High Spatial Resolution Remote Sensing Data Computing Pattern Based on Feature Primitives

Abstract

With the developments in satellite sensor technology, data acquisition technology developed rapidly and with the start of a series of space-based observation network for Earth science, such as EOS, GTOS, ECOS, GOOS and etc., high performance processing and analysis of tremendous data becomes the bottleneck faced by earth observation. According to the differences of the computational behavior and the computing emphasis, this paper divides high spatial resolution remote sensing data computation into two classes: deep-computation and active-computation. Deep-computation (from data to features) is to extract the feature primitives through certain methods, so deep-computation emphasizes particularly on computing amount. Active-computation (from features to knowledge) is based on the feature primitives obtained by deep-computation. Firstly the spatial relationships between the feature primitives are computed, then the decisions can be made effectively and efficiently with domain knowledge and domain models through web services, so deep-computation emphasizes particularly on intelligence of computation. Finally, by using the distributed computing technique, a case study of information extraction and target recognition from remote sensing image based on feature primitives was given to illustrate and testify the ideas mentioned above. Experimental results shows that the high spatial resolution remote sensing data computation pattern based on feature primitives is feasible and it is practically meaningful to resolve the problem of huge geo-spatial data computation.