Research On Landslide Information Extraction In Baoxing Base On GF-1 Image

Our country mountainous area is broad, under the influence of human factors, geological tectonic movement and rainfall, the landslide disasters occur frequently. It often leads to the village of buried, traffic disruption, farmland damage, etc., and causes huge losses of life and property of people. According to statistics, China’s annual economic loss caused by landslide disaster of more than billion. How to quickly find and extract landslide information of landslide disaster prevention and mitigation are extremely important. Remote sensing technology has the characteristics such as large detection, strong timeliness, low cost, etc. with the development of high-resolution, hyperspectral remote sensing technology, through the remote sensing technology for landslide information extraction has entered a new stage.The traditional method of landslide extraction is mainly based on field investigation and artificial visual interpretation to identification and classification of disaster information. The method has some disadvantages such as the long period of information extraction, the low accuracy of classification and high cost. The pixel-oriented method can realize rapid extraction of landslide information, and it has been widely used in low-resolution images. however it has the salt-pepper effect in landslide extraction base on high-resolution images. The object-oriented method merges the pixels with similar features into an object, can effectively avoid pixel-oriented method of salt-pepper effect and improve the classification accuracy.This paper taken FengTong town as the research area, using object-oriented method to identification and classification of landslide information base on DEM and GF-1 image. The main results and innovations are summarized as follows:(1)By analyzing the landslide characters of research area, put forward a quick and effective way of landslide extraction with EDVI,soil index and first principal component information, It was validated through the pixel-oriented and object-oriented methods. It was shown that the object-oriented method can successfully avoid the effect of salt-pepper caused by the spectral diversity of high-resolution image and made the results more scientific and accurate.(2)The object-oriented method was used for the extraction of landslide information, and the results show that object-oriented method has higher classification accuracy(Kappa coefficient 88%)。multi-scale segment algorithm was applied to build up different layers according to the size of the segment scale. Then analyzed the spectrum, texture and context relationship according to the feature category extract from different layer. set up rules of landslide information extraction, and extracted the landslide information.(3)Extracted more than 40 landslide by Object-oriented method, these landslides are mainly concentrated in the southwest of the study area, the ratio of the length to width is 1-2.5, the height of the slope is 1500-2500, and the slope is 35-55 degrees.(4) Put forward a landslide metadata registration mechanism base on the DOA architecture. It can be able to unified data management and lay the foundation for the follow-up study.