CN110333494A - A kind of InSAR timing deformation prediction method, system and relevant apparatus - Google Patents
A kind of InSAR timing deformation prediction method, system and relevant apparatus Download PDFInfo
- Publication number
- CN110333494A CN110333494A CN201910286195.3A CN201910286195A CN110333494A CN 110333494 A CN110333494 A CN 110333494A CN 201910286195 A CN201910286195 A CN 201910286195A CN 110333494 A CN110333494 A CN 110333494A
- Authority
- CN
- China
- Prior art keywords
- deformation
- model
- timing
- convolutional neural
- neural networks
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/88—Radar or analogous systems specially adapted for specific applications
- G01S13/885—Radar or analogous systems specially adapted for specific applications for ground probing
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/88—Radar or analogous systems specially adapted for specific applications
- G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
- G01S13/90—Radar or analogous systems specially adapted for specific applications for mapping or imaging using synthetic aperture techniques, e.g. synthetic aperture radar [SAR] techniques
- G01S13/9021—SAR image post-processing techniques
- G01S13/9023—SAR image post-processing techniques combined with interferometric techniques
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/417—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section involving the use of neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Remote Sensing (AREA)
- Radar, Positioning & Navigation (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Electromagnetism (AREA)
- Computer Networks & Wireless Communication (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A kind of InSAR timing deformation prediction method provided herein, comprising: obtain the history deformation data of target object；History deformation data is inputted into depth convolutional neural networks model, output timing deformation prediction result；Wherein, the network structure of depth convolutional neural networks model has merged the network structure of U-Net model and the network structure of DenseNet model.This method will acquire the history deformation data input depth convolutional neural networks model of target object, output timing deformation prediction result.The network structure of U-Net model and the network structure of DenseNet model are merged due to the network structure of depth convolutional neural networks model, so the precision of the timing deformation prediction result of output is higher.The application also provides a kind of InSAR timing deformation prediction system, equipment and computer readable storage medium, all has above-mentioned beneficial effect.
Description
Technical field
This application involves InSAR timing deformation prediction field, in particular to a kind of InSAR timing deformation prediction method is
System, equipment and computer readable storage medium.
Background technique
As the mankind surround InSAR (Synthetic Aperture Radar Interferometry, synthetic aperture thunder
Up to interferometry) concentrate carry out to the research in terms of earth's surface deformation monitoring, the mankind more recognize to earth's surface deformation prediction study
Importance, deformation prediction abnormal conditions early warning and in time processing in play central role, have important research
Value.Some methods based on model are applied to the deformation prediction of certain known variant trend.For example, hyperbolic model
Being successfully applied to prediction, there is the extra large land of filling out of deceleration trend to settle；Grey Markov model is with being used to predict Beijing
Face sedimentation.
Currently, more common InSAR timing deformation prediction method is based on experience and numerical model, but by this method
After the deformation results and terrestrial reference point data of prediction carry out accuracy evaluation, the ratio of precision of the timing deformation prediction of this method is obtained
It is lower.
Therefore, how to improve the precision of InSAR timing deformation prediction is the skill of those skilled in the art's urgent need to resolve
Art problem.
Summary of the invention
The purpose of the application is to provide a kind of InSAR timing deformation prediction method, system, equipment and computer-readable storage
Medium can be improved the precision of InSAR timing deformation prediction.
In order to solve the above technical problems, the application provides a kind of InSAR timing deformation prediction method, comprising:
Obtain the history deformation data of target object；
The history deformation data is inputted into depth convolutional neural networks model, output timing deformation prediction result；Wherein,
The network structure of the depth convolutional neural networks model has merged the network structure of U-Net model and the net of DenseNet model
Network structure.
Preferably, described that the history deformation data is inputted depth convolutional neural networks model, output timing deformation is pre-
Survey result, comprising:
Model training is carried out using depth convolutional neural networks method, obtains the depth convolutional neural networks model；
The history deformation data is inputted into the depth convolutional neural networks model, exports the timing deformation prediction knot
Fruit.
Preferably, after obtaining the depth convolutional neural networks model, further includes:
Whether the precision for verifying the depth convolutional neural networks model reaches preset precision threshold；
If so, executing described by the history deformation data input depth convolutional neural networks model, output institute
The step of stating timing deformation prediction result.
Preferably, the history deformation data for obtaining target object, comprising:
The COSMO-SkyMed image of the Ground Deformation of preset duration is obtained, and as the history deformation data.
Preferably, the history deformation data for obtaining target object, comprising:
The multi-space baseline synthetic aperture radar image of the Ground Deformation of preset duration is obtained, and as the history deformation
Data.
The application also provides a kind of InSAR timing deformation prediction system, comprising:
History deformation data obtains module, for obtaining the history deformation data of target object；
Timing deformation prediction result output module, for the history deformation data to be inputted depth convolutional neural networks mould
Type, output timing deformation prediction result；Wherein, the network structure of the depth convolutional neural networks model has merged U-Net mould
The network structure of type and the network structure of DenseNet model.
Preferably, the timing deformation prediction result output module, comprising:
Model training unit obtains the depth volume for carrying out model training using depth convolutional neural networks method
Product neural network model；
Timing deformation prediction result output unit, for the history deformation data to be inputted the depth convolutional Neural net
Network model exports the timing deformation prediction result.
