CN116109931A - Automatic urban ground subsidence recognition and classification method - Google Patents
Automatic urban ground subsidence recognition and classification method Download PDFInfo
- Publication number
- CN116109931A CN116109931A CN202310237618.9A CN202310237618A CN116109931A CN 116109931 A CN116109931 A CN 116109931A CN 202310237618 A CN202310237618 A CN 202310237618A CN 116109931 A CN116109931 A CN 116109931A
- Authority
- CN
- China
- Prior art keywords
- ground subsidence
- local
- points
- insar
- sedimentation
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 49
- 238000004062 sedimentation Methods 0.000 claims abstract description 64
- 238000010586 diagram Methods 0.000 claims abstract description 23
- 238000013145 classification model Methods 0.000 claims abstract description 20
- 238000000605 extraction Methods 0.000 claims abstract description 19
- 238000001514 detection method Methods 0.000 claims description 24
- 239000011159 matrix material Substances 0.000 claims description 15
- 238000002372 labelling Methods 0.000 claims description 8
- 238000005259 measurement Methods 0.000 claims description 8
- 238000012549 training Methods 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 238000005305 interferometry Methods 0.000 claims description 4
- 239000006185 dispersion Substances 0.000 claims description 3
- 238000005315 distribution function Methods 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000010998 test method Methods 0.000 claims description 2
- 238000004458 analytical method Methods 0.000 abstract description 4
- 239000008186 active pharmaceutical agent Substances 0.000 description 9
- 238000012544 monitoring process Methods 0.000 description 9
- 238000005516 engineering process Methods 0.000 description 5
- 239000013049 sediment Substances 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- 230000004927 fusion Effects 0.000 description 3
- 239000002689 soil Substances 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 238000007596 consolidation process Methods 0.000 description 2
- 230000000875 corresponding effect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000001556 precipitation Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000000926 separation method Methods 0.000 description 2
- 230000001629 suppression Effects 0.000 description 2
- 238000001276 Kolmogorov–Smirnov test Methods 0.000 description 1
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 238000012952 Resampling Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000001427 coherent effect Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000005336 cracking Methods 0.000 description 1
- 230000008021 deposition Effects 0.000 description 1
- 238000011049 filling Methods 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000013535 sea water Substances 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/176—Urban or other man-made structures
-
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/766—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using regression, e.g. by projecting features on hyperplanes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A30/00—Adapting or protecting infrastructure or their operation
- Y02A30/60—Planning or developing urban green infrastructure
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the field of geological disasters, and discloses an automatic urban ground subsidence identification and classification method, which comprises the following steps: constructing a two-layer hybrid network, and identifying high coherence points and low coherence points in the SAR image by utilizing a multi-key extraction technology; performing Kriging interpolation on the high coherence point and the low coherence point to obtain an InSAR deformation speed diagram, and separating local sedimentation and a corresponding InSAR local deformation speed diagram from the InSAR deformation speed diagram; constructing a local sedimentation sample set based on multi-source data based on the InSAR local deformation velocity diagram; carrying out feature extraction on the InSAR deformation speed and the local sedimentation sample set based on the multi-channel pyramid structure to obtain a feature image; constructing a ground subsidence recognition classification model, inputting a characteristic image into the ground subsidence recognition classification model to perform characteristic extraction, and realizing recognition and classification of ground subsidence. The invention provides technical support for constructing intelligent and modern geological disaster analysis and early warning.
Description
Technical Field
The invention belongs to the field of geological disasters, and particularly relates to an automatic urban ground subsidence recognition and classification method.
Background
Ground subsidence is a phenomenon that the earth surface slowly sinks due to underground solid or fluid exploitation, and when serious, geological disasters such as ground cracking, high-rise collapse, seawater backflow and the like can be caused. Thus, achieving automatic identification and classification of large-scale ground subsidence is critical to assessing potential geological disasters and improving disaster resistance in areas.
In recent years, in the aspect of ground subsidence monitoring, the DSInSAR technology remarkably improves the space-time resolution and the data processing precision of deformation monitoring. However, as the density of InSAR monitoring points increases, manual analysis tends to be time consuming and labor intensive when applied to large scale deformation monitoring. At present, most researches still rely on traditional methods such as threshold value, clustering, hot-spot-like and the like to analyze InSAR results, but the feasibility is not high and the efficiency is low. Therefore, it is urgent how to efficiently and intelligently post-process and analyze InSAR results.
Disclosure of Invention
The invention aims to provide an automatic identification and classification method for urban ground subsidence, which combines InSAR deformation monitoring results and multisource auxiliary data to realize automatic identification and classification of ground subsidence, and greatly improves feasibility and efficiency of early identification of geological disasters so as to solve the problems in the prior art.
