Studies on parallel and distributed RS image issuance system based on SVM

Abstract

Remote sensing (RS) data has the property of large volume, multi-sensor, multi-resolution, multi-spectrum and multi-epoch, which introduces some difficulties for RS image geo-information sharing and issuance. The following are two prominent difficulties: firstly, background servers need to handle heavy data processing. Secondly, there is no efficient organization of RS image data for storage and management on the background. This paper adopts parallel computers based on sharing virtual memory (SVM) as background servers for the RS images distributed issuance system, and analyzes parallel procedure design for image data based on parallel computer clusters. Furthermore this paper has analyzed and tested the parallel process of image extraction, image matching and image compressing of the background. Test results have indicated that these parallel computers could promote the background performance distinctly. As for the organization of image data, this paper has constructed a solid index mechanism of 'Pyramid, Block, Layer, Epoch' according to the properties of RS data and has established a corresponding logical database. RS data still distribute on parallel computers in files, not in a database as blobs. All server side processes are performed through the logical database. According to the above ideas, this paper has designed the issuance system, which provides storage, management, query and browsing of RS image data in a distributed environment.