阅读说明：本技术 一种滑坡形变监测方法与可视化服务平台 (Landslide deformation monitoring method and visual service platform ) 是由 张鹏 孙恺悦 王煜 孙学宏 于 2021-09-03 设计创作，主要内容包括：本发明涉及遥感技术的技术领域,特别是涉及一种滑坡形变监测方法与可视化服务平台,从滑坡区域特征出发,利用光学与雷达影像融合带来的优势,提出一种基于光学、雷达数据融合的滑坡监测系统方法,解决传统滑坡形变监测方法中精度低、工作量大、成本高的问题；包括：滑坡识别模块、形变处理模块、风险评估预警模块、数据管理模块和可视化模块；滑坡识别模块,用于遥感影像中滑坡区域的识别；形变处理模块,用于提取滑坡区域内形变特征；风险评估预警模块,用于对各个滑坡发生灾害的风险进行评估和预警；数据管理模块,用于存储系统内各项数据；可视化分析模块,用于滑坡区域形变数据分析及可视化展示。(The invention relates to the technical field of remote sensing technology, in particular to a landslide deformation monitoring method and a visual service platform, and provides a landslide monitoring system method based on optical and radar data fusion by starting from landslide region characteristics and utilizing advantages brought by optical and radar image fusion, so that the problems of low precision, large workload and high cost in the traditional landslide deformation monitoring method are solved; the method comprises the following steps: the system comprises a landslide identification module, a deformation processing module, a risk assessment early warning module, a data management module and a visualization module; the landslide identification module is used for identifying a landslide area in the remote sensing image; the deformation processing module is used for extracting deformation characteristics in a landslide area; the risk assessment early warning module is used for assessing and early warning the risk of disaster occurrence of each landslide; the data management module is used for storing various data in the system; and the visual analysis module is used for analyzing and visually displaying the deformation data of the landslide area.)

1. The utility model provides a landslide deformation monitoring visual service platform which characterized in that includes: the system comprises a landslide identification module, a deformation processing module, a risk assessment early warning module, a data management module and a visualization module;

the landslide identification module is used for identifying a landslide area in the remote sensing image;

the deformation processing module is used for extracting deformation characteristics in a landslide area;

the risk assessment early warning module is used for assessing and early warning the risk of disaster occurrence of each landslide;

the data management module is used for storing various data in the system;

and the visual analysis module is used for analyzing and visually displaying the deformation data of the landslide area.

2. The landslide deformation monitoring visual service platform of claim 1, wherein the landslide identification module: selecting remote sensing image data and an identification area, preprocessing the remote sensing image by the module, extracting an area and a boundary of a landslide target in the image by using a landslide identification model based on a convolutional neural network, and finally storing a landslide area identification polygon result in a data management module.

3. The landslide deformation monitoring visualization service platform of claim 2, wherein the deformation processing module: in a deformation processing module, performing D-InSAR processing on a target area on the basis of a time sequence SAR image to obtain a short-term deformation diagram of the whole target area; and then matching the landslide identification area with the short-term deformation graph to obtain the short-term deformation of the landslide in the target area, and finally storing landslide deformation information and a deformation rating result in a data management module.

4. The landslide deformation monitoring visualization service platform of claim 3, wherein the risk assessment early warning module: dividing the landslide into three types of high risk, low risk and no risk according to a preset deformation threshold; aiming at high-risk landslide, warning is timely sent out through an early warning module, meanwhile, follow-up monitoring frequency is set to be high frequency, the landslide area is mainly monitored for a long time, for low-risk landslide, besides the early warning, the follow-up monitoring frequency is set to be low frequency, optimization of data operation resources and monitoring efficiency is achieved, for no-risk landslide, the follow-up monitoring frequency is ignored until overall deformation investigation of the next period is carried out, for different risk landslide in deformation risk assessment, when a follow-up monitoring scheme is timely adjusted, risk information can be sent to a local responsible person in a short message mode.

5. The landslide deformation monitoring visualization service platform of claim 4 wherein the data management module: the data not only comprises optical and radar remote sensing data required by landslide deformation monitoring, but also comprises landslide short-term and long-term deformation analysis results, disaster data, vector data and chart documents, and is based on a big data storage technology.

6. The landslide deformation monitoring visualization service platform of claim 5 wherein the visualization analysis module: a data analysis and visualization platform developed based on a WEB technology is compiled by using Javascript language, different satellite remote sensing image maps are manufactured based on tile data of remote sensing images, landslide related data are read, and landslide area marks, deformation analysis, risk levels and landslide basic information in the maps are displayed.

7. A landslide deformation monitoring method is characterized by comprising the following steps:

step 1, constructing a loess landslide identification network model based on a multi-source remote sensing image; the multi-source remote sensing image loess landslide identification network model mainly comprises a main network and a cavity space convolution pooling pyramid module; the main network part uses the current popular residual error network or various variant networks of the residual error network, firstly, the input network is a characteristic diagram with the size of C multiplied by H multiplied by W, C represents the channel number of the characteristic diagram, H is the length of the characteristic diagram, W is the width of the characteristic diagram, in order to fuse the spectral characteristic and the terrain characteristic of the landslide, RGB data and DEM data corresponding to a landslide sample are merged into 4-channel data to be input into the main network to obtain two outputs of a shallow characteristic and a deep characteristic, the deep characteristic is used as the input of an ASPP module, the ASPP structure uses four expansion convolutions with different ratios and a global average pooling to respectively extract a local characteristic and a global characteristic, the five output characteristics with different scales are superposed and then are convolved by 1x1 to obtain a one-dimensional characteristic diagram fusing multi-scale characteristic information, a Decoder module fuses the shallow characteristic and the multi-scale characteristic, then, obtaining a landslide mass recognition prediction map with the same size as the input data through upsampling;

step 2, establishing a historical loess landslide sample database based on the high-resolution second satellite image artificial visual interpretation result, training a network model on the sample database, predicting a new remote sensing image by using the trained network model, and identifying a landslide area on the remote sensing image, wherein the specific implementation comprises the following substeps:

step 2.1, adopting a high-resolution remote sensing image provided by a high-resolution second satellite covering a research area as a data source, and finally obtaining a complete image in an experimental area by the high-resolution second remote sensing image through operations of radiometric calibration, orthometric correction, geometric registration, image fusion, image cutting and image splicing pretreatment;

step 2.2, the landslide is judged by using the terrain visibility index in an auxiliary mode, the landslide visibility and a remote sensing visual interpretation method of the landslide are combined, a landslide sample is established, the polygonal vector is used for marking along the edge of the landslide sample, the polygonal edge vector edge of each landslide example is ensured to be more accurate and reliable, and no omitted and unmarked landslide sample exists in the image;

step 2.3, cutting landslide image samples and making labels, analyzing vector files marked on the landslide samples, enabling each group of polygon coordinates to correspond to a landslide sample example, amplifying external rectangles of each landslide sample marked in an experimental area according to a proportion of 1.2-2 times at random to serve as cutting sizes, enabling each cut landslide sample to contain a certain amount of background areas except landslides, and finally generating a gray graph with the same size as the cutting size for each landslide example according to the polygon coordinates, wherein one image only contains two categories: processing image labels into 0 and 1, wherein the pixel value in the range of a landslide example is 1, the non-landslide is 0, and respectively storing optical image data and a gray scale map corresponding to the processed landslide sample into image and label folders by the same name to serve as a landslide sample database;

step 2.4, cutting corresponding elevation data, simultaneously cutting digital elevation model images according to the same coordinate and range, respectively using the digital elevation model images as auxiliary data in the training and testing process, and inputting the auxiliary data into a network together for training and testing;

step 3, training the landslide area identification model by using the labeled landslide sample data set after the construction of the network is completed; aiming at the problems that a self-built landslide data set has few learning samples and is easy to generate overfitting in a complex network, a data enhancement technology is adopted to expand an original data set, when an image sample is processed, landslide sample data is amplified through random cutting, rotating, stretching, horizontal turning and offset operations on the original image to obtain partial image amplification operation front and back results, the same operation is carried out on data corresponding to the landslide sample to ensure the mutual matching of various types of data, and finally the augmented landslide data set is according to 9: 1, dividing the proportion into a training set and a verification set, and performing iterative training on the convolutional neural network by using a training set sample to obtain an optimal landslide area identification model;

step 4, carrying out regional deformation treatment, selecting a target region, carrying out remote sensing data treatment by using a D-InSAR technology aiming at a research region to obtain a short-term deformation result of the target region, and subsequently carrying out remote sensing data treatment by using an SBAS-InSAR technology aiming at high-risk and low-risk landslide points to obtain a long-term time sequence deformation treatment result;

step 5, risk assessment and early warning are carried out, a deformation threshold value is set, a landslide deformation result is compared with the deformation threshold value, a landslide exceeding the deformation threshold value is subjected to deformation risk analysis, the landslide is divided into three levels, namely a high risk level, a low risk level and a risk-free level, band operation and classification operation are carried out on the basis of a processed remote sensing image, and relevant landslide hazard factors such as earth surface coverage are extracted; performing GIS space analysis by using DEM data, and extracting related terrain and hydrological factors such as gradient, slope direction, and valley distance;

and 6, evaluating a deformation result, namely comparing the deformation result of the target landslide experiment with the field inspection result to obtain a difference quantitative analysis report of the deformation result and the field inspection result, and realizing the accuracy evaluation of the deformation processing flow.

Technical Field

The invention relates to the technical field of remote sensing technology, in particular to a landslide deformation monitoring method and a visual service platform.

Background

Landslide is a common natural disaster, which seriously threatens the natural environment and the safety of people's lives and properties, and landslide disaster deformation monitoring provides important data support for the prevention and treatment work of landslide disasters. In the early days, landslide hazard data was obtained primarily by field investigation by field workers, a method which is time consuming, labor intensive and dangerous. The remote sensing technology is an emerging measurement technology in the 20 th century, has good observation timeliness, strong macroscopic property and rich information quantity, is widely used in various fields, and has higher and higher observation precision along with the continuous progress of the technology, thereby playing a more important role in landslide monitoring.

The remote sensing monitoring method can rapidly acquire landslide information in a large range, has a wider monitoring range and higher monitoring efficiency compared with conventional monitoring means, is lower in economic consumption, and is one of indispensable monitoring means for geological disasters such as landslides. The SynthetIc Aperture Radar (SAR) has a great potential in monitoring geological disasters, such as a common SynthetIc radar interference system (interferometric SAR), a Differential interferometric SAR (dinsar), and the like, radar waves emitted by the technologies can penetrate through cloud fog, the SynthetIc aperture radar has all-weather working capability, and ground information can be more accurately acquired through an interferogram generated by the SynthetIc aperture radar.

The identification of landslide geological disasters is the important factor in landslide deformation monitoring and analysis. With the deep research on landslide properties, some automatic landslide area identification methods are gradually proposed, and mainly can be divided into two types, namely a landslide identification method based on optical remote sensing images and a landslide identification method based on microwave remote sensing technology. The method based on the optical image mainly identifies the landslide according to the special texture, hue, geometric form and other typical characteristics of the landslide in the image, and the method based on the microwave remote sensing mainly finds the landslide by identifying the micro deformation of the earth surface. The traditional landslide identification method based on the optical remote sensing image relies on various manually designed features to identify and judge landslides. However, because the difference of optical characteristics between different landslides on the remote sensing image is large, the hue, the geometric shape, the texture, the shadow and the figure of each landslide example are different. Therefore, the landslide is identified by utilizing a method for artificially designing features, the identification accuracy is not high, the generalization performance of the algorithm is poor, and the landslide identification effect is not ideal. In recent years, convolutional neural networks in deep learning exhibit strong performance in image processing-related tasks such as image classification, target detection, semantic segmentation, and the like. Therefore, there are also a few researchers using convolutional neural networks for landslide identification. However, these studies are still in the infancy and only a few simple network models have been designed and tried. The use of a more complex network structure and the adoption of a more targeted design is an effective method for improving the landslide identification accuracy. For the landslide identification task, how to make the model pay attention to and learn the essential characteristics of the landslide better is a key problem for designing the automatic landslide identification model.

The related scheme of the prior art is as follows: GPS monitoring

The GPS is a short name of a Global Positioning System (Global Positioning System), and performs space backward intersection measurement by using signals transmitted by satellites so as to determine three-dimensional coordinates of a ground point to be measured, and reflects the displacement of the point to be measured by using the difference between a new value and an initial coordinate on the basis of distance, coordinates and angles to achieve the purpose of monitoring deformation. The method can be used for monitoring the surface displacement change of the landslide at different deformation stages, can accurately measure the change rate, can also be used for continuous monitoring, is widely used in landslide surface deformation monitoring, and can achieve millimeter level continuous monitoring precision. Each GPS antenna in the GPS monitoring system can receive original data at the same time, then the information is transmitted to a landslide deformation monitoring control center (figure 1) through a wireless network, then the calculation software of the control center calculates the information flow of each monitoring point in real time by adopting methods such as static differential displacement measurement and the like, the three-dimensional coordinates of each monitoring point are obtained, the landslide displacement condition is analyzed, and warning can be sent out if necessary.

