Proceedings Article | 6 August 2009
KEYWORDS: Landslides, Remote sensing, Visualization, Image enhancement, Image filtering, Radiometric corrections, Vegetation, Image segmentation, Image analysis, Lithium
After Wenchuan earthquake disaster, landslide disaster becomes a common concern, and remote sensing becomes more
and more important in the application of landslide monitoring. Now, the method of interpretation and recognition for
landslides using remote sensing is visual interpretation mostly. Automatic recognition of landslide is a new and difficult
but significative job. For the purpose of seeking a more effective method to recognize landslide automatically, this
project analyzes the current methods for the recognition of landslide disasters, and their applicability to the practice of
landslide monitoring. Landslide is a phenomenon and disaster triggered by natural and artificial reasons that a part of
slope comprised of rock, soil and other fragmental materials slide alone a certain weak structural surface under the
gravitation. Consequently, according to the geo-science principle of landslide, there is an obvious change in the sliding
region between the pre-landslide and post-landslide, and it can be described in remote sensing imagery, so we develop
the new approach to identify landslides, which uses change detection based on texture analysis in multi-temporal
imageries. Preprocessing the remote sensing data including the following aspects of image enhancement and filtering,
smoothing and cutting, image mosaics, registration and merge, geometric correction and radiation calibration, this paper
does change detection base on texture characteristics in multi-temporal images to recognize landslide automatically.
After change detection of multi-temporal remote sensing images based on texture analysis, if there is no change in
remote sensing image, the image detected is relatively homogeneous, the image detected shows some clustering
characteristics; if there is part change in image, the image detected will show two or more clustering centers; if there is
complete change in remote sensing image, the image detected will show disorderly and unsystematic. At last, this paper
takes some landslides at the Parry Lake as a case to implement the effectiveness of the new method in the application of
landslide identification, which takes SPOT-5(Oct 10, 2003) and ALOS-AVNIR2(Sep 19, 2007) as the respective data
sources of pre-landslide and post-landslide. The result shows that the method based on change detection is available of
landslide information in arid area and other area where there is not obvious spectral difference between landslide mass
and the background. Certainly, it will be more available of such area where there is obvious spectrum difference between
landslide region and the background.