Preferably, the InSAR timing deformation prediction system further include:
Whether precision test module, the precision for verifying the depth convolutional neural networks model reach preset precision
Threshold value；
The timing deformation prediction result output unit is specially the essence for working as the depth convolutional neural networks model
When degree reaches the preset precision threshold, then the history deformation data is inputted into the depth convolutional neural networks model,
Export the unit of the timing deformation prediction result.
The application also provides a kind of equipment, comprising:
Memory and processor；Wherein, the memory is for storing computer program, and the processor is for executing institute
The step of InSAR timing deformation prediction method described above is realized when stating computer program.
The application also provides a kind of computer readable storage medium, and the computer-readable recording medium storage has computer
The step of program, the computer program realizes InSAR timing deformation prediction method described above when being executed by processor.
A kind of InSAR timing deformation prediction method provided herein, comprising: obtain the history texturing variables of target object
According to；The history deformation data is inputted into depth convolutional neural networks model, output timing deformation prediction result；Wherein, described
The network structure of depth convolutional neural networks model has merged the network structure of U-Net model and the network knot of DenseNet model
Structure.
This method will acquire the history deformation data input depth convolutional neural networks model of target object, output timing shape
Become prediction result.Due to the network structure of the depth convolutional neural networks model merged U-Net model network structure and
The network structure of DenseNet model, so the precision of the timing deformation prediction result of output is higher.The application also provides one kind
InSAR timing deformation prediction system, equipment and computer readable storage medium, all have above-mentioned beneficial effect, no longer superfluous herein
It states.
Detailed description of the invention
In order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
The embodiment of application for those of ordinary skill in the art without creative efforts, can also basis
The attached drawing of offer obtains other attached drawings.
Fig. 1 is a kind of flow chart of InSAR timing deformation prediction method provided by the embodiment of the present application；
Fig. 2 is a kind of structural block diagram of InSAR timing deformation prediction system provided by the embodiment of the present application.
Specific embodiment
The core of the application is to provide a kind of InSAR timing deformation prediction method, can be improved InSAR timing deformation prediction
Precision.Another core of the application is to provide a kind of InSAR timing deformation prediction system, equipment and computer-readable storage medium
Matter.
To keep the purposes, technical schemes and advantages of the embodiment of the present application clearer, below in conjunction with the embodiment of the present application
In attached drawing, the technical scheme in the embodiment of the application is clearly and completely described, it is clear that described embodiment is
Some embodiments of the present application, instead of all the embodiments.Based on the embodiment in the application, those of ordinary skill in the art
Every other embodiment obtained without making creative work, shall fall in the protection scope of this application.
Currently, more common InSAR timing deformation prediction method is based on experience and numerical model, but by this method
After the deformation results and terrestrial reference point data of prediction carry out accuracy evaluation, the ratio of precision of the timing deformation prediction of this method is obtained
It is lower.InSAR timing deformation prediction method provided by the present application, can be improved the precision of InSAR timing deformation prediction, specifically asks
With reference to Fig. 1, Fig. 1 is a kind of flow chart of InSAR timing deformation prediction method provided by the embodiment of the present application, when the InSAR
Sequence deformation prediction method specifically includes:
S101, the history deformation data for obtaining target object；
The embodiment of the present application obtained the history deformation data of target object before this, did not limited specifically at this target object
It is fixed, corresponding setting should be made according to the actual situation by those skilled in the art, which can be Ground Deformation.Into one
Step ground, which can be the Ground Deformation for presetting area, such as target object can be Hong Kong International Airport (Hong
Kong International Airport, HKIA) Ground Deformation.Due to applicant in the experimental study stage mainly by Hong Kong
The Ground Deformation of International airport is studied as target object, so hereafter when necessary can be with the earth's surface shape of Hong Kong International Airport
It is illustrated for change situation, but it should be noted that the target object in the application is not restricted to the ground of Hong Kong International Airport
Table deformation.Hong Kong International Airport is that two island (Chek Lap Kok, Lam Chau) is being relied on to fill out sea formation, is in the world
One of maximum artificial island, using airport region as the survey region of this experiment.Airport is there are two types of deformation tendency, and first
It kind is after the completion of nineteen ninety-five filling out sea, due to the trend of reclaimed land Continuous Settlement caused by the consolidation banketed, long-term is filled out
Extra large ground settlement trend typically appears as deceleration trend, and hyperbolic model can be used and predicted.Second of trend be due to
The season linear thermal expansion of high level infrastructure caused by temperature change (such as building and high building).
Further, the history deformation data of above-mentioned acquisition target object, generally includes: obtaining the earth's surface shape of preset duration
The COSMO-SkyMed image of change, and as history deformation data.The embodiment of the present application does not limit above-mentioned preset duration specifically
It is fixed, corresponding setting should be made according to the actual situation by those skilled in the art, such as in the experimental study stage that Hong Kong is international
The COSMO-SkyMed image of the Ground Deformation (timing node is from 2013.10.04 to 2016.01.04) on airport is as history shape
Parameter evidence.Specifically, 28 COSMO-SkyMed images (timing node is from 2013.10.04 to 2016.01.04) has been selected
The time series (as shown in table 1, table 1 is COSMO-SkyMed image list and relevant parameter) of deformation is extracted, and will
2014.12.06 the main image as acquisition image.In order to increase density a little, inventor has detected PS and DS point simultaneously, finally
More than 1,500,000 points are detected, and have obtained the time series of corresponding height, averaged deformation speed and deformation.Deformation is flat
The range of equal speed is that -28.2mm/y arrives 15.1mm/y, standard deviation 2.2mm/y.