The related scheme of the prior art is as follows: surface fracture monitoring

The geometrical form, the combination relation and the mechanical property of the landslide cracks can judge the distribution range, the thickness of a landslide body, the mechanical mechanism and the development stage of the landslide, so that the property and the development trend of the landslide are objectively evaluated and predicted, and an important basis is provided for monitoring and early warning of landslide disasters and investigation and treatment of the landslide. The crack monitoring is suitable for landmark crack opening and closing and dislocation change monitoring in various deformation stages of the landslide. In the landslide surface crack monitoring during landslide surface deformation monitoring, monitoring contents comprise the pulling speed of a crack and the expansion conditions of two ends, and common measuring instruments comprise a telescopic instrument, a dislocation instrument, a dividing ruler and the like; in the monitoring of the cracks on the landslide surface, the monitoring contents comprise the pulling speed of the cracks and the expansion conditions of two ends, and common measuring instruments comprise a stretching instrument, a dislocation instrument, a micrometer and the like. The monitoring method is simple to operate, feasible in precision, adjustable in measuring range and convenient and rapid in data acquisition, and the manual measurement reading is high in working strength and insufficient in long-term stability.

The prior art adopts a third related scheme: InSAR monitoring method

InSAR is a short name of Synthetic Aperture Radar interferometry (interferometric Synthetic Aperture Radar) technology, and the principle is that two SAR images are obtained by imaging the same area twice on adjacent repeated satellite orbits through a Radar satellite, the phase difference of the two SAR images is calculated to obtain an image after interference processing, and then three-dimensional information and elevation change information of the earth surface can be extracted from interference fringes through phase unwrapping and differential interference processing. The InSAR monitoring method adopts radar and satellite monitoring, so that the method has the advantages of all weather, high precision, high resolution, large coverage, long observation span, convenience, rapidness and the like, is sensitive to the geometric shape and the surface relief of ground objects, can penetrate through soil and vegetation, can obtain macroscopic static information, and can also obtain continuous surface deformation dynamic information of a certain area. Compared with the GPS, the InSAR monitoring result is more continuous, has stronger space coverage advantage and spatial resolution, has good performance in landslide monitoring at home and abroad, and also carries out landslide monitoring by using the technical complementation of the GPS and the InSAR. However, in the data acquisition process, many factors bring errors, which affect the quality of the interferogram. If the sudden landslide time is short, the interference images are irrelevant, and a decorrelation problem is generated; in high mountain areas, radar shadowing and radar beam eclipsing may occur. In addition, interferograms with good correlation have strong dependence on seasons and weather conditions, and high humidity and high vegetation coverage can adversely affect the correlation.

Disclosure of Invention

The invention aims to provide a landslide monitoring system method based on optical and radar data fusion based on the advantages brought by optical and radar image fusion from landslide area characteristics, and solves the problems of low precision, large workload and high cost in the traditional landslide deformation monitoring method.

The invention also aims to establish a visual service platform by applying the method, so as to realize landslide disaster monitoring, quick response and visual analysis.

The invention relates to a landslide deformation monitoring visual service platform, which comprises: the system comprises a landslide identification module, a deformation processing module, a risk assessment early warning module, a data management module and a visualization module;

the landslide identification module is used for identifying a landslide area in the remote sensing image;

the deformation processing module is used for extracting deformation characteristics in a landslide area;

the risk assessment early warning module is used for assessing and early warning the risk of disaster occurrence of each landslide;

the data management module is used for storing various data in the system;

and the visual analysis module is used for analyzing and visually displaying the deformation data of the landslide area.

Further, the landslide identification module: selecting remote sensing image data and an identification area, preprocessing the remote sensing image by the module, extracting an area and a boundary of a landslide target in the image by using a landslide identification model based on a convolutional neural network, and finally storing a landslide area identification polygon result in a data management module.

