Intelligent Identification Of Geological Landslide Based On Multi-source Data Fusion

As one of the most dangerous disasters,geological landslide causes severe loss to human life and property,which urgently requests reliable detection technologies.The prevention and control of geological disaster has become one of key works in the 13th five-years-plan and has achieved fruitful results.However,at present,geological landslide detection is mainly carried out by manual interpretation and expert judgment,which is time-consuming,inefficient and severely dependent on experts'experience.Meanwhile,more and more remote sense data from different platforms are produced,and deep learning-based artificial intelligence methods have been widely applied in geographical landslide detection.The intelligent landslide detection technology,which is based on multiple-source data fusion and deep learning,has become the trend of geological landslide detection.In this thesis,data from four sources,i.e.,Interferometric Synthetic Aperture Radar(InSAR),Digital Elevation Model(DEM),geological basement,and satellite-borne optical remote sensing image,will be fused in deep learning models to distill sufficient information for reliable landslide detection.The main work and innovation of this thesis are as follows.Firstly,a framework,consisting of InSAR deformation aggregation area recognition,slope unit generation,and important geological features subtraction,is proposed to build InSAR deformation aggregation areas,generate slope units and extract essential features of potential landslide spots,via which feature dataset can be generated for collection of feature attributes and label attributes of slope units.Then,an enhanced road recognition model is developed.Acting as both the disaster bearing bodies and the causing elements of landslide,road recognition is of importance for landslide detection.The developed model can enhance the performance in the case of road interruption and edge burr problem.Finally,a comprehensive detection model is developed,which fuses essential features of landslide and then reliably identifies potential landslide points.This thesis uses Massachusetts roads dataset to verify the proposed road recognition model,and the experimental results testify that the effectiveness of the proposed model.The proposed framework and corresponding comprehensive detection model are also evaluated on real dataset.The experimental results show that the proposed method can identify the potential hazard points of landslide accurately.