In order to achieve the above purpose, the invention provides an automatic urban ground subsidence identification and classification method, which comprises the following steps:
constructing a two-layer hybrid network, and identifying high coherence points and low coherence points in the SAR image by utilizing a multi-point extraction technology based on the two-layer hybrid network;
performing Kriging interpolation on the high coherence point and the low coherence point to obtain an InSAR deformation speed diagram, and separating local sedimentation and a corresponding InSAR local deformation speed diagram from the InSAR deformation speed diagram;
constructing a local sedimentation sample set based on multi-source data based on the InSAR local deformation velocity diagram;
performing feature extraction on the InSAR deformation speed and the local sedimentation sample set based on a multi-channel pyramid structure to obtain a feature image;
constructing an initial ground subsidence recognition classification model, and training the ground subsidence recognition classification model based on the local subsidence sample set to obtain a target ground subsidence recognition classification model;
and inputting the characteristic images into the target ground subsidence recognition classification model to perform characteristic extraction so as to realize recognition and classification of ground subsidence.
Optionally, the process of obtaining the SAR image includes: registering the multi-baseline SLC images, and carrying out differential interferometry on the registered images to obtain SAR images.
Optionally, the process of identifying the high coherence point and the low coherence point using the multiple point extraction technique includes:
setting an initial amplitude dispersion threshold value in a first layer network to obtain a high-quality high-coherence point; based on the high-quality high-coherence points, a Delaunay triangular network and a self-adaptive encryption network are constructed, and then phase difference processing is carried out on the adjacent high-quality high-coherence points, so that a differential interference diagram is obtained, and further, the height and deformation speed information of the high-quality coherence points is obtained;
in the second layer network, a local star network is built in multiple directions by taking high-quality high-coherence points detected by the first layer network as reference points and taking four corner points of SAR images as starting points; and expanding the rest high coherence points and all low coherence points based on the local star network, and obtaining the height and deformation speed information of all the high coherence points and all the low coherence points in the vertical downward direction based on the direction of gradient movement.
Optionally, the process of obtaining the differential interferogram includes: and carrying out parameter estimation on a sensing matrix of the height and deformation speed information to be estimated by adopting an M estimator and a ridge estimator, and combining the reconstructed backscattering coefficients to obtain the differential interference diagram.
Optionally, the process of identifying all low coherence points includes: in a second layer network, identifying the same particle based on a KS (K-nearest neighbor) test method, estimating a covariance matrix by using a weight factor, and estimating a Phase based on a Phase-Linking method of a coherence coefficient to obtain all low coherence points;
the weight factor obtaining process comprises the following steps: introducing a first pixel point and a second pixel point, acquiring the statistical distance and the spatial distance of the two pixel points, and combining the statistical distance and the spatial distance by using a kernel function to acquire a weight factor; the statistical distance is the absolute difference of the largest empirical distribution function of the two pixel points, and the spatial distance is the Euclidean distance of the two pixel points.
Optionally, the process of separating out the local sedimentation comprises: and converting the InSAR deformation speed map from a space domain to a frequency domain based on two-dimensional fast Fourier transform, and separating regional sedimentation and local sedimentation by utilizing band-pass filtering.
Optionally, the process of constructing the local sedimentation sample set based on the multi-source data includes: and (3) presetting a local sedimentation area to meet the condition, screening local sedimentation samples in the InSAR local deformation velocity diagram, and attaching category labels to the local sedimentation samples based on the correlation of the spatial characteristics of sedimentation and auxiliary data.
Optionally, the process of constructing the initial ground subsidence recognition classification model includes: introducing the central coordinates of the predicted candidate areas, the height and width of the circumscribed rectangle of the predicted directional candidate areas and the offset of the top and the middle point of the right side of the circumscribed rectangle to describe the directional area prediction network, so as to generate a high-quality rotation candidate frame; and correcting the position of the rotation candidate frame based on the rotation candidate frame and the characteristic image, so as to obtain accurate positions and categories.
Optionally, the process of training the ground subsidence recognition classification model based on the local subsidence sample set includes: when the cross joint overlapping of one detection frame and any marking frame exceeds a first preset threshold value, or the detection frame and one marking frame have the highest area overlapping, and the area overlapping is higher than a second preset threshold value, the detection frame is a sedimentation target; otherwise, the detection frame is a background;
the area threshold is the ratio of the area of the union of the detection frame and the labeling frame to the area of the intersection of the detection frame and the labeling frame.
Optionally, the target ground subsidence identification classification model has a loss function L 1 The definition is as follows:
wherein N is the number of samples, F cls Is cross entropy loss and label of measurement label
And a network predicted tag p i Differences between F reg Is a detection frame delta for measuring network output by smoothing L1 loss i And labeling frame->
Offset between them.