Further, the deformation processing module: in a deformation processing module, performing D-InSAR processing on a target area on the basis of a time sequence SAR image to obtain a short-term deformation diagram of the whole target area; then, matching the landslide identification area with the short-term deformation graph to obtain the short-term deformation amount of the landslide in the target area, and finally storing landslide deformation information and a deformation rating result in a data management module as shown in fig. 3;

further, the risk assessment early warning module: dividing the landslide into three types of high risk, low risk and no risk according to a preset deformation threshold value, such as that the horizontal displacement is not more than 12 mm/month, the vertical displacement is not more than 5 mm/month and the like; aiming at high-risk landslide, warning is timely sent out through an early warning module, meanwhile, follow-up monitoring frequency is set to be high frequency, the landslide area is mainly monitored for a long time, for low-risk landslide, besides the early warning, the follow-up monitoring frequency is set to be low frequency, optimization of data operation resources and monitoring efficiency is achieved, for no-risk landslide, the follow-up monitoring frequency is ignored until overall deformation investigation of the next period is carried out, for different risk landslide in deformation risk assessment, when a follow-up monitoring scheme is timely adjusted, risk information can be sent to a local responsible person in a short message mode.

Further, the data management module: the data not only comprises optical and radar remote sensing data required by landslide deformation monitoring, but also comprises landslide short-term and long-term deformation analysis results, disaster data, vector data and chart documents, and is based on a big data storage technology.

Further, the visualization analysis module: a data analysis and visualization platform developed based on a WEB technology is compiled by using Javascript language, different satellite remote sensing image maps are manufactured based on tile data of remote sensing images, landslide related data are read, and landslide area marks, deformation analysis, risk levels and landslide basic information in the maps are displayed.

The invention discloses a landslide deformation monitoring method, which comprises the following steps:

step 1, constructing a loess landslide identification network model based on a multi-source remote sensing image, wherein the network is used for learning the characteristics of various landslides on the remote sensing image; the multi-source remote sensing image loess landslide identification network model mainly comprises a backbone network (backbone network) and a cavity space convolution pooling pyramid module (solar spatial pyramid) and comprises a main network (backbone network) and a cavity space convolution pooling pyramid module (solar spatial pyramid module); the backbone Network part uses the currently popular Residual Network (ResNet) or various variant networks of the Residual Network, and the like, the structure is shown in FIG. 5, firstly, the input Network is a characteristic diagram with the size of C × H × W, C represents the channel number of the characteristic diagram, H is the length of the characteristic diagram, W is the width of the characteristic diagram, in order to fuse the spectral characteristic and the topographic characteristic of the landslide, RGB data and DEM data corresponding to the landslide sample are merged into 4-channel data to be input into the backbone Network to obtain two outputs of a shallow characteristic and a deep characteristic, the deep characteristic is used as the input of an ASPP module, the ASPP structure uses four expansion convolutions with different ratios and a global average pooling to respectively extract the local characteristic and the global characteristic, the five output characteristics with different scales are superposed and then are subjected to dimension adjustment through 1x1 convolution to obtain a one-dimensional characteristic diagram fusing multi-scale characteristic information, the Decoder module fuses shallow layer features and multi-scale features and then obtains a landslide mass recognition prediction map with the same size as input data through up-sampling;

step 2, establishing a historical loess landslide sample database based on the high-resolution second satellite image artificial visual interpretation result, training a network model on the sample database, predicting a new remote sensing image by using the trained network model, and identifying a landslide area on the remote sensing image, wherein the specific implementation comprises the following sub-steps as shown in fig. 1:

step 2.1, adopting a high-resolution remote sensing image provided by a high-resolution second satellite covering a research area as a data source, and finally obtaining a complete image in an experimental area by the high-resolution second remote sensing image through operations of radiometric calibration, orthometric correction, geometric registration, image fusion, image cutting and image splicing pretreatment (figure 2);

step 2.2, the landslide is judged by using the terrain visibility index in an auxiliary mode, the landslide visibility and a remote sensing visual interpretation method of the landslide are combined, a landslide sample is established, the polygonal vector is used for marking along the edge of the landslide sample, the polygonal edge vector edge of each landslide example is ensured to be more accurate and reliable, and no omitted and unmarked landslide sample exists in the image;

step 2.3, cutting landslide image samples and making labels, analyzing vector files marked by the landslide samples, enabling each group of polygon coordinates to correspond to a landslide sample example, amplifying external rectangles of each landslide sample marked in an experimental area according to a proportion of 1.2-2 times at random to serve as cutting sizes, enabling each cut landslide sample to contain a certain amount of background areas except landslides, and finally generating a gray scale map (Mask) with the same size as the cutting size for each landslide example according to the polygon coordinates, wherein one image only comprises two categories: processing image labels into 0 and 1, wherein the pixel value in the range of a landslide example is 1, the non-landslide is 0, and respectively storing optical image data and a gray scale map corresponding to the processed landslide sample into image and label folders by the same name to serve as a landslide sample database;

step 2.4, cutting corresponding elevation data, simultaneously cutting digital elevation model images according to the same coordinate and range, respectively using the digital elevation model images as auxiliary data in the training and testing process, and inputting the auxiliary data into a network together for training and testing;

step 3, training the landslide area identification model by using the labeled landslide sample data set after the construction of the network is completed; aiming at the problems that a self-built landslide data set has few learning samples and is easy to generate overfitting in a complex network, a data enhancement technology is adopted to expand an original data set, when an image sample is processed, landslide sample data is amplified through random cutting, rotating, stretching, horizontal overturning and offset operations on the original image, results before and after partial image amplification operations are obtained, the results before and after the partial image amplification operations are shown in figure 3, the same operation is carried out on data corresponding to the landslide sample to ensure the mutual matching of various types of data, and finally the augmented landslide data set is subjected to the following steps of: 1, dividing the proportion into a training set and a verification set, and performing iterative training on the convolutional neural network by using a training set sample to obtain an optimal landslide area identification model;

step 4, carrying out regional deformation treatment, selecting a target region, carrying out remote sensing data treatment by using a D-InSAR technology aiming at a research region to obtain a short-term deformation result of the target region, and subsequently carrying out remote sensing data treatment by using an SBAS-InSAR technology aiming at high-risk and low-risk landslide points to obtain a long-term time sequence deformation treatment result;

step 5, risk assessment and early warning are carried out, deformation threshold values (such as deformation amounts in the horizontal and vertical directions in unit time) are set, landslide deformation results are compared with the deformation threshold values, landslide exceeding the deformation threshold values is subjected to deformation risk analysis, the landslide is divided into three levels, namely a high risk level, a low risk level and a risk-free level, waveband operation and classification operation are carried out on the basis of processed remote sensing images, and relevant landslide hazard factors such as ground surface coverage are extracted; performing GIS space analysis by using DEM data, and extracting related terrain and hydrological factors such as gradient, slope direction, and valley distance;

and 6, evaluating a deformation result, namely comparing the deformation result of the target landslide experiment with the field inspection result to obtain a difference quantitative analysis report of the deformation result and the field inspection result, and realizing the accuracy evaluation of the deformation processing flow.

Compared with the prior art, the invention has the beneficial effects that:

(1) the method utilizes deep learning and InSAR technology to realize automatic extraction of deformation of the target landslide region;

(2) the method realizes landslide automatic risk rating by utilizing cloud computing;

(3) the invention solves the problems of complex environment, difficult landslide inspection and the like of a target area;

(4) the method is suitable for deformation monitoring research of other infrastructures.

Drawings

FIG. 1 is a GPS monitoring system;

FIG. 2 is a block diagram of a landslide deformation monitoring visualization service platform of the present invention;

FIG. 3 is a left region deformation diagram; the right is a time sequence deformation line graph;

FIG. 4 is a schematic of a risk assessment pre-warning;

fig. 5 is a schematic structural diagram of a landslide identification model.

Detailed Description

The following detailed description of embodiments of the present invention is provided in connection with the accompanying drawings and examples. The following examples are intended to illustrate the invention but are not intended to limit the scope of the invention.

From the perspective of data analysis, visualization and rapid and high-quality service capability, a set of efficient and easily-extensible visualization platform is established, landslide disaster monitoring, rapid response and visualization analysis are realized in urgent need, the visualization platform and the field have low coupling degree, and the requirements on the software expansibility and the universality of the visualization platform can be met through a hierarchical and componentized framework.

Based on the technical scheme, the rapid response and visualization service platform based on the multi-source satellite remote sensing data is provided, the platform is established based on a container and a micro-service framework, various satellite remote sensing data analysis and visualization technologies are integrated, and near real-time disaster monitoring service is achieved.