1 COSMO-SkyMed image list of table and relevant parameter
Since DCNN (Deep Convolutional Neural Networks, depth convolutional neural networks) requires data
It is uniformly distributed on time and Spatial Dimension, so the above discrete point cannot be used directly for deep learning.In order to obtain in space
Equally distributed data on scale, in ArcGIS using inverse distance-weighting (IDW) method to each time series deformation map into
Row space interpolation.And since DCNN training process is more time-consuming, in order to improve computational efficiency, so setting interpolated resolution to
6m, the i.e. half of SAR image resolution ratio.Then temporal interpolation is carried out to deformation data using cubic spline curve, it will
2013.10.04 the image at time point has obtained on time dimension equally distributed deformation data (between the time as starting image
It is divided into one month).In order to verify monitoring data, 90 terrestrial reference point datas that it is provided with Airport Authority are compared
Compared with.The elevation and deformation quantity of datum mark can be calculated with the terrestrial reference data of two different time points and be obtained, that is, utilize 2013/
2014 and 2015 terrestrial reference data, time interval are about 1 year.Acquisition time and interpolation institute by ground data
The Time Inconsistency of the InSAR data obtained, inventor are compared with the two time closest time series data, gained
The root-mean-square error (RMSE) arrived is 3.3mm, this is consistent with the InSAR deformation precision under the influence of by uncertain factor.Therefore,
The time series of the deformation of InSAR detection after temporal-spatial interpolating can be as the list entries of subsequent study.
Further, the history deformation data of above-mentioned acquisition target object, generally includes: obtaining the earth's surface shape of preset duration
The multi-space baseline synthetic aperture radar image of change, and as history deformation data.
Multidate synthetic aperture radar interferometry (Multi-temporal Synthetic Aperture Radar
Interferometry, MT-InSAR) it is to utilize multi-space baseline synthetic aperture radar (Synthetic Aperture
Radar, SAR) image measures the time series of Ground Deformation.Compared to difference synthetic aperture radar interferometry
(Differential SAR Interferometry, D-InSAR) technology, MT-InSAR technology can be effectively reduced space-time and go
Correlation effect significantly improves the precision of deformation monitoring.Most representative MT-InSAR method is permanent scattering interference method
(Persistent Scatterer Interferometry, PSInSAR), it uses single main image data set identify PS point
(such as artificiality and day stone) can measure sub-meter grade landform and grade deformation.In order to improve consistency, reduce landform
Influence for deformation monitoring has scholar to propose Small Baseline Subset method, empty when being reassembled as single main image collection to have shorter
Multiple main image data sets of baseline；In order to extract the deformation of unstable settling zone, STAMPS (Standford
Methodfor Persistent Scatter) algorithm is suggested, and this method is suitable for monitoring a large amount of pixels and association in time
The lower region of property, such as volcano etc.；In order to reinforce the measurement for Low coherence region (such as mountain area or vegetation region),
SqueeSAR technology is suggested, it is the extension carried out to the method for PSInSAR, can be used for detecting to adjacent pixel with similar
Distributed diffusion (Distributed Scatterer, DS) point of scattering mechanism.In previous research, also to MT-InSAR
The precision of deformation monitoring carried out assessment, and European Space Agency has carried out test representative twice: PSIC4 and
Terrafirma Validation project.It is worth noting that, difference test team is not using in the test of the latter
Measurement result with processing method has carried out cross validation.The results show that the standard deviation of speed of deformation is in 0.4-0.5mm/y model
In enclosing, the deformation of time series is within the scope of 1.1-4mm.This result is ERS and ENVISAT data in speed of deformation and deformation
The substantially situation of timing provides reference, this demonstrates the accuracy of MT-InSAR technology theoretically to a certain extent.
S102, history deformation data is inputted into depth convolutional neural networks model, output timing deformation prediction result；Its
In, the network structure of depth convolutional neural networks model has merged the network structure of U-Net model and the net of DenseNet model
Network structure.
InSAR timing deformation prediction can be considered as a Time-space serial study and forecasting problem in the embodiment of the present application, grind
Study carefully regional deformation monitoring data temporal-spatial interpolating the result is that X × Y image, the deformation of the entire survey region of t moment can be with table
It is shown as a two-dimensional matrix Xt∈Rt×x×y, give m continuous timing deformation quantity [Xt+1,Xt+2,…,Xt+m] it is used as input item, n is a
Timing deformation [Xt+m+1,Xt+m+2,…,Xt+m+n] it is used as output item, w indicates the external data of deformation incitant and impact factor
Number, prediction deformation will form a new timing results, by changing m and n Parameter analysis input and output deformation number to pre-
The influence of precision is surveyed, and determines optimal m and n.