The invention has the technical effects that:
the improvement of InSAR technology can generate a large number of deformation measurement points, so that the geological disaster assessment is very time-consuming and labor-consuming, and therefore, the invention provides a multichannel deep convolutional neural network model based on a multi-scale rotation detection frame, the model integrates InSAR measurement and multi-source auxiliary data, the separation of regional sedimentation and local sedimentation and the automatic identification and classification of local sedimentation are realized, and the technical support is provided for the construction of intelligent and modern geological disaster analysis and early warning of large data technology.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application, illustrate and explain the application and are not to be construed as limiting the application. In the drawings:
FIG. 1 is an extraction flow chart of an InSAR multi-point extraction technique in an embodiment of the present invention;
fig. 2 is a schematic diagram of a ground subsidence recognition classification model based on a scale rotation frame depth convolutional neural network in an embodiment of the invention.
Detailed Description
It should be noted that, in the case of no conflict, the embodiments and features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the accompanying drawings in conjunction with embodiments.
It should be noted that the steps illustrated in the flowcharts of the figures may be performed in a computer system such as a set of computer executable instructions, and that although a logical order is illustrated in the flowcharts, in some cases the steps illustrated or described may be performed in an order other than that illustrated herein.
Example 1
As shown in fig. 1, the embodiment provides an automatic urban ground subsidence identifying and classifying method, which includes:
step one: inSAR multi-key extraction technology suitable for large-range deformation robust monitoring
In consideration of the fact that the space-time distribution of the atmospheric phase of the sedimentation area is complex, the accuracy and stability of deformation estimation are improved by constructing a two-layer network and introducing a robust estimator (an M estimator and a ridge estimator). In addition, in order to solve the problem of missing measurement points caused by non-urban area coherence loss during large-scale detection, the embodiment adopts a multi-key extraction technology to jointly detect a high-coherence PS point and a low-coherence DS point, and Phase-Linking algorithm based on a coherence coefficient is used for reconstructing Phase and CPU/GPU parallel resolving parameters, so that refined deformation information is quickly inverted on the premise of ensuring estimation accuracy and robustness.
Specifically, with the aid of the SRTM DEM image, a multi-baseline Sentinel one number (Sentinel-1) single vision complex (SLC) image is strictly registered, and then differential interferometry is performed on the registered image by adopting 8×2 multi-vision pairs. In order to mitigate the stratified atmosphere effects in the interferogram, this embodiment uses the data of the generic atmospheric correction product GACOS for phase compensation during the interferometry process.
Aiming at monitoring of large-range deformation, the embodiment designs an InSAR multi-key extraction method based on a two-layer hybrid network to identify PS and DS points, and the method does not need to perform prior atmospheric signal removal in the whole research area.
In the first layer network, an initial amplitude dispersion threshold (typically less than 0.3) is set to select high quality PS candidate points, which are then connected in a dense Delaunay triangulation network. After phase-differentiating two adjacent PS candidate points, the signal model can be expressed as:
y＝AΥ
wherein y= [ y ] 1 ，…，y Ns ] T (Ns is the number of SAR images) represents the differential interferogram, y is the reconstructed backscatter coefficient, and a represents the sensing matrix containing the height and deformation speed information to be estimated.
In order to reduce the error of phase unwrapping, a robust M estimator is adopted to carry out parameter estimation on the height and deformation speed information, and whether arc segments are reserved or not during network construction is determined by a time coherence coefficient PS t1 =0.72 decision. The embodiment integrates the relative parameters through network adjustment, and adopts the ridge estimator to reduce the condition number of the adjustment matrix, thereby improving the stability of adjustment operation.
The second layer network uses the PS points detected by the first layer network as references to construct a local star network to extend the remaining PS points and all DS points. In order to ensure that coherent target points are detected to the maximum extent, the second-layer network expansion takes four corner points of the SAR image as starting points to perform multi-direction expansion, and in addition, the newly identified PS points are updated to reference points in time to realize space continuous expansion.
For detection of DS points, the embodiment adopts a simple KS test method to identify the same mass points, only statistical information is considered and spatial information is not considered in the traditional covariance matrix estimation, and in order to improve the accuracy of covariance matrix estimation, the statistical distance and the spatial distance of two pixel points (x 1 and x 2) are introduced as weight estimation covariance matrices. The statistical distance represents the maximum absolute difference of the empirical distribution function of the two pixel points, the spatial distance represents the Euclidean distance of the two pixel points, and a weight factor is calculated by combining the two by using a kernel function and is used for estimating a covariance matrix C:
wherein Ω is a set of co-dots; z is a complex observer vector; psi is complex matrix containing two-to-two image interference phases; w (x 1, x 2) is a weight factor; d (x 1, x 2) is the spatial distance; sigma (x 1, x 2) is the statistical distance; r is (r) d And r σ Is a scale factor.
After obtaining the weighted covariance matrix, the SquesAR uses the Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm to make a maximum likelihood estimate of the phase. However, this algorithm requires inverting the covariance matrix, but the covariance matrix may not be a positive definite matrix, which guarantees that the matrix is not invertible. To solve this problem, the present embodiment estimates the Phase using a Phase-Linking method based on a coherence coefficient more efficiently. The time coherence threshold for identifying other PS points and DS points is set to PS respectively t2 =0.7 and DS t2 ＝0.65。
Finally, the present embodiment calculates the deformation speed and time series deformation of all PS and DS points along the line of sight (LOS) direction. Assuming that the surface deformation is dominated by land subsidence, the present embodiment converts the LOS-oriented deformation to a vertical deformation by dividing it by cos θ (where θ is the angle of incidence). In steep regions, the present embodiment assumes that the grade motion is in a vertically downward direction, converting LOS-oriented deformation to vertically downward deformation by grade.
Step two: separating localized sedimentation and localized sedimentation that are superimposed
The coastal delta areas tend to be more susceptible to both regional and local sedimentation under the combined action of sediment consolidation and human activity. Regional sedimentation can be directly identified from the InSAR deformation speed map; however, detection of local sedimentation is very time consuming because of their wide distribution and variety of shapes. In addition, regional sedimentation often exists in superposition with local sedimentation, which can mask the local sedimentation, and potential geological disasters cannot be found in time.
To facilitate identification of localized sedimentation, the present embodiment separates regional sedimentation from localized sedimentation in a deformation velocity map. In the embodiment, the Kerling interpolation is carried out on the PS/DS points, and an InSAR deformation speed diagram with the spatial resolution of 60 meters is output. To avoid uncertainty in interpolation, this embodiment only retains pixels adjacent to the PS/DS point, with a buffer of 100 meters for subsequent processing. Based on the property that regional sedimentation has a low frequency, while local sedimentation has a high frequency, the present embodiment uses a two-dimensional Fast Fourier Transform (FFT) to transform the InSAR deformation velocity map from the spatial domain to the frequency domain, followed by band-pass filtering to separate the region and the local sedimentation. In addition, this embodiment eliminates extremely high frequency signals in the FFT process, considering that severe tropospheric effects (such as storms) can cause spike noise with extremely high frequencies.
Considering the spatial resolution of the Sentinel-1 image, the algorithm uses two windows of 20 x 20 pixels and 300 x 300 pixels in the frequency domain image to decompose long, medium, and short wavelength signals. The spatial wavelengths of the resulting signals were >3570 meters, 238-3570 meters, <238 meters, respectively, corresponding to regional sedimentation, local sedimentation and noise.
Step three: constructing a local sedimentation sample set based on multisource data
In addition to InSAR deformation monitoring results, this example also collected four other types of subsidence-related auxiliary data, including geologic lithology data (fourth interval sediment thickness, lithology type, and fault line), land cover data (land cover and sea fill distribution), terrain data (elevation, slope, and terrain humidity index), and climate data (average annual precipitation, soil humidity). These data will be used together for training of the network.
Considering that the local sedimentation areas on the InSAR deformation speed diagram are different in size, direction and shape, the embodiment uses a rotation detection frame to more accurately identify the range of the sedimentation sample. In addition, this example also assigns a class to each sediment sample, for a total of seven classes: sediment consolidation, sea fill settlement, construction settlement, farmland settlement, pond settlement, landslide movement and unclassified as shown in table 1:
TABLE 1
Specifically, the steps of labeling the local sedimentation sample are as follows:
(1) taking the InSAR deformation speed diagram as a reference, projecting all data to the same coordinate system WGS84_50N, and resampling (the spatial resolution is 60 multiplied by 60 m); (2) in the separated InSAR local deformation velocity diagram, screening a local sedimentation sample, and if a certain area meets the following conditions: a) The characteristic of a 'sedimentation bowl' is presented, namely the sedimentation speed is changed from the center to the periphery, and the maximum sedimentation speed is more than 25 mm/year; b) The area of the smallest external rectangle is between 0.05 and 5.5 square kilometers; c) The driving factor of sedimentation can be found in the multisource assistance data, then this region is considered to be a local sedimentation zone; (3) the labeled local sedimentation samples are attached with category labels, and the determination of the categories is mainly based on the spatial characteristics of sedimentation and the correlation of the spatial characteristics and auxiliary data (refer to table 1). For example, if the noted subsidence is located at both the soft soil deposition area and the artificial earth surface, and its boundaries are highly correlated with the artificial earth surface, it is designated as building subsidence.
Step four: multi-channel pyramid structure extraction feature fusing InSAR and multi-source auxiliary data
As described above, the present embodiment relates to 11 data sets in total, including InSAR deformation speed, geological lithology data (fourth-period sediment thickness, lithology type and fault line), land coverage data (land coverage and sea filling distribution), topographic data (elevation, gradient and topographic humidity index) and climatic data (average annual precipitation, soil humidity), and is typically three channels in natural image processing, and in order to enable effective fusion of the individual data sets and the contained features, the present embodiment designs a multi-channel pyramid structure extraction feature, and each input channel of the network corresponds to one specific data set. Considering that different data sets contain different semantic features, the multi-channel pyramid structure in the embodiment enables the output features to have both the visual information of the bottom layer and the semantic information of the high layer through a continuous up-sampling and cross-layer fusion mechanism.
Step five: constructing a ground subsidence recognition classification model based on a multi-scale rotating frame depth convolution neural network
As shown in fig. 2, the model is divided into two stages, the first stage mainly generates a high-quality rotation candidate frame, and the second stage is used for correcting the position of the rotation candidate frame to obtain an accurate position and category.
The first stage of directed area prediction network adopts six parameters (x, y, w, h, delta alpha, delta beta) for effectively describing a candidate area. (x, y) is the center coordinates of the predicted candidate region, (h, w) is the height and width of the bounding rectangle of the predicted directional candidate region, (Δα, Δβ): is the offset of the top and right midpoints of the bounding rectangle. By each position on the feature map obtained in step four, the directed region prediction network generates 3 rotation candidate boxes, so its regression branches contain 6×3=18 outputs, which will be further decoded. With these six parameters, the present embodiment may not introduce redundant network structures, and generate a high-quality rotation candidate frame.
The second stage takes the feature map generated in the fourth stage and the series of candidate boxes generated in the first stage as inputs, and obtains accurate positions and categories by correcting the positions of the rotation candidate boxes. Specifically, a rotation RoI alignment method is used to extract feature vectors of a fixed size, which are unchanged in rotation, from the matched feature graphs, each feature vector is input into two fully connected layers, and finally the probability that the candidate box belongs to k+1 class (K objects and 1 background, k=7 in the present embodiment) and the position offset thereof to the K class objects are output.
Step six: training the recognition model in the fifth step by using the sedimentation monitoring multi-source data sample set constructed in the third step, and inputting the fusion features in the fourth step into the recognition model in the fifth step for feature extraction, so that the ground sedimentation is recognized and classified in a large range.
In order to train the network, the present embodiment defines a settlement target and a background, respectively. A detected rotating box is considered a sedimentation target when it overlaps more than 0.5 with the cross-over union (IoU) of any one of the label boxes, or it overlaps a label box with the highest IoU and IoU is higher than 0.3; otherwise, it is considered as background. Furthermore, if the detected rotation box does not correspond to any label box, this sedimentation target is considered as missed.
area (gΣl) represents the area of the union of the detection frame and the label frame, and rea (gΣl) represents the area of the intersection of the detection frame and the label frame. Based on this, the loss function L of the target ground subsidence recognition classification model 1 The definition is as follows:
wherein N is the number of samples, F cls Is cross entropy loss and label of measurement label
And a network predicted tag p i Differences between F reg Is a detection frame delta for measuring network output by smoothing L1 loss i And labeling frame->
Offset between them.
The model is trained in an end-to-end fashion using random gradient descent while optimizing the first and second stage networks in step five. During the reasoning process, the network tends to generate highly overlapping directed candidate boxes. In the first stage, the present invention retains 2000 candidate boxes per layer of the feature pyramid and reduces redundancy by Non-maximum suppression (Non-Maximum Suppression, NMS). In the second phase, the threshold of IoU in the polygonal NMS (poly NMS) is set to 0.1.
Step seven: after the network training is completed, a group of multichannel images are input, and the position and the category of the local sedimentation can be detected and used as output.
Advances in InSAR technology can generate a large number of deformation measurement points, which makes it time-consuming and labor-consuming to evaluate geological disasters. According to the multichannel deep convolutional neural network model based on the multi-scale rotation detection frame, inSAR measurement and multi-source auxiliary data are integrated, separation of regional subsidence and local subsidence and automatic identification and classification of the local subsidence are achieved, and technical support is provided for large data technology construction intelligentization and modern geological disaster analysis and early warning.
The foregoing is merely a preferred embodiment of the present application, but the scope of the present application is not limited thereto, and any changes or substitutions easily contemplated by those skilled in the art within the technical scope of the present application should be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (10)
1. An automatic urban ground subsidence recognition and classification method is characterized by comprising the following steps:
constructing a two-layer hybrid network, and identifying high coherence points and low coherence points in the SAR image by utilizing a multi-point extraction technology based on the two-layer hybrid network;
performing Kriging interpolation on the high coherence point and the low coherence point to obtain an InSAR deformation speed diagram, and separating local sedimentation and a corresponding InSAR local deformation speed diagram from the InSAR deformation speed diagram;
constructing a local sedimentation sample set based on multi-source data based on the InSAR local deformation velocity diagram;
performing feature extraction on the local sedimentation sample set and the InSAR deformation speed based on a multi-channel pyramid structure to obtain a feature image, wherein the InSAR deformation speed is obtained based on the InSAR deformation speed map;
constructing an initial ground subsidence identification classification model, and training the initial ground subsidence identification classification model based on the local subsidence sample set to obtain a target ground subsidence identification classification model;
and inputting the characteristic images into the target ground subsidence recognition classification model to perform characteristic extraction so as to realize recognition and classification of ground subsidence.
2. The method for automatically identifying and classifying urban ground subsidence according to claim 1, wherein,
the process of obtaining the SAR image includes: registering the multi-baseline SLC images, and carrying out differential interferometry on the registered images to obtain SAR images.
3. The method for automatically identifying and classifying urban ground subsidence according to claim 1, wherein,
the process of identifying high coherence points and low coherence points using multi-point extraction techniques includes:
setting an initial amplitude dispersion threshold value in a first layer network of the two-layer hybrid network to obtain a high-quality high-coherence point; based on the high-quality high-coherence points, a Delaunay triangular network and a self-adaptive encryption network are constructed, and then phase difference processing is carried out on the adjacent high-quality high-coherence points, so that a differential interference diagram is obtained, and further, the height and deformation speed information of the high-quality coherence points is obtained;
in a second layer network of the two-layer hybrid network, a local star network is built in multiple directions by taking high-quality high-coherence points detected by a first layer network as reference points and taking four corner points of SAR images as starting points; and expanding the rest high coherence points and all low coherence points based on the local star network, and obtaining the height and deformation speed information of all the high coherence points and all the low coherence points in the vertical downward direction based on the direction of gradient movement.
4. The method for automatically identifying and classifying urban ground subsidence according to claim 3,
the process of obtaining the differential interferogram comprises: and carrying out parameter estimation on a sensing matrix of the height and deformation speed information to be estimated by adopting an M estimator and a ridge estimator, and combining the reconstructed backscattering coefficients to obtain the differential interference diagram.
5. The method for automatically identifying and classifying urban ground subsidence according to claim 3,
the process of identifying all low coherence points includes: identifying the same mass point based on a KS (K-nearest neighbor) test method in a second layer network of the two-layer hybrid network, estimating a covariance matrix by using a weight factor, and estimating phases based on a Phase-Linking method of a coherence coefficient to obtain all low coherence points;
the weight factor obtaining process comprises the following steps: introducing a first pixel point and a second pixel point in the SAR image, acquiring the statistical distance and the spatial distance of the two pixel points, and combining the statistical distance and the spatial distance by using a kernel function to acquire a weight factor; the statistical distance is the absolute difference of the largest empirical distribution function of the two pixel points, and the spatial distance is the Euclidean distance of the two pixel points.
6. The method for automatically identifying and classifying urban ground subsidence according to claim 1, wherein,
the process of separating out the local sedimentation includes: and converting the InSAR deformation speed map from a space domain to a frequency domain based on two-dimensional fast Fourier transform, and separating regional sedimentation and local sedimentation by utilizing band-pass filtering.
7. The method for automatically identifying and classifying urban ground subsidence according to claim 1, wherein,
the process of constructing a local sedimentation sample set based on multi-source data includes: and presetting a meeting condition of local sedimentation, screening a local sedimentation sample in the InSAR local deformation speed diagram, and attaching a class label to the local sedimentation sample based on the correlation of the spatial characteristics of sedimentation and auxiliary data.
8. The method for automatically identifying and classifying urban ground subsidence according to claim 1, wherein,
the process for constructing the initial ground subsidence recognition classification model comprises the following steps: introducing the central coordinates of the predicted candidate areas, the height and width of the circumscribed rectangle of the predicted directional candidate areas and the offset of the top and the middle point of the right side of the circumscribed rectangle to describe the directional area prediction network, so as to generate a high-quality rotation candidate frame; and correcting the position of the rotation candidate frame based on the rotation candidate frame and the characteristic image to obtain accurate positions and categories, and further completing the construction of an initial ground subsidence recognition classification model.
9. The method for automatically identifying and classifying urban ground subsidence according to claim 1, wherein,
the process of training the ground subsidence recognition classification model based on the local subsidence sample set includes: setting detection frames and marking frames on the InSAR deformation speed map, and when the cross joint overlapping of one detection frame and any marking frame exceeds a first preset threshold value, or the detection frame and one marking frame have the highest area overlapping, and the area overlapping is higher than a second preset threshold value, the detection frame is a sedimentation target; otherwise, the detection frame is a background;
the area threshold is the ratio of the area of the union of the detection frame and the labeling frame to the area of the intersection of the detection frame and the labeling frame.
10. The method for automatically identifying and classifying urban ground subsidence according to claim 1, wherein,
the loss function L of the target ground subsidence recognition classification model 1 The definition is as follows:
wherein N is the number of samples, F cls Is cross entropy loss and label of measurement label
And a network predicted tag p i Differences between F reg Is a detection frame delta for measuring network output by smoothing L1 loss i And labeling frame->
Offset between them. />
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310237618.9A CN116109931B (en) | 2023-03-02 | 2023-03-02 | Automatic urban ground subsidence recognition and classification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310237618.9A CN116109931B (en) | 2023-03-02 | 2023-03-02 | Automatic urban ground subsidence recognition and classification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116109931A true CN116109931A (en) | 2023-05-12 |
| CN116109931B CN116109931B (en) | 2024-03-15 |
Family
ID=86261772
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310237618.9A Active CN116109931B (en) | 2023-03-02 | 2023-03-02 | Automatic urban ground subsidence recognition and classification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116109931B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117607865A (en) * | 2023-10-30 | 2024-02-27 | 云南大学 | Subway line deformation detection method, device, system and storage medium |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104123464A (en) * | 2014-07-23 | 2014-10-29 | 中国国土资源航空物探遥感中心 | Method for inversion of ground feature high elevation and number of land subsidence through high resolution InSAR timing sequence analysis |
| US20160238704A1 (en) * | 2014-11-19 | 2016-08-18 | Conocophillips Company | Satellite geodesy and reservoir performance |
| CN108196252A (en) * | 2017-12-05 | 2018-06-22 | 中国资源卫星应用中心 | A kind of ore deposit Subsidence Area settles dynamic monitoring method |
| CN108663017A (en) * | 2018-08-13 | 2018-10-16 | 伟志股份公司 | A kind of method of ground settlement along monitoring city underground |
| CN110132237A (en) * | 2019-05-05 | 2019-08-16 | 四川省地质工程勘察院 | A kind of method of urban ground deformation disaster EARLY RECOGNITION |
| CN111798135A (en) * | 2020-07-06 | 2020-10-20 | 天津城建大学 | High-speed rail settlement hazard assessment method based on multi-source data integration |
| CN113238228A (en) * | 2021-05-10 | 2021-08-10 | 首都师范大学 | Level constraint-based InSAR three-dimensional surface deformation acquisition method, system and device |
| CN113405447A (en) * | 2020-05-19 | 2021-09-17 | 湖南北斗微芯产业发展有限公司 | Track traffic deformation monitoring method, device and equipment integrating InSAR and GNSS |
| CN113625241A (en) * | 2021-07-14 | 2021-11-09 | 北京市水文地质工程地质大队(北京市地质环境监测总站) | Differential settlement monitoring and early warning method |
| CN115494501A (en) * | 2022-09-29 | 2022-12-20 | 南宁市勘测设计院集团有限公司 | Urban infrastructure deformation monitoring method based on high-resolution PS-InSAR |
| US20220403721A1 (en) * | 2021-06-17 | 2022-12-22 | Halliburton Energy Services, Inc. | Systems and methods for automated gas lift monitoring |
-
2023
- 2023-03-02 CN CN202310237618.9A patent/CN116109931B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104123464A (en) * | 2014-07-23 | 2014-10-29 | 中国国土资源航空物探遥感中心 | Method for inversion of ground feature high elevation and number of land subsidence through high resolution InSAR timing sequence analysis |
| US20160238704A1 (en) * | 2014-11-19 | 2016-08-18 | Conocophillips Company | Satellite geodesy and reservoir performance |
| CN108196252A (en) * | 2017-12-05 | 2018-06-22 | 中国资源卫星应用中心 | A kind of ore deposit Subsidence Area settles dynamic monitoring method |
| CN108663017A (en) * | 2018-08-13 | 2018-10-16 | 伟志股份公司 | A kind of method of ground settlement along monitoring city underground |
| CN110132237A (en) * | 2019-05-05 | 2019-08-16 | 四川省地质工程勘察院 | A kind of method of urban ground deformation disaster EARLY RECOGNITION |