As shown in fig. 2, the landslide deformation monitoring visualization service platform of the present invention includes: the system comprises a landslide identification module, a deformation processing module, a risk assessment early warning module, a data management module and a visualization module;

the landslide identification module is used for identifying a landslide area in the remote sensing image;

the deformation processing module is used for extracting deformation characteristics in a landslide area;

the risk assessment early warning module is used for assessing and early warning the risk of disaster occurrence of each landslide;

the data management module is used for storing various data in the system;

and the visual analysis module is used for analyzing and visually displaying the deformation data of the landslide area.

Further, the landslide identification module: selecting remote sensing image data and an identification area, preprocessing the remote sensing image by the module, extracting an area and a boundary of a landslide target in the image by using a landslide identification model based on a convolutional neural network, and finally storing a landslide area identification polygon result in a data management module.

Further, the deformation processing module: in a deformation processing module, performing D-InSAR processing on a target area on the basis of a time sequence SAR image to obtain a short-term deformation diagram of the whole target area; then, matching the landslide identification area with the short-term deformation graph to obtain the short-term deformation amount of the landslide in the target area, and finally storing landslide deformation information and a deformation rating result in a data management module as shown in fig. 3;

further, the risk assessment early warning module: dividing the landslide into three types of high risk, low risk and no risk according to a preset deformation threshold value, such as that the horizontal displacement is not more than 12 mm/month, the vertical displacement is not more than 5 mm/month and the like; aiming at high-risk landslide, warning is timely sent out through an early warning module, meanwhile, follow-up monitoring frequency is set to be high frequency, the landslide area is mainly monitored for a long time, for low-risk landslide, besides the early warning, the follow-up monitoring frequency is set to be low frequency, optimization of data operation resources and monitoring efficiency is achieved, for no-risk landslide, the follow-up monitoring frequency is ignored until overall deformation investigation of the next period is carried out, for different risk landslide in deformation risk assessment, when a follow-up monitoring scheme is timely adjusted, risk information can be sent to a local responsible person in a short message mode.

Further, the data management module: the data not only comprises optical and radar remote sensing data required by landslide deformation monitoring, but also comprises landslide short-term and long-term deformation analysis results, disaster data, vector data and chart documents, and is based on a big data storage technology.

Further, the visualization analysis module: the method comprises the steps that a data analysis and visualization platform developed based on a WEB technology is compiled by using Javascript language, different satellite remote sensing image maps are manufactured on the basis of tile data of remote sensing images, landslide related data are read, and display of landslide area marks, deformation analysis, risk levels and landslide basic information in the maps is achieved;

the system bottom layer relies on virtualized basic resource facilities, Kubernets are used as a main operation environment, a dynamically telescopic micro-service architecture is realized, support for a cloud virtual environment and a GPU server is realized, the technical standard of the remote sensing big data service available for production is determined based on landslide disaster monitoring service requirements, the micro-service management technology is tightly combined, complete decoupling with the bottom layer basic facilities is realized, and the system has modern cloud native application characteristics of automatic task distribution, automatic data processing, automatic level expansion, load balancing, central configuration management and the like.

The invention discloses a landslide deformation monitoring method, which comprises the following steps:

step 1, constructing a loess landslide identification network model based on a multi-source remote sensing image, wherein the network is used for learning the characteristics of various landslides on the remote sensing image; the multi-source remote sensing image loess landslide identification network model mainly comprises a backbone network (backbone network) and a cavity space convolution pooling pyramid module (solar spatial pyramid) and comprises a main network (backbone network) and a cavity space convolution pooling pyramid module (solar spatial pyramid module); the backbone Network part uses the currently popular Residual Network (ResNet) or various variant networks of the Residual Network, and the like, the structure is shown in FIG. 5, firstly, the input Network is a characteristic diagram with the size of C × H × W, C represents the channel number of the characteristic diagram, H is the length of the characteristic diagram, W is the width of the characteristic diagram, in order to fuse the spectral characteristic and the topographic characteristic of the landslide, RGB data and DEM data corresponding to the landslide sample are merged into 4-channel data to be input into the backbone Network to obtain two outputs of a shallow characteristic and a deep characteristic, the deep characteristic is used as the input of an ASPP module, the ASPP structure uses four expansion convolutions with different ratios and a global average pooling to respectively extract the local characteristic and the global characteristic, the five output characteristics with different scales are superposed and then are subjected to dimension adjustment through 1x1 convolution to obtain a one-dimensional characteristic diagram fusing multi-scale characteristic information, the Decoder module fuses shallow layer features and multi-scale features and then obtains a landslide mass recognition prediction map with the same size as input data through up-sampling;

step 2, establishing a historical loess landslide sample database based on the high-resolution second satellite image artificial visual interpretation result, training a network model on the sample database, predicting a new remote sensing image by using the trained network model, and identifying a landslide area on the remote sensing image, wherein the specific implementation comprises the following sub-steps as shown in fig. 1:

step 2.1, adopting a high-resolution remote sensing image provided by a high-resolution second satellite covering a research area as a data source, and finally obtaining a complete image in an experimental area by the high-resolution second remote sensing image through operations of radiometric calibration, orthometric correction, geometric registration, image fusion, image cutting and image splicing pretreatment (figure 2);

step 2.2, the landslide is judged by using the terrain visibility index in an auxiliary mode, the landslide visibility and a remote sensing visual interpretation method of the landslide are combined, a landslide sample is established, the polygonal vector is used for marking along the edge of the landslide sample, the polygonal edge vector edge of each landslide example is ensured to be more accurate and reliable, and no omitted and unmarked landslide sample exists in the image;

step 2.3, cutting landslide image samples and making labels, analyzing vector files marked by the landslide samples, enabling each group of polygon coordinates to correspond to a landslide sample example, amplifying external rectangles of each landslide sample marked in an experimental area according to a proportion of 1.2-2 times at random to serve as cutting sizes, enabling each cut landslide sample to contain a certain amount of background areas except landslides, and finally generating a gray scale map (Mask) with the same size as the cutting size for each landslide example according to the polygon coordinates, wherein one image only comprises two categories: processing image labels into 0 and 1, wherein the pixel value in the range of a landslide example is 1, the non-landslide is 0, and respectively storing optical image data and a gray scale map corresponding to the processed landslide sample into image and label folders by the same name to serve as a landslide sample database;

step 2.4, cutting corresponding elevation data, simultaneously cutting digital elevation model images according to the same coordinate and range, respectively using the digital elevation model images as auxiliary data in the training and testing process, and inputting the auxiliary data into a network together for training and testing;

step 3, training the landslide area identification model by using the labeled landslide sample data set after the construction of the network is completed; aiming at the problems that a self-built landslide data set has few learning samples and is easy to generate overfitting in a complex network, a data enhancement technology is adopted to expand an original data set, when an image sample is processed, landslide sample data is amplified through random cutting, rotating, stretching, horizontal overturning and offset operations on the original image, results before and after partial image amplification operations are obtained, the results before and after the partial image amplification operations are shown in figure 3, the same operation is carried out on data corresponding to the landslide sample to ensure the mutual matching of various types of data, and finally the augmented landslide data set is subjected to the following steps of: 1, dividing the proportion into a training set and a verification set, and performing iterative training on the convolutional neural network by using a training set sample to obtain an optimal landslide area identification model;

step 4, carrying out regional deformation treatment, selecting a target region, carrying out remote sensing data treatment by using a D-InSAR technology aiming at a research region to obtain a short-term deformation result of the target region, and subsequently carrying out remote sensing data treatment by using an SBAS-InSAR technology aiming at high-risk and low-risk landslide points to obtain a long-term time sequence deformation treatment result;

step 5, risk assessment and early warning are carried out, deformation threshold values (such as deformation amounts in the horizontal and vertical directions in unit time) are set, landslide deformation results are compared with the deformation threshold values, landslide exceeding the deformation threshold values is subjected to deformation risk analysis, the landslide is divided into three levels, namely a high risk level, a low risk level and a risk-free level, waveband operation and classification operation are carried out on the basis of processed remote sensing images, and relevant landslide hazard factors such as ground surface coverage are extracted; performing GIS space analysis by using DEM data, and extracting related terrain and hydrological factors such as gradient, slope direction, and valley distance;

and 6, evaluating a deformation result, namely comparing the deformation result of the target landslide experiment with the field inspection result to obtain a difference quantitative analysis report of the deformation result and the field inspection result, and realizing the accuracy evaluation of the deformation processing flow.

The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.