The Research Of Parallel Mosaicking For Large Scale Remote Sensed Images Based On Task-tree Scheduling

Remote sensing technology has come a long way and still in high level of activitytill now. Signal processing for captured remote sensed images is the main technicalmethod to obtain necessarily knowledge, as the demand continue to raise, one singleimage can’t afford the whole information for applications such as environmentanalysis and region hazard warning, in these cases remote sensed image mosaic havebeen brought up. However remote sensed image mosaic is relevant to lots of remotesensed image processing, and those inevitable procedures make remote sensed imagemosaic rather lengthy. In order to improve the timeliness of these applications weneed to find a way to enhance the utilization of hardware and software.To solve these problems, this paper studies the remote sensed image mosaictechnology, parallel programming technology, task scheduling method, we alsoanalysis the existent Large scale remote sensed image mosaic algorithms, the lack ofefficiency and extendibility are always the major issues. We provide a Task-Treebased parallel mosaicking for Large scale remote sensed images algorithm, decouplethe Large scale mosaicking problem into several small ones, those small tasks will bepassed to computing nodes under the rules of logical relationship. This method use theadvantage of high performance cluster to compute and monitor all the process, wepass all the tasks whose condition is sufficient into ready queue, and will pass thosetasks into running queue when there are available nodes who will run an MPI parallelmosaic program, this multi-level parallel structure can make full use ofexisting facilities.Our method is a crucial path and dynamic scheduled–state queues method. Usingdynamic DAG(Directed Acyclic Graph) to build the TTM(Task-tree Mosaic) system.TTM system can automatically produce the task tree with less height, optimal theparallelism and reduce the CPU time. We simultaneously arrange the ability to correctthe false output and retransmission the false one, those functions provide a very goodfoundation for our system.This paper shows how to achieve the parallel mosaic system TTM for Large scales remote sensed images as designed above. We also experimentally proof theadvantage of running time and extendibility is enhanced under certain highperformance cluster circumstances.