| CN113405447A (en) * | 2020-05-19 | 2021-09-17 | 湖南北斗微芯产业发展有限公司 | Track traffic deformation monitoring method, device and equipment integrating InSAR and GNSS |
| CN111798135A (en) * | 2020-07-06 | 2020-10-20 | 天津城建大学 | High-speed rail settlement hazard assessment method based on multi-source data integration |
| CN113238228A (en) * | 2021-05-10 | 2021-08-10 | 首都师范大学 | Level constraint-based InSAR three-dimensional surface deformation acquisition method, system and device |
| US20220403721A1 (en) * | 2021-06-17 | 2022-12-22 | Halliburton Energy Services, Inc. | Systems and methods for automated gas lift monitoring |
| CN113625241A (en) * | 2021-07-14 | 2021-11-09 | 北京市水文地质工程地质大队(北京市地质环境监测总站) | Differential settlement monitoring and early warning method |
| CN115494501A (en) * | 2022-09-29 | 2022-12-20 | 南宁市勘测设计院集团有限公司 | Urban infrastructure deformation monitoring method based on high-resolution PS-InSAR |
Non-Patent Citations (2)
| Title |
|---|
| 马培峰: ""coupling insar and numerical modeling for characterizing landslide movements under complex loads in urbanized hillslopes"", 《LANDSLIDES》 * |
| 马培峰: ""监测城市基础设施健康的星载MT-InSAR方法介绍"", 《测绘学报》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117607865A (en) * | 2023-10-30 | 2024-02-27 | 云南大学 | Subway line deformation detection method, device, system and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116109931B (en) | 2024-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Guzzetti et al. | Landslide inventory maps: New tools for an old problem | |
| Balaguer et al. | Definition of a comprehensive set of texture semivariogram features and their evaluation for object-oriented image classification | |
| CN101706577B (en) | Method for monitoring roadbed subsidence of express way by InSAR | |
| Mather et al. | An evaluation of Landsat TM spectral data and SAR-derived textural information for lithological discrimination in the Red Sea Hills, Sudan | |
| US11238307B1 (en) | System for performing change detection within a 3D geospatial model based upon semantic change detection using deep learning and related methods | |
| US11587249B2 (en) | Artificial intelligence (AI) system and methods for generating estimated height maps from electro-optic imagery | |
| US11747468B2 (en) | System using a priori terrain height data for interferometric synthetic aperture radar (IFSAR) phase disambiguation and related methods | |
| US11636649B2 (en) | Geospatial modeling system providing 3D geospatial model update based upon predictively registered image and related methods | |
| CN116109931B (en) | Automatic urban ground subsidence recognition and classification method | |
| Véga et al. | A sequential iterative dual-filter for Lidar terrain modeling optimized for complex forested environments | |
| Aplin | 2 Comparison of simulated IKONOS and SPOT HRV imagery for | |
| Xu et al. | Feature-based constraint deep CNN method for mapping rainfall-induced landslides in remote regions with mountainous terrain: An application to Brazil | |
| Ghasemi et al. | Urban classification using preserved information of high dimensional textural features of Sentinel-1 images in Tabriz, Iran | |
| Tarsha-Kurdi et al. | New approach for automatic detection of buildings in airborne laser scanner data using first echo only | |
| WO2022271963A1 (en) | Automated computer system and method of road network extraction from remote sensing images using vehicle motion detection to seed spectral classification | |
| Mhangara et al. | Road extraction using object oriented classification | |
| Chen et al. | Susceptibility-guided landslide detection using fully convolutional neural network | |
| Pfeifer et al. | LiDAR data filtering and digital terrain model generation | |
| Cai et al. | Automatic identification of active landslides over wide areas from time-series InSAR measurements using Faster RCNN | |
| Warner et al. | Pine Island Glacier (Antarctica) velocities from Landsat7 images between 2001 and 2011: FFT-based image correlation for images with data gaps | |
| Lian et al. | Extraction of high-accuracy control points using ICESat-2 ATL03 in urban areas | |
| Saba et al. | Comparison of pixel, sub-pixel and object-based image analysis techniques for co-seismic landslides detection in seismically active area in Lesser Himalaya, Pakistan | |
| Kim et al. | Tree and building detection in dense urban environments using automated processing of IKONOS image and LiDAR data | |
| Yong et al. | Site characterization using integrated imaging analysis methods on satellite data of the Islamabad, Pakistan, region | |
| CN113238228A (en) | Level constraint-based InSAR three-dimensional surface deformation acquisition method, system and device |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |