The Research Of Parallel Task Scheduling Algorithm In The Ship Image Processing

The image information taken by ship is the important basis of strategic decisions formodern ships. However, influenced by the marine environment, motion blur will be happenedon the sensing images. The existence of the blurred images will affect the ship managers’strategic decisions, which will not meet the requirements of the reliability of the ships. So weneed to restore the blurred images accurately and quickly. The amount of images taken byships and the amount of calculation in the process of image restoration will be very huge. Ifwe use a single processor for image restoration, obviously, we can’t meet the requirements ofreal-time and rapidity. Therefore, we should bring in parallel computing technique for imagerestoration. How the task scheduling algorithm works directly affects how the parallelcomputing performs. So researching the task scheduling algorithm of ship image processingparallel computing has an important impact on the ship image parallel recovery.First, based on the typical cluster system, we build ship cluster image processing systemand analyze the specific part of the system and its characteristics, and provides hardware andsoftware for the bringing in of parallel computing in ship image processing. Ship clusterimage processing system is a typical cluster system, but it also has its own characteristics.This paper analyses the general process of image restoration. In order to solve the problemssuch as large amount of calculation for images restoration process and too much time waste,we come up with that combining the parallel-computing technique and ship image processing.Dividing the process of image restoration, we will get the DAG model diagram of imagerestoration parallel computing. Based on the DAG model diagram, we come up with the shipparallel image restoration algorithm.Secondly, ship cluster image processing system has the characteristics of heterogeneity.The existence of the heterogeneity makes parallel processing algorithm costing differentexecution time on each nodes of the cluster system, which finally will lead to the loadimbalance and affect the efficiency of parallel computing. So selecting a kind of appropriatescheduling algorithm for parallel image restoration is very important. In the research of taskscheduling algorithm, this paper analyzes the key path algorithm in detail. In the DAG modeldiagram, this algorithm can handle the key task first and assign the task evenly, but it has theproblem of excess communication loss and confusion of the task assignment. So this paperintroduces the task duplication into the key path algorithm. The task duplication technique cancopy the father task to the same node as the task, which will reduce the communication loss between them. This algorithm not only solves the problems of the key path algorithm, but alsoimproves the node CPU utilization.Finally, this paper conducts the simulations for the two kinds of scheduling algorithm onthe ship cluