireless communication modules, adopt self-organizing network protocols, and automatically adjust the connection relationship according to signal strength and distance to form a stable communication network. The network has a self-healing function. When a certain node fails or the communication link is interrupted, other nodes can automatically reorganize the network to ensure uninterrupted data transmission.

[0099] Through dynamic networking, the robots can share measurement data and position information in real time, and cooperate to complete complex tasks, such as grid measurement of a large area, to improve work efficiency. In cooperative operations, the robots can dynamically adjust work strategies according to the task progress and data feedback to achieve intelligent collaboration and improve the overall measurement accuracy and reliability.

[0100] Content Three: Cloud Decision-making Layer

[0101] 3.1 Data Fusion Method

[0102] Adopt a feature-level fusion method to extract common features of satellite data and ground measurement data, such as deformation trends, etc., for fusion processing to reduce data redundancy. Use the coordinate of the measurement robot to invert the local atmospheric delay, generate a regional atmospheric phase map through spatial interpolation, and correct the InSAR atmospheric phase. Apply the weighted average algorithm to assign different weights according to the reliability and importance of the data source to achieve data fusion and improve the accuracy and reliability of the fusion result.

[0103] 3.2 Application of Fusion Results

[0104] The fused data can be used to generate a high-precision terrain deformation map, visually display the regional deformation situation, and provide an important basis for geological disaster early warning, such as early warning of disasters such as landslides and settlements. Provide accurate data support for engineering construction. For example, in large-scale infrastructure construction, optimize the construction plan according to the fused data to ensure project safety and quality.

[0105] 3.3 LSTM Neural Network Deformation Prediction Model

[0106] The Long Short-Term Memory (LSTM) network is a special type of recurrent neural network that is good at processing time series data. Through memory cells and gating mechanisms, it effectively solves the problem of gradient vanishing in traditional RNNs and can accurately capture long-term dependencies in time series. In deformation prediction, LSTM can learn and fuse the temporal patterns of deformation data, such as seasonal changes and periodic fluctuations, to provide an accurate model for future deformation prediction.

[0107] 3.4 Data Transmission and System Collaboration

[0108] The spatial perception layer transmits the processed satellite data to the cloud decision-making layer through a communication link, and the measurement data of the ground execution layer is also uploaded to the cloud in real time. After the cloud decision-making layer fuses and analyzes the data, it feeds back the instructions to the ground execution layer. The data transmission uses an encryption protocol to ensure data security and integrity and prevent data leakage and tampering.

[0109] Based on the fused data and prediction results, the cloud decision-making layer formulates an overall decision-making plan, such as disaster warning levels, engineering intervention measures, etc., and issues decision-making instructions to the ground execution layer. The ground execution layer adjusts the measurement strategy or takes corresponding actions according to the instructions, such as evacuating dangerous areas or carrying out emergency reinforcement during disaster warnings, to achieve closed-loop system collaboration.

[0110] A multi-scale deformation monitoring method based on the collaborative operation of InSAR and total stations specifically includes the following steps:

[0111] Step 1: Receive multi-band SAR data from the SAR satellite constellation, denoise and correct the data, and then process it using time series InSAR technology.

[0112] Process 10+ SAR images, generate an initial deformation field using time series InSAR, with a deformation accuracy of 5-10 mm. In this invention, the SBAS-InSAR technology is taken as an example to introduce the process of time series InSAR data processing.

[0113] Suppose there are N+1 SAR images, which are arranged in chronological order as t0, t1, t2........t N , and their corresponding time series phase values are φ(t0), φ(t1), φ(t2)..........φ(t N ). Among them, the deformation phase of φ(t0) is set to 0.

[0114] (1) Select one of them as the common master image, and register the other N images to this common master image, so that all SAR images have the same spatial coordinates.

[0115] (2) Set shorter time baselines and spatial baseline thresholds for interference combination to obtain M differentially interferometric images after multi-look processing, where Establish an interference pair list based on the acquisition time points of the master and slave images of each interferogram. The first column is the acquisition time IE of the respective master images of the M differentially interferometric images j , and the second column is the acquisition time IS of the corresponding slave images j , and the corresponding unwrapped interferometric phase is δφ j , (j = 1, 2... M).

[0116] (3) For the j-th interferogram obtained from the master image t A and the slave image t B , taking the pixel with azimuth coordinate x and range coordinate r as an example, establish an interferometric phase model as shown in Equation (1):

[0117]

[0118] Among them, d(t A , x, r), d(t B , x, r) are the deformations at the pixel (x, r) corresponding to the t A image and the t B image respectively; B ⊥,j is the vertical baseline of the j-th interferogram; Δh is the elevation residual at the pixel (x, r) of the j-th interferogram; R is the distance from the pixel (x, r) of the j-th interferogram to the radar sensor; θ is the incident angle corresponding to the pixel (x, r) of the j-th interferogram; φ atm (t A , x, r), φ atm (t B , x, r) are the atmospheric delay phases at the pixel (x, r) corresponding to the t A image and the t B image respectively; Δn j is the noise phase of the j-th interferogram; λ represents the radar wavelength.

[0119] Without considering the atmospheric and noise phases, the model can be simplified to the following formula:

[0120]

[0121] Since the φ(t0), φ(t1), φ(t2)..........φ(t N ) obtained by solving the model δφ = Aφ will have discontinuities in the calculation results due to large jumps in the deformation phases of some pixel points in the interference pairs. To obtain a physically meaningful settlement sequence, convert the phase in Equation (2) to the average phase rate v between two image pairs, as shown in Equation (3):

[0122]

[0123] Model the average deformation rate between image pairs. Then, Equation (3) can be transformed into:

[0124]

[0125] That is, it is simplified to:

[0126]

[0127] Apply SVD decomposition to matrix B, and the minimum norm solution of the velocity vector v can be obtained. According to the settlement velocity of each time interval, integrate the velocity in the time domain for each time period to obtain the deformation amount of each time period.

[0128] Based on the above large-scale time-series InSAR monitoring results, the surface deformation anomaly area can be identified (set the threshold deformation rate ≥ 8 mm / month or cumulative displacement ≥ 30 mm).

[0129] Step 2: Dynamic networking of measurement robots

[0130] According to the spatial distribution of the anomaly area, dynamically deploy the measurement robot network. The node spacing D is adaptively adjusted according to the deformation gradient, and the calculation formula

[0131] D = 500 / (1 + 0.2×G) (6)

[0132] where G is the deformation rate, with the unit of mm / month.

[0133] Deliver the measurement robot to the target position, unfold the tripod and complete the initial positioning. The measurement robot realizes millimeter-level positioning through multi-sensor fusion (GNSS + inertial navigation + laser ranging). The core formula is the dynamic positioning error model

[0134] ΔP(t) = αgv(t) + βga(t) + γgw(t) + ε (7)

[0135] where α, β, γ are the sensor weight coefficients, v(t) is the velocity, a(t) is the acceleration, w(t) is the angular velocity, and ε is the noise error (usually ≤ 0.3 mm).

[0136] After positioning, conduct network communication tests to ensure that the data transmission delay ≤ 1 s. The measurement robot can be selected as Trimble SX10 (angular measurement accuracy 0.5″, ranging accuracy 1 mm + 1 ppm); the communication module can be selected as Huawei 5G industrial router (time delay ≤ 20 ms). Prisms are laid at key points or the target is directly automatically tracked and aimed for real-time three-dimensional coordinate monitoring.

[0137] Step 3: Data fusion and collaborative monitoring

[0138] (1) Unify the spatio-temporal reference of InSAR and the total station, and ensure the time synchronization of InSAR and total station data (error < 1 hour). In addition, convert the three-dimensional deformation data of the total station to the line-of-sight direction (LOS direction) of InSAR:

[0139] d LOS =d E cosθsinα + d N cosθcosα + d U sinθ (8)

[0140] where θ is the radar incident angle, α is the satellite heading angle, d E , d N , d U are the deformation amounts in the east, north, and vertical directions.

[0141] (2) Use the total station data to correct the InSAR atmospheric delay error. At the total station monitoring point, the atmospheric delay phase difference is:

[0142] Δφ atm =φ InSAR -φ robot (9)

[0143] where φ InSAR represents the original interferometric phase of InSAR, and φ robot is the LOS phase (only including the true deformation) converted from the actually measured deformation of the total station.

[0144] Use the Δφ atm at the total station point to estimate the atmospheric phase distribution over the entire area through Kriging spatial interpolation:

[0145]

[0146] where the weight λ i is optimized by the semivariogram to ensure unbiasedness and minimum variance.

[0147] Deduct the atmospheric delay phase difference from the original interferometric phase of InSAR:

[0148] φ corrected =φ InSAR -Δφ atm (x, y) (11)

[0149] The total station uploads the coordinate data (X, Y, Z) every 10 minutes, and the cloud platform updates the GCP database in real time. Perform iterative optimization of InSAR phase resolution until the atmospheric residual phase ≤ 0.5 rad.

[0150] (3)Perform monitoring data fusion on the overlapping points to improve the monitoring accuracy of the points. The fusion monitoring accuracy can reach 1.2 mm. Specifically, perform weighted averaging on the deformation data of the overlapping points to obtain the fused deformation value ΔH f ,

[0151] ΔH f =W InSAR gΔH InSAR +W robot gΔH robot (12)

[0152] where ΔH InSAR and ΔH robot are the deformation values of InSAR and the total station respectively. The weight coefficients are calculated by the following formula:

[0153]

[0154] where σ InSAR and σ robot are the standard deviations of the measurement errors of InSAR and the total station respectively.

[0155] Step 4: Cloud decision-making and task scheduling

[0156] (1) Transmit the satellite data and ground measurement data to the cloud server. After the cloud decision-making layer fuses and analyzes the data, use the LSTM neural network for short-term deformation trend prediction, and shorten the warning response time to 48 hours. The present invention adopts a single-feature, many-to-many LSTM neural network prediction model. The model mainly includes four modules: an input layer, a hidden layer (LSTM layer and Dropout layer), network training, and an output layer. In the network structure, the LSTM layer and Dropout layer are designed as stacked hidden layers.

[0157] Among them, the input layer will complete the preprocessing of the fused deformation monitoring data, including the division and standardization of samples to meet the data requirements of the LSTM layer; the hidden layer learns the non-linear correlation between settlement time series through LSTM unit cells, and uses the Dropout layer to randomly inactivate a certain proportion of neurons, thereby reducing the mutual connection between neurons and preventing the prediction model from overfitting; the network training module uses the ADAM (Adaptive Moment Estimation) optimizer to optimize the network weights; the output layer includes the trained network and its predicted values.

[0158] To measure the prediction accuracy of the network model, select the root mean square error (Mean Square Error, MSE), mean absolute error (Mean Absolute Error, MAE), and average prediction accuracy as the accuracy evaluation indicators. The calculation formulas are as follows:

[0159]

[0160]

[0161]

[0162] In the formula, represents the predicted value of the i-th high coherence point, Y i represents the monitored value of this point, and Pnum represents the total number of high coherence points in the settlement area.

[0163] (2) The cloud decision-making layer formulates an overall decision-making plan based on the fusion data and prediction results. When real-time monitored deformation occurs in any area of the fusion deformation field or the trend of the LSTM model exceeds the threshold (such as 15 mm / month set for slope monitoring), the cloud triggers and pushes warning information:

[0164] Level 1 warning (yellow): Notify by email;

[0165] Level 2 warning (orange): Start encrypted observation by the measuring robot (frequency changes from 1 time / hour to 1 time / 5 minutes);

[0166] Level 3 warning (red): Link the emergency management system to generate a reinforcement plan.

[0167] If sudden deformation is detected in the fusion data (such as a rate change > 20%), automatically increase the robot sampling frequency to 50 Hz.

[0168] Experiments show that the fusion monitoring accuracy of the present invention can reach 1.2 mm, and the warning response time is shortened to 48 hours, which is applicable to scenarios such as landslides, mining area subsidence, and dam deformation.

[0169] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-scale deformation monitoring system based on the collaborative operation of InSAR and total station, characterized in that, It includes a spatial perception layer, a ground execution layer, and a cloud decision-making layer. The spatial perception layer is used to receive multi-band SAR data, perform data preprocessing, and generate a regional deformation rate map and identify surface deformation anomaly areas using time-series InSAR processing technology. The ground execution layer dynamically deploys a network of measurement robots according to the spatial distribution of the anomaly areas. Each intelligent measurement robot is equipped with multiple sensors to obtain its own position, attitude, and surrounding environment information in real time. The robots establish connections through wireless communication modules and use self-organizing network protocols to automatically adjust the connection relationship according to signal strength and distance to form a stable communication network. The cloud decision-making layer uses a feature-level fusion method to extract the common features of satellite data and ground measurement data for fusion processing, then uses an LSTM neural network for short-term deformation trend prediction. Finally, according to the fusion data and prediction results, it formulates an overall decision-making plan and sends decision instructions to the ground execution layer. When the real-time monitored deformation in any area of the fusion deformation field or the trend of the LSTM model exceeds the threshold, the cloud decision-making layer triggers and pushes warning information.

2. The multi-scale deformation monitoring system based on the collaborative operation of InSAR and total station according to claim 1, characterized in that The spatial perception layer includes an SAR data reception module, a data preprocessing module, and a time-series InSAR processing module. The SAR data reception module is used to receive multi-band SAR data from an SAR satellite constellation. The data preprocessing module is used to denoise the received SAR data and then perform data correction to eliminate the errors caused by satellite attitude and orbital deviation factors, making the data consistent with the actual geographical coordinates. The time-series InSAR processing module selects multiple master images to perform interference processing on multi-temporal SAR data of the same area, obtains the overall deformation information of the area, generates a regional deformation rate map, intuitively shows the deformation speed differences in different areas, and identifies surface deformation anomaly areas.

3. A multi-scale deformation monitoring system based on the collaborative operation of InSAR and a total station according to claim 2, characterized in that The spatial perception layer also includes an SBAS / PSI processing module. Through time-series interferometric measurement technology, stable scatterers are selected as monitoring points, and minute deformation information is extracted through multi-temporal data for large-scale overall monitoring.

4. A multi-scale deformation monitoring system based on the collaborative operation of InSAR and a total station according to claim 1, characterized in that The sensors equipped on the intelligent measurement robots include a laser rangefinder and an inertial measurement unit. Using advanced navigation algorithms, combined with map data and sensor information, the robots autonomously plan the optimal travel path, adjust the navigation strategy in real time, and through dynamic networking, the robots share measurement data and position information in real time to collaborate on complex tasks.

5. The multi-scale deformation monitoring system based on the collaborative operation of InSAR and total station according to claim 1, wherein The cloud decision-making layer includes a data fusion processing module, an LSTM neural network deformation prediction module, a data transmission and system coordination processing module, and a warning module. The data fusion processing module uses a feature-level fusion method to extract the common features of satellite data and ground measurement data, fuse the deformation trends, inversely calculate the local atmospheric delay amount using the coordinates of the measurement robots, generate a regional atmospheric phase map through spatial interpolation, correct the InSAR atmospheric phase, and use a weighted average algorithm to assign different weights according to the reliability and importance of the data sources to achieve data fusion. The LSTM neural network deformation prediction module includes an input layer, a hidden layer, a network training module, and an output layer. Among them, the LSTM layer and the Dropout layer are designed as a stacked hidden layer. The input layer will complete the preprocessing of the fused deformation monitoring data, including the division and standardization of samples to meet the data requirements of the LSTM layer. The hidden layer learns the non-linear correlation between settlement time series through LSTM unit cells and uses the Dropout layer to randomly inactivate a certain proportion of neurons. The network training module uses the ADAM optimizer to optimize the network weights. The output layer includes the trained network and its predicted values. The data transmission and system collaborative processing module receives the satellite data processed by the space perception layer and the measurement data of the ground execution layer, formulates an overall decision-making plan based on the fused data and prediction results, and issues decision-making instructions to the ground execution layer. The early warning module is used to trigger and push early warning information when any region in the fused deformation field monitors deformation in real time or the trend of the LSTM model exceeds the threshold.

6. A multi-scale deformation monitoring method based on the collaborative operation of InSAR and total station, which is implemented by using the monitoring system described in any one of claims 1-5, characterized in that It includes the following steps: (1) Receive multi-band SAR data of the SAR satellite constellation, denoise and correct the data, and then use the time-series InSAR technology to obtain the overall deformation information of the region and identify the abnormal regions of surface deformation. (2) Measure the dynamic networking of robots (2-1) Dynamically deploy a measurement robot network according to the spatial distribution of the abnormal regions. The node spacing D is adaptively adjusted according to the deformation gradient. D = 500 / (1 + 0.2×G) where G is the deformation rate, with the unit of mm / month. (2-2) Deliver the measurement robot to the target position, unfold the triangular support and complete the initial positioning. The measurement robot realizes millimeter-level positioning through multi-sensor fusion. The dynamic positioning error model is: ΔP(t) = αgv(t) + βga(t) + γgw(t) + ε where α, β, and γ are sensor weight coefficients, v(t) is the velocity, a(t) is the acceleration, w(t) is the angular velocity, and ε is the noise error. (2-3) Conduct network communication tests after positioning, lay prisms at key points or directly automatically track and aim at the target for real-time three-dimensional coordinate monitoring. (3) Data fusion and collaborative monitoring Unify the spatio-temporal reference of InSAR and the measurement robot, correct the InSAR atmospheric delay error using the measurement robot data, estimate the atmospheric phase distribution of the entire region using spatial interpolation method with the robot points, and use the weighted average algorithm for the overlapping points, and assign different weights according to the reliability and importance of the data sources for monitoring data fusion. (4) Cloud decision-making and task scheduling Transmit the satellite data and ground measurement data to the cloud server. After the cloud decision-making layer fuses and analyzes the data, use the LSTM neural network to predict the short-term deformation trend. Select the root mean square error, mean absolute error, and mean prediction accuracy as the accuracy evaluation indicators. According to the fused data and prediction results, formulate an overall decision-making plan. When any region in the fused deformation field monitors deformation in real time or the trend of the LSTM model exceeds the threshold, the cloud triggers and pushes early warning information.

7. A multi-scale deformation monitoring method based on the collaborative operation of InSAR and a total station according to claim 6, characterized in that The data processing process of the time-series InSAR technology in step (1) is as follows: For N + 1 SAR images, they are arranged in chronological order as t0, t1, t2........t N , and their corresponding temporal phase values are φ(t0), φ(t1), φ(t2)..........φ(t N ), where the deformation phase of φ(t0) is set to 0; (1-1) Select one of them as the common master image, and register the other N images to this common master image, so that all SAR images have the same spatial coordinates; (1-2) Set shorter time baseline and spatial baseline thresholds for interference combination to obtain M multi-looked differential interferograms, where Establish an interference pair list based on the master and slave image time points of each interferogram. The first column is the acquisition time IE of the respective master images of the M differential interferograms j , and the second column is the acquisition time IS of the corresponding slave images j , and the corresponding unwrapped interference phase is δφ j , (j = 1, 2... M); (1-3) For the main image t A and the slave image t B For the j-th interferogram obtained, taking the pixel with the azimuth coordinate as x and the range coordinate as r as an example, an interferometric phase model is established as shown in Equation (1): where d(t A , x, r) and d(t B , x, r) are the deformations at the pixel (x, r) corresponding to the t A image and the t B image respectively; B ⊥,j is the vertical baseline of the j-th interferogram; Δh is the elevation residual at the pixel (x, r) of the j-th interferogram; R is the distance from the pixel (x, r) of the j-th interferogram to the radar sensor; θ is the incident angle corresponding to the pixel (x, r) of the j-th interferogram; φ atm (t A , x, r) and φ atm (t B , x, r) are the atmospheric delay phases at the pixel (x, r) corresponding to the t A image and the t B image respectively; Δn j is the noise phase of the j-th interferogram; λ represents the radar wavelength; Without considering the atmospheric and noise phases, the model is simplified to the following formula: Convert the phase in formula (2) into the average phase rate v between two image pairs, as shown in formula (3): Model the average deformation rate between image pairs, then formula (3) is transformed into: That is, it is simplified to: Apply SVD decomposition to matrix B to obtain the minimum norm solution of the velocity vector v. According to the settlement velocity in each time interval, integrate the velocities in each time period in the time domain to obtain the deformation amount in each time period. Identify the surface deformation anomaly area based on the above large-scale time-series InSAR monitoring results.

8. A multi-scale deformation monitoring method based on the collaborative operation of InSAR and a total station according to claim 6, characterized in that The specific process of step (3) is as follows: (3-1) Unify the spatio-temporal benchmarks of InSAR and the total station, ensure the time synchronization of InSAR and total station data, and convert the three-dimensional deformation data of the total station to the InSAR line-of-sight direction: d LOS = d E cosθsinα + d N cosθcosα + d U sinθ(8) where θ is the radar incident angle, α is the satellite heading angle, and d E , d N , d U are the deformation amounts in the east, north, and vertical directions; (3-2) Use the total station data to correct the InSAR atmospheric delay error. At the total station monitoring point, the atmospheric delay phase difference is: Δφ atm = φ InSAR - φ robot (9) Among them, φ InSAR represents the original interferometric phase of InSAR, and φ robot is the LOS phase converted from the deformation actually measured by the total station; Using Δφ of the robot points atm , estimating the atmospheric phase distribution of the entire area through Kriging interpolation: Among them, the weight λ i is optimized by the semivariogram to ensure unbiasedness and minimum variance; Deduct the atmospheric delay phase difference from the original InSAR interference phase: φ corrected = φ InSAR - Δφ atm (x,y)(11) The total station uploads coordinate data (X, Y, Z) every 10 minutes, and the cloud platform updates the GCP database in real time to perform InSAR phase solution iterative optimization until the atmospheric residual phase ≤ 0.5 rad; (3-3)Perform weighted averaging on the deformation data of the coincident points to obtain the fused deformation value ΔH f , ΔH f = W InSAR gΔH InSAR + W robot gΔH robot (12) where ΔH InSAR and ΔH robot are the deformation values of InSAR and the total station respectively, and the weight coefficients are calculated by the following formula: where, σ InSAR and σ robot are the standard deviations of the measurement errors of InSAR and the robot, respectively.

9. A multi-scale deformation monitoring method based on the collaborative operation of InSAR and total station according to claim 6, characterized in that The formulas for the root mean square error, mean absolute error, and mean prediction accuracy in step (4) are as follows: In the formula, represents the predicted value of the i-th high coherence point, and Y i represents the monitored value of this point, and Pnum represents the total number of high coherence points in the settlement area.

10. A multi-scale deformation monitoring method based on the collaborative operation of InSAR and total station according to claim 6, characterized in that, The warning information includes Level 1 warning: Notify by email; Level 2 warning: Start encrypted observation of the total station; Level 3 warning: Link the emergency management system to generate a reinforcement plan; If sudden deformation is detected in the fused data, automatically increase the sampling frequency of the robot.

Images