Landslide Extraction From Sar Image Intensity And Coherence Coefficient

Landslide is a natural geological disaster with multiple characteristics such as being frequent,sudden,destructive,and widespread,which seriously affects transportation as well as industrial and agricultural production,causing great loss of people's lives and property.Southwest China,with abundant rainfall,wide mountainous area and large topographic fluctuation,is a high incidence area of landslide disaster,and it is difficult to obtain disaster information and rescue.It's essential for post-disaster rescue,disaster management and prevention of secondary landslide disasters to obtain accurate information of landslide location,shape and coverage.To acquire relevant information,remote sensing is one of the useful measures which has the advantages of wide coverage,low cost and rapid data acquisition,etc.However,it is difficult for optical remote sensing to obtain effective information about landslides in mountainous areas in time due to the influence of light and weather.The Synthetic Apeture Radar(SAR)satellite has the capability of all-day and all-weather observation of the earth,and can timely acquire the observation images of landslide areas and thus extract landslide information.Therefore,this master thesis works on the SAR image landslide extraction method.The specific work and achievements are as follows:(1)Working on the data selection problem of extracting landslide from SAR image.Based on the case of “6·24” Xinmo village landslide in Mao county,analyzing the influence on the landslide extraction with SAR image's geometric distortion,such as layover and foreshortening,by using the satellite data from Sentinel-1A.And it reached a conclusion that an image in which the chosen aspect of landslide wasn't directed towards the same as the incident direction of radar wave is a rational choice to avoid the effect of the image foreshortening.(2)Describing three unsupervised classification methods of remote sensing images,and a landslide extraction method based on SAR multi-temporal intensity information is proposed.Temporal SAR intensity images of Xinmo village in Mao county are taken as the data,and the EM algorithm is used to excute the landslide extraction experiment based on SAR intensity information.The experimental results show that the correct rate of landslide extraction using temporal SAR image intensity information is only 3.96%.The main reason is that the intensity information of SAR image is not so different among different types of objects,and the geometric distortions such as layover,foreshortening and speckle noise in SAR image leads to the fact that the intensity information of SAR image in these areas can not reflect the real scattering characteristics of objects.Therefore,it is difficult to distinguish the actual types of objects only based on intensity information.Experimental analysis shows that traditional remote sensing classification based on SAR image intensity information is not suitable for landslide extraction of SAR images.(3)Aiming at the problem of extracting landslide by using unsupervised classification method based on SAR image intensity information,and considering the characteristics of surface change caused by landslide,change detection method is introduced for landslide extraction by using the change of SAR intensity information at different time.Based on the case of “6·24” Xinmo village landslide in Mao county,producing the change image by using mean ratio detection(MRD),and then obtaining the landslide extraction results through the process of Otsu segmentation and mathematical morphology.The experiment shows that using change detection method based on intensity can reduce the error caused by geometric distortion such as layover,foreshortening and shadow,and the accuracy of landslide extraction is better than that of unsupervised classification method.However,the correct rate of landslide extraction by change detection method is only about 15%.This is mainly due to the high vegetation coverage in the experimental area.And the instability of vegetation's scattering characteristics in time brings a lot of false change information which leads to the unsatisfactory accuracy of landslide extraction.The experiment shows that it's also difficult to extract landslide effectively,especially that in the vegetation covered area,when changing detection method based on SAR image intensity information.(4)As there are lots of problems in using SAR intensity information to extract landslide,moreover,considering that SAR image contains both intensity and phase,and the coherence between the landslide and the surrounding environment(e.g.vegetation)varies with time and space,this thesis proposes a change detection method based on coherence coefficient to extract landslide.On the basis of analyzing the temporal coherence of the study area,selecting the coherence coefficient maps that one is across the occurrence time of landslide and the other is after the landslide as data,adopting change detection method for landslide extraction.The experimental results show that the change detection method based on coherence coefficient eliminates the false change caused by vegetation,compared with the detection of landslide based on the change of intensity information,and the correct rate of landslide is over 90%.The results show that the landslide extraction by change detection method can overcome the problem that the SAR images can not reflect real information in the area with landslides geometric distortion,and adopting coherence coefficient as data for change detection avoids false change information caused by vegetation.Therefore,the landslide in vegetation-covered area can be effectively extracted by using coherence coefficient change detection which combines both intensity and phase of SAR images.