The depth convolutional neural networks model in the embodiment of the present application is illustrated below: the ability to express of deep learning
Define the tightness degree of relationship between model and problem to be solved.For different training content and task type, network knot
The difference of structure can largely influence its performance.In numerous model structures of deep learning, convolutional neural networks
(Convolutional Neural Network) is current deep learning towards in the algorithm of image procossing, widely used height
Network structure is imitated, it comprises the computing units being largely classified, and handle visual correlation information by different level in the feed forward mode.It is each
Layer computing unit can be considered for extracting in input picture the feature filters of characteristic in a certain respect, be exported a series of corresponding special
Levy the characteristic pattern of expression.Convolutional neural networks do the operation feature in image convolution and pond with it, it is easier to capture data not
Vertic features, such as the invariance of translation and rotation, greatly reduce the number of parameters in network structure.The present invention utilizes convolution
Classical Open Framework Caffe in field of neural networks, to improve model tormulation ability to greatest extent as target, construct it is a kind of towards
The depth convolutional neural networks structure that InSAR deformation develops.Depth convolutional neural networks model in the application is based on U-Net mould
Type and DenseNet modelling.Wherein, U-net model is a kind of full connection convolutional Neural net towards image, semantic segmentation
Network has the characteristics that network structure is symmetrical and convolutional layer jump connects, shows in the task of many computer visions excellent
Anisotropic energy.The network structure of DenseNet model alleviates deep neural network gradient disperse problem, improves feature in network
In propagation efficiency, emphasize that network characterization is multiplexed, allow the network to it is more preferable must learn and capture space in input data according to
Rely feature.In the model of the application, the characteristics of we have merged U-Net model and DenseNet model simultaneously, is removed last
Active coating after one layer of convolutional layer allows network to export continuous Time-space serial value；By the continuous convolution layer of U-Net model
Replace with the structure of DenseNet model.More importantly, we increase external input layer in network the last layer, with mould
The preamble output of type is merged.External input includes geologic data, project data, hydrographic data, meteorological data etc., is added outer
Portion's input allows model preferably to learn the time-space relationship that different external conditions act on lower earth's surface variation, improves the pre- of model
Survey ability.
The embodiment of the present application is not especially limited the timing deformation prediction result of output, depending on needing according to the actual situation.
For example, by the prediction result of the depth convolutional neural networks model of experimental stage the application and the prediction result of hyperbolic model into
Row comparison, it is specific as follows: on Hong Kong International Airport's runway the prediction result of some settlement point P1 can reflect DCNN method for
The superiority and inferiority of continuous sedimentation prediction result.The averaged deformation speed of P1 point sedimentation is -10.2mm/y, in the embodiment of the present application, due to
Period is shorter, the approximate linear trend of the deformation data of P1.From 2014.04.24,2015.12.02 and 2017.02.24 tri-
The terrestrial reference data of a timing node are it is found that the sinking speed of P1 point is gradually being accelerated.It is assumed that the deformation results of monitoring are
In correct situation, the root mean square of the data from 2014.10.04 to 2016.01.04 in the deformation data of prediction can be calculated
Error (RMSE), i.e., interior error, interior error are able to reflect the consistent degree between monitoring data and prediction data.Wherein, DCNN
Error is 0.6mm in method, and the interior error of hyperbolic model method is that 0.3mm illustrates to predict much smaller than the precision of monitoring deformation
Deformation results and monitoring deformation have preferable consistency.It then calculates prediction deformation and terrestrial reference point data exists
2015.12.02 with the absolute error of 2017.02.24 time point deformation, i.e., outer error.Since ground data and InSAR data are adopted
The timing node of collection is not completely the same, therefore the data of timing node arest neighbors has been selected to compare.DCNN method is pre-
Survey generate outer error be 2.4mm, hyperbolic model prediction generate outer error be 3.2mm, this result substantially with monitoring deformation
Precision it is consistent.The above results show that the inside and outside error of DCNN and hyperbolic model prediction result is smaller, with monitoring data
Deformation data it is consistent, institute can be used in two ways predict continuous sedimentation.
P2 (elevation 30.1m) point is located on building, can be used for assessing prediction of the DCNN for building seasonal variety
Ability.It can be derived that from the timing diagram of P2 point, thermal expansion causes the time series of deformation to present apparent seasonal trend.
During parameter evaluation, using linear deformation model, linear deformation speed can be carried out in the case where there is seasonal deformation
Correction.Therefore, the average speed of P2 point deformation is not 0mm/y, but 1.5mm/y, it can be deduced that DCNN method can be effective
The seasonal deformation that prediction is generated due to thermal expansion, however hyperbolic model then unpredictable seasonal deformation this out.DCNN
The interior error that method generates is 0.3mm, and the interior error that hyperbolic model method generates is 1mm, illustrates predicting seasonal deformation
When, DCNN method will be obviously due to hyperbolic model method.In fact, the performance of the prediction technique based on model is largely
Depending on the applicability of model, hyperbolic model is suitable for the prediction of continuous sedimentation, but is not particularly suited for the pre- of seasonal deformation
It surveys.And DCNN method, using data as driven factor, advantage is to be totally independent of used model, adaptively
Excavate time mode.The seasonal variety for the next year that DCNN method is predicted is less than detection data.Due to asking for poor fitting
Topic, single linear deformation model may make the seasonal variety of prediction generate deviation.Lesser seasonal variety may be by
The interference between thermal expansion and sedimentation in monitoring data during originally incorrect seasonal variety or Associate learning
Caused by.This error can be adjusted by introducing temperature data, but this is within the scope of the discussion of this research.
Finally the spatial model of the deformation of prediction is analyzed, the spatial model of the two deformation generally presents
Consistency.For example, the average value of the accumulative deformation at 2015.12.04 time point increase to from-the 4mm of monitoring data deformation map it is pre-
- the 3.7mm of measured data deformation map, standard deviation are reduced to 5.42mm from 5.46mm.Compared with the millimeter class precision of monitoring data, this
The small gap of kind can be ignored substantially, illustrate that DCNN method prediction result is in close proximity to the deformation monitored.
