Landslide Susceptibility Based On Time-series InSAR Technology Combined With Back Propagation

China is one of the countries with the most frequent geological disasters,of which landslides disasters account for more than 70%,in the world.Landslides have caused enormous casualties and economic losses.In landslide research and disaster monitoring,the deformation is a reflection of the current state of motion stability and is the most direct physical quantity reflecting the current stability and motion state of the landslide body.At present.there are many methods for deformation monitoring,among which,Synthetic Aperture Radar(SAR)has short time interval,high accuracy and large monitoring range all-day and all-weather and many other advantages compared with optical ramote sensing and traditional measurement technology.Especially in the southwestern part of C hina,where the terrain is complex and varied and the vegetation is dense,the long-band radar signal can effectively penetrate the cloud and vegetation canopy to achieve large-scale high-density monitoring.Interferometric Synthetic Aperture Radar(InSAR)technology is one of the most important technical means to extract topographic information from radar remote sensing.However.conventional Differential Interferometric Synthetic Aperture Radar(DInSAR)technology is susceptible to temporal and spatial decorrelation,atmospheric delay and terrain error,and the monitoring accuracy of long-term sequence deformation monitoring is low.To cope with these drawbacks,various time series methods have been developed,such as Permanent Scattering InSAR(PSI)and Small Baseline Subset(SBAS).However.due to the limitations of the technology itself,it's not satisfactory for its application in large gradient deformation regions such as landslides and debris flows.Therefore,this paper combines time series InSAR technology and Offset-tracking technology to improve the monitoring ability of SAR technology in landslide deformation monitoring.Based on the obtained deformations,combined with the topographic information,soil type.land use,geological type.vegetation coverage and rainfall information of the study area,the deep learning method was used to evaluate the sensitivity of the landslide body.The main tasks are as follows:(1)The usability of existing SAR data for InSAR monitoring in a mountainous area of Guizhou Province is studied,and the detection capabilities of different SAR data under complex terrain covered by multiple vegetation are analyzed.(2)Based on the combination of SBAS technology and Offset-tracking technology,the surface deformation monitoring of the monitoring areas is carried out by using ALOS-PALSAR 2 data and Sentinel 1 data.The two results are compared with each other to verify the accuracy of the monitoring.The monitoring objectives include deformation of small gradients and large ones,and obtained comprehensive deformation monitoring results.At the same time,according to the obtained deformation monitoring results,the detailed field investigation and verification of the region with large deformation rate in the detection area proves the large-area deformation monitoring capability and application value of SAR technology.(3)The terrain information,soil type,land use,geological type,vegetation cover and rainfall information of the study area were used as the landslide sensitivity evaluation factors,and the landslide sensitivity evaluation was carried out based on the deep learning method.On this basis,the deformation is added as a new evaluation factor,and the evaluation results of the two are compared with the accounting information of the field survey.Then this paper introduces three improved BP neural networks to evaluate the accuracy and compares the fitting accuracy of three improved network models with traditional network models.The results show that the accuracy of landslide sensitivity evaluation after adding the deformation is significantly improved.And the fitting accuracy of BP neural network model based on improved algorithm is higher than that of traditional BP neural network model.