Research On Landslide Information Extraction Method Based On Optical Remote Sensing Image

Landslide is one of the most widespread natural geological disasters in the world.Landslide disasters are increasingly occurring in some areas,which poses a major threat to people’s transportation and safety of life and property.After the landslide occurs,it has important guiding significance to detect and obtain landslide information quickly and accurately for carrying out emergency rescue work.The unmanned aerial vehicle(UAV)remote sensing technology has been widely used in emergency rescue because of its advantages of fast imaging speed,high resolution and low cost In the absence of pre-disaster auxiliary data,can the accuracy of landslides extraction only from the optical image of UAV to meet the emergency requirement?Is there a change detection method or technical process based on the existing optical image before the disaster and after the disaster,which can not only improve the speed of landslide information acquisition,but also improve the extraction accuracy?In view of the above two questions,this paper studies the method of landslide hazard extraction based on UAV optical image and.The main research contents are as follows:(1)A method process for quickly extracting landslide information based on single-time UAV visible light image was proposed.The visible light image of the UAVs in the three areas of the"4.20 Lushan Earthquake"(Lushan County,Taiping Town,Baosheng Township)were selected,and the object-oriented method was used to obtain the suspected landslide area by using landslide features such as texture,spectrum and shape.The accuracy analysis was carried out on the pixel angle.It is concluded that the method is more suitable for extracting landslide location information in the research area with low vegetation coverage and less interference from other objects.The accuracy of landslide location information extraction in the study area is 80.06%,while the accuracy of landslide extraction in the study area with low vegetation coverage but more other land objects is only 63.75%.It shows that the method of extracting landslide accuracy is more unstable,which is greatly affected by DEM resolution and the overall extraction accuracy is low.(2)For the case of low-precision accuracy of single-time UAV visible light image extraction with low vegetation coverage but more other land objects,we selected the pre-disaster(March 2012)space image and post-disaster(April 2013)UAV aerial image of Lushan County and used object-oriented method to segment them.The post-classification comparison change detection method was selected to detected the change of the two classified images.Furthermore,the location and size of the landslide disaster were extracted,and the extraction accuracy arrives 80.7%.From the landslide extraction results,the change detection method based on post-classification comparison does not require the same type of sensor,and the accuracy of landslide extraction is higher than that based on single-time UAV visible image.(3)In order to improve the accuracy of landslide detection based on post-classification comparison change detection method,the shadow detection index(SDI)was constructed in the preprocessing process and the color constancy algorithm was introduced to remove the shadow in the post-disaster image.Then the method of post-classification comparison change detection was used to extract landslide,and the extraction accuracy is up to 90.1%,which is significantly higher than that of the method of change detection without considering landslide shadow.(4)Considering the suddenness of landslides and the difficulty in obtaining pre-disaster high-resolution images,discussed how to improve the method of quickly extracting landslide information based on visible light images of single-time UAVs.By analyzing the advantages and disadvantages of different visible vegetation indices and combined with the landslide texture and spectral characteristics,analyze texture features in the pre-processing process and combine the radiance estimation method based on vegetation index to enhance the bare land information in the research area.Then,combined with the method(proposed in(1))of quickly extracting landslide information from the visible light image of the single-time UAV is used to extract and analyze the landslide information in the visible light image of the UAV in the local area of Lushan County.The correct extraction accuracy is 91.3%.However,this method is complicated and greatly affected by the selection of texture features and thresholds.(5)At the end of the paper,the process of landslide extraction from four-band(visible and near infrared)optical images is briefly discussed,and the role of near infrared band in the process of landslide extraction from optical images is emphasized.In this paper,four-band Landsat7 images of visible and near infrared are used to simulate the four-band images of UAV.The landslide is extracted by the principal component analysis change detection method combined with NDVI vegetation index,and the extraction accuracy is 63.49%.The main reason for the low accuracy is that the Landsat7 image resolution is lower than that of the UAV image,but the method is simple and fast,and can be used as a technical reference for extracting landslides from four-band UAV images.Based on the research of the optical image extraction landslide method of the UAV,some landslide extraction functions are realized by programming.The experiments show that the landslide extraction method based on object-oriented and change detection high-resolution optical remote sensing image can extract landslide information quickly and effectively,which provides a reference for the rapid extraction landslide from high-resolution optical remote sensing image.