2015.12.04 accumulative deformation histogram has also reflected prediction deformation and has monitored the consistency between deformation, micro- in histogram
Small difference shows that in the correct situation of monitoring data, the result of DCNN method tends to underestimate the amount of sedimentation.Hong Kong is international
The deformation of airport overwhelming majority point is smaller or nothing, it is believed that is the principal component of Hong Kong International Airport's dynamic change.It is a certain when predicting
When the deformation of a point, the space-time characteristic of all the points can be added in trained queue by DCNN method, therefore in Hong Kong International Airport
Under the influence of principal component, most of point tends to tend towards stability.
Further, above-mentioned that history deformation data is inputted into depth convolutional neural networks model, output timing deformation prediction
As a result, generally including: carrying out model training using depth convolutional neural networks method, obtain depth convolutional neural networks model；
History deformation data is inputted into depth convolutional neural networks model, output timing deformation prediction result.The embodiment of the present application will instruct
Practice multi-source data and be divided into 80% training group and 20% validation group, tests the degree of convergence and generalization ability of training pattern.Training
Process will be separately operable in two computing platforms, be the CUDA (Compute based on NVIDIA independently built respectively
Unified Device Architecture) the GPU cloud computing service that provides of GPU parallel computation environment and Amazon
(Amazon's EC2GPU).Synchronous operation will improve the time interval of model adjustment, accelerate the rhythm that experiment is carried out, and be used for
Contrast test independently builds the performance of parallel computing platform.
It further, usually can also include: verifying depth convolution mind after obtaining depth convolutional neural networks model
Whether the precision through network model reaches preset precision threshold；History deformation data is inputted into depth convolution if so, executing
The step of neural network model, output timing deformation prediction result.Above-mentioned preset precision threshold is not especially limited at this,
Corresponding setting should be made according to the actual situation by those skilled in the art.In the experimental study stage, by carrying out prediction result
Verifying to determine the precision of depth convolutional neural networks model.For example, utilizing the inside and outside of all monitoring points of Hong Kong International Airport
Error verifies the accuracy of prediction result, analyzes large error Producing reason.Experiment show that the average value of interior error is
0.3mm, compared with the millimeter class precision of InSAR data, this error be can be ignored.DCNN method can regard input data
For the data being absolutely correct, and the monitoring data result of input is approached in learning process as far as possible.Therefore, the deformation of input
Time series needs carry out pretreatment, so that it is guaranteed that the high-precision of prediction result.The region of interior error larger (> 1mm) is mainly
Building or big settling zone have higher fluctuation in the deformation data in these regions, therefore are highly susceptible to error
It influences.In conjunction with 90 terrestrial reference point datas on 2015 and 2016/2017 year two timing node, prediction can be verified
The validity of value of the deformation between 2016.02.04 to 2017.01.04.In order to avoid generating local error, datum mark is uniform
It is distributed in airport.The experimental results showed that the relationship between prediction deformation and terrestrial reference data, the two are generally consistent
, RMSE=3mm, Pearson correlation coefficient R=0.55.Biggish outer error (> 5mm) mainly appears on building, can
It can be because that is, ground data may be in the case where thermal expansion condition does not occur at building bottom position caused by season linear thermal expansion deviation
Collected data are set, and corresponding InSARS point may be to be selected in the building top position with larger thermal expansion.
In general, the precision of the time series of the deformation of prediction is suitable with the monitoring precision of deformation, this shows the effective of short-term forecast
Property.Because prediction deformation with monitoring deformation it is close, other than error can more be monitored deformation precision influence.Precision is not
It is uncertain consistent in certainty and InSAR monitoring data, such as the deformation of reference point, position and time difference etc..
A kind of InSAR timing deformation prediction method that the application proposes selects Hong Kong International Airport as survey region, uses
The reclaimed land of duration sedimentation and the building of expanded by heating influential effect verify this method as research main object.Data
The deformation data for selecting COSMO-SkyMed image from 2013.10.04 to 2016.01.04, in the effect of DCNN method
Under, predict the new deformation data from 2014.10.04 to 2017.01.04.Two masters can be obtained from the above experiment
Want conclusion:
1.DCNN is unrelated with the time mode of deformation, it can not only predict continuity sedimentation but also can predict by seasonal heat
Expansion effects have periodic deformation process, and hyperbolic ray mode can only predict that continuity settles, this embodies data-driven
The advantage of method.Also demonstrating DCNN method simultaneously can be as a kind of universal method, to predict under more complicated time mode
Deformation process.
The average interior error (0.3mm) that 2.DCNN method generates is negligible, shows DCNN output result and input knot
Fruit is close.Precision using 90 ground reference point data verification prediction deformation is 3mm, suitable with monitoring data precision, is shown
Validity of the DCNN method in short-term forecast.
Therefore, DCNN method is that a kind of tool is promising, and the method that can be used in predicting InSAR deformation data can
Applied to establishing early warning system.And in fact, deformation is influenced by many external factor, the thickness of extra large filler is such as filled out,
Temperature, land type etc., this research also lack the study to such data.In research from now on, these external datas will be received
Enter and further quantify the relationship between deformation and influence factor, improves the robustness of prediction model.
A kind of InSAR timing deformation prediction method provided by the present application, the history deformation data that will acquire target object are defeated
Enter depth convolutional neural networks model, output timing deformation prediction result.Due to the network knot of depth convolutional neural networks model
Structure has merged the network structure of U-Net model and the network structure of DenseNet model, so the timing deformation prediction knot of output
The precision of fruit is higher.
It to a kind of InSAR timing deformation prediction system provided by the embodiments of the present application, equipment and computer-readable deposits below
Storage media is introduced, InSAR timing deformation prediction system, equipment and computer readable storage medium described below and above
The InSAR timing deformation prediction method of description can correspond to each other reference.
Referring to FIG. 2, Fig. 2 is a kind of structural frames of InSAR timing deformation prediction system provided by the embodiment of the present application
Figure；The InSAR timing deformation prediction system includes:
History deformation data obtains module 201, for obtaining the history deformation data of target object；
Timing deformation prediction result output module 202, for history deformation data to be inputted depth convolutional neural networks mould
Type, output timing deformation prediction result；Wherein, the network structure of depth convolutional neural networks model has merged U-Net model
The network structure of network structure and DenseNet model.
Based on the above embodiment, timing deformation prediction result output module 202 in the present embodiment, generally includes:
Model training unit obtains depth convolution mind for carrying out model training using depth convolutional neural networks method
Through network model；
Timing deformation prediction result output unit, for history deformation data to be inputted depth convolutional neural networks model,
Output timing deformation prediction result.
Based on the above embodiment, the InSAR timing deformation prediction system usually can also include: in the present embodiment
Whether precision test module, the precision for verifying depth convolutional neural networks model reach preset precision threshold
Value；
The precision that timing deformation prediction result output unit specially works as depth convolutional neural networks model reaches preset
When precision threshold, then history deformation data is inputted into depth convolutional neural networks model, the list of output timing deformation prediction result
Member.
The application also provides a kind of equipment, comprising: memory and processor；Wherein, memory is for storing computer journey
The step of sequence, processor is for realizing the InSAR timing deformation prediction method of above-mentioned any embodiment when executing computer program.
The application also provides a kind of computer readable storage medium, and computer-readable recording medium storage has computer journey
Sequence, the step of InSAR timing deformation prediction method of above-mentioned any embodiment is realized when computer program is executed by processor.
The computer readable storage medium may include: USB flash disk, mobile hard disk, read-only memory (Read-Only
Memory, ROM), random access memory (Random Access Memory, RAM), magnetic or disk etc. is various to deposit
Store up the medium of program code.
Each embodiment is described in a progressive manner in specification, the highlights of each of the examples are with other realities
The difference of example is applied, the same or similar parts in each embodiment may refer to each other.For embodiment provide system and
Speech, since it is corresponding with the method that embodiment provides, so being described relatively simple, related place is referring to method part illustration
?.
Professional further appreciates that, unit described in conjunction with the examples disclosed in the embodiments of the present disclosure
And algorithm steps, can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate hardware and
The interchangeability of software generally describes each exemplary composition and step according to function in the above description.These
Function is implemented in hardware or software actually, the specific application and design constraint depending on technical solution.Profession
Technical staff can use different methods to achieve the described function each specific application, but this realization is not answered
Think beyond the scope of this invention.
The step of method described in conjunction with the examples disclosed in this document or algorithm, can directly be held with hardware, processor
The combination of capable software module or the two is implemented.Software module can be placed in random access memory (RAM), memory, read-only deposit
Reservoir (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technology
In any other form of storage medium well known in field.
It to a kind of InSAR timing deformation prediction method provided herein, system, equipment and computer-readable deposits above
Storage media is described in detail.Specific examples are used herein to illustrate the principle and implementation manner of the present application,
The description of the example is only used to help understand the method for the present application and its core ideas.It should be pointed out that for this technology
For the those of ordinary skill in field, under the premise of not departing from the application principle, several improvement can also be carried out to the application
And modification, these improvement and modification are also fallen into the protection scope of the claim of this application.
Claims (10)
1. a kind of InSAR timing deformation prediction method characterized by comprising
Obtain the history deformation data of target object；
The history deformation data is inputted into depth convolutional neural networks model, output timing deformation prediction result；Wherein, described
The network structure of depth convolutional neural networks model has merged the network structure of U-Net model and the network knot of DenseNet model
Structure.
2. InSAR timing deformation prediction method according to claim 1, which is characterized in that described by the history deformation
Data input depth convolutional neural networks model, output timing deformation prediction result, comprising:
Model training is carried out using depth convolutional neural networks method, obtains the depth convolutional neural networks model；
The history deformation data is inputted into the depth convolutional neural networks model, exports the timing deformation prediction result.
3. InSAR timing deformation prediction method according to claim 2, which is characterized in that obtaining the depth convolution
After neural network model, further includes:
Whether the precision for verifying the depth convolutional neural networks model reaches preset precision threshold；
If so, execution is described to input the depth convolutional neural networks model for the history deformation data, when exporting described
The step of sequence deformation prediction result.
4. InSAR timing deformation prediction method according to claim 1, which is characterized in that the acquisition target object
History deformation data, comprising:
The COSMO-SkyMed image of the Ground Deformation of preset duration is obtained, and as the history deformation data.
5. InSAR timing deformation prediction method according to claim 1, which is characterized in that the acquisition target object
History deformation data, comprising:
The multi-space baseline synthetic aperture radar image of the Ground Deformation of preset duration is obtained, and as the history texturing variables
According to.
6. a kind of InSAR timing deformation prediction system characterized by comprising
History deformation data obtains module, for obtaining the history deformation data of target object；
Timing deformation prediction result output module, for the history deformation data to be inputted depth convolutional neural networks model,
Output timing deformation prediction result；Wherein, the network structure of the depth convolutional neural networks model has merged U-Net model
The network structure of network structure and DenseNet model.
7. InSAR timing deformation prediction system according to claim 6, which is characterized in that the timing deformation prediction knot
Fruit output module, comprising:
Model training unit obtains the depth convolution mind for carrying out model training using depth convolutional neural networks method
Through network model；
Timing deformation prediction result output unit, for the history deformation data to be inputted the depth convolutional neural networks mould
Type exports the timing deformation prediction result.
8. InSAR timing deformation prediction system according to claim 7, which is characterized in that further include:
Whether precision test module, the precision for verifying the depth convolutional neural networks model reach preset precision threshold
Value；
The precision that the timing deformation prediction result output unit specially works as the depth convolutional neural networks model reaches
When to the preset precision threshold, then the history deformation data is inputted into the depth convolutional neural networks model, output
The unit of the timing deformation prediction result.
9. a kind of equipment characterized by comprising
Memory and processor；Wherein, the memory is for storing computer program, the processor by execute it is described based on
It realizes when calculation machine program such as the step of InSAR timing deformation prediction method described in any one of claim 1 to 5.
10. a kind of computer readable storage medium, which is characterized in that the computer-readable recording medium storage has computer journey
Sequence is realized when the computer program is executed by processor as InSAR timing deformation described in any one of claim 1 to 5 is pre-
The step of survey method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910286195.3A CN110333494A (en) | 2019-04-10 | 2019-04-10 | A kind of InSAR timing deformation prediction method, system and relevant apparatus |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910286195.3A CN110333494A (en) | 2019-04-10 | 2019-04-10 | A kind of InSAR timing deformation prediction method, system and relevant apparatus |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110333494A true CN110333494A (en) | 2019-10-15 |
Family
ID=68138994
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910286195.3A Pending CN110333494A (en) | 2019-04-10 | 2019-04-10 | A kind of InSAR timing deformation prediction method, system and relevant apparatus |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110333494A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110838303A (en) * | 2019-11-05 | 2020-02-25 | 南京大学 | Voice sound source positioning method using microphone array |
| CN111105347A (en) * | 2019-11-19 | 2020-05-05 | 贝壳技术有限公司 | Method, device and storage medium for generating panoramic image with depth information |
| CN111860158A (en) * | 2020-06-15 | 2020-10-30 | 中国测绘科学研究院 | Time sequence InSAR high-coherence point extraction method fusing 1D-CNN and BiLSTM neural network |
| CN111929683A (en) * | 2020-07-28 | 2020-11-13 | 中国自然资源航空物探遥感中心 | Landslide deformation accumulation area prediction model generation method and landslide deformation accumulation area prediction method |
| CN112526515A (en) * | 2020-11-05 | 2021-03-19 | 山西省交通科技研发有限公司 | Surface deformation detection method based on synthetic aperture radar interferometry |
| US11055835B2 (en) | 2019-11-19 | 2021-07-06 | Ke.com (Beijing) Technology, Co., Ltd. | Method and device for generating virtual reality data |
| CN113139349A (en) * | 2021-05-12 | 2021-07-20 | 江西师范大学 | Method, device and equipment for removing atmospheric noise in InSAR time sequence |
| CN116299245A (en) * | 2023-05-11 | 2023-06-23 | 中山大学 | Time sequence InSAR deformation rate result self-adaptive mosaic correction method |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160019458A1 (en) * | 2014-07-16 | 2016-01-21 | Deep Learning Analytics, LLC | Systems and methods for recognizing objects in radar imagery |
| CN106940443A (en) * | 2017-01-16 | 2017-07-11 | 洪都天顺（深圳）科技有限公司 | Complicated city infrastructure PSInSAR deformation methods of estimation under the conditions of cloud-prone and raining |
| CN107292410A (en) * | 2016-03-30 | 2017-10-24 | 中国石油天然气股份有限公司 | tunnel deformation prediction method and device |
| CN108846199A (en) * | 2018-06-12 | 2018-11-20 | 华能澜沧江水电股份有限公司 | Ultra-high arch dam deformation space-time sequence prediction method based on space-time integration |
| CN109061641A (en) * | 2018-07-06 | 2018-12-21 | 中南大学 | A kind of InSAR timing earth's surface deformation monitoring method based on sequential adjustment |
| CN109543237A (en) * | 2018-10-29 | 2019-03-29 | 苏州科技大学 | Foundation pit displacement prediction method based on GA-BP neural network |
-
2019
- 2019-04-10 CN CN201910286195.3A patent/CN110333494A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160019458A1 (en) * | 2014-07-16 | 2016-01-21 | Deep Learning Analytics, LLC | Systems and methods for recognizing objects in radar imagery |
| CN107292410A (en) * | 2016-03-30 | 2017-10-24 | 中国石油天然气股份有限公司 | tunnel deformation prediction method and device |
| CN106940443A (en) * | 2017-01-16 | 2017-07-11 | 洪都天顺（深圳）科技有限公司 | Complicated city infrastructure PSInSAR deformation methods of estimation under the conditions of cloud-prone and raining |
| CN108846199A (en) * | 2018-06-12 | 2018-11-20 | 华能澜沧江水电股份有限公司 | Ultra-high arch dam deformation space-time sequence prediction method based on space-time integration |
