ABSTRACT
Aerial remote sensing technology can obtain surface information periodically, quickly, comprehensively and non-contact, which has an absolute advantage in the identification of ancient landslides. However, the extraction of landslide information based on remote sensing is only a visual interpretation of the image combined with the morphological characteristics of the landslide. Its accuracy is high, but the degree of automation is relatively low, and it is not conducive to rapid and large-scale extraction. In order to overcome this disadvantage, this paper intends to study a multi-level object construction method based on content awareness. On this basis, an object depth learning feature expression method based on the combination of full convolution network and multi-level object model is developed to automatically learn and extract space, shape, color and other feature information on multiple scales. This can improve the theory of image-based disaster scene understanding and pattern classification, and provide some theoretical and technical support for smart city and environmental perception.