| CN109061641A (en) * | 2018-07-06 | 2018-12-21 | 中南大学 | A kind of InSAR timing earth's surface deformation monitoring method based on sequential adjustment |
| CN109543237A (en) * | 2018-10-29 | 2019-03-29 | 苏州科技大学 | Foundation pit displacement prediction method based on GA-BP neural network |
Non-Patent Citations (3)
| Title |
|---|
| 李格: "基于深度学习的单目图像深度估计", 《中国优秀博硕士学位论文全文数据库（硕士）信息科技辑》 * |
| 王成等: "基于COSMO-SkyMed雷达影像的山地城市地表形变监测研究", 《电子测量技术》 * |
| 许石罗: "基于多源遥感影像的动态滑坡灾害空间预测模型研究", 《中国优秀博硕士学位论文全文数据库（博士）基础科学辑》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110838303A (en) * | 2019-11-05 | 2020-02-25 | 南京大学 | Voice sound source positioning method using microphone array |
| CN110838303B (en) * | 2019-11-05 | 2022-02-08 | 南京大学 | Voice sound source positioning method using microphone array |
| US11055835B2 (en) | 2019-11-19 | 2021-07-06 | Ke.com (Beijing) Technology, Co., Ltd. | Method and device for generating virtual reality data |
| CN111105347B (en) * | 2019-11-19 | 2020-11-13 | 贝壳找房(北京)科技有限公司 | Method, device and storage medium for generating panoramic image with depth information |
| CN111105347A (en) * | 2019-11-19 | 2020-05-05 | 贝壳技术有限公司 | Method, device and storage medium for generating panoramic image with depth information |
| US11721006B2 (en) | 2019-11-19 | 2023-08-08 | Realsee (Beijing) Technology Co., Ltd. | Method and device for generating virtual reality data |
| CN111860158A (en) * | 2020-06-15 | 2020-10-30 | 中国测绘科学研究院 | Time sequence InSAR high-coherence point extraction method fusing 1D-CNN and BiLSTM neural network |
| CN111860158B (en) * | 2020-06-15 | 2024-02-20 | 中国测绘科学研究院 | Time sequence InSAR high coherence point extraction method fusing 1D-CNN and BiLSTM neural network |
| CN111929683A (en) * | 2020-07-28 | 2020-11-13 | 中国自然资源航空物探遥感中心 | Landslide deformation accumulation area prediction model generation method and landslide deformation accumulation area prediction method |
| CN112526515A (en) * | 2020-11-05 | 2021-03-19 | 山西省交通科技研发有限公司 | Surface deformation detection method based on synthetic aperture radar interferometry |
| CN113139349A (en) * | 2021-05-12 | 2021-07-20 | 江西师范大学 | Method, device and equipment for removing atmospheric noise in InSAR time sequence |
| CN116299245A (en) * | 2023-05-11 | 2023-06-23 | 中山大学 | Time sequence InSAR deformation rate result self-adaptive mosaic correction method |
| CN116299245B (en) * | 2023-05-11 | 2023-07-28 | 中山大学 | Time sequence InSAR deformation rate result self-adaptive mosaic correction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110333494A (en) | A kind of InSAR timing deformation prediction method, system and relevant apparatus | |
| D'Ambrosio et al. | A cellular automata model for soil erosion by water | |
| CN103871102B (en) | A kind of road three-dimensional fine modeling method based on elevational point and road profile face | |
| CN106327577B (en) | Dimensional topography optimal method based on local curvature's entropy and quad-tree structure | |
| Hirabayashi et al. | Projection of glacier mass changes under a high-emission climate scenario using the global glacier model HYOGA2 | |
| CN105974479B (en) | The chromatography 2D/3D anisotropy Depth Domain velocity modeling method of GPU space lattice body | |
| CN102521835B (en) | Method for extracting point cloud data of tree height based on space three-dimensional template | |
| CN109840260A (en) | A kind of extensive real-time rainfall automatic Observation station ranked data processing method based on dynamic interpolation | |
| CN102147479A (en) | Modelling method of reservoir space physical property parameters | |
| CN113610301B (en) | Prediction algorithm for representing ground settlement time sequence evolution | |
| CN112633544A (en) | Predicted wind speed correction method and device | |
| CN111744183B (en) | Illumination sampling method and device in game and computer equipment | |
| CN114970302B (en) | Regional groundwater condition prediction method based on groundwater monitoring system | |
| CN108986024A (en) | A kind of regularly arranged processing method of laser point cloud based on grid | |
| US12014536B2 (en) | Target detection method and device for high-resolution remote sensing image | |
| CN108733952B (en) | Three-dimensional characterization method for spatial variability of soil water content based on sequential simulation | |
| CN107452059A (en) | A kind of three-dimensional interpolation modeling method, equipment and its storage device | |
| CN106846247A (en) | Space environment data multi scale display method, method for establishing model and device | |
| CN113987969A (en) | Land-water dynamic simulation model of multi-elevation scale water flow network | |
| CN116051777B (en) | Super high-rise building extraction method, apparatus and readable storage medium | |
| CN108062631A (en) | Urban waterlogging methods of risk assessment, device and terminal | |
| CN106875382A (en) | Tailings Dam size values extracting method and device | |
| KR101450524B1 (en) | Apparatus for applicating intertidal zone in the landuse information | |
| CN207008056U (en) | Difference cloud cluster extrapolation precipitation predicting device based on Hu squares | |
| CN108805981A (en) | A kind of branch channel type delta front training image method for building up |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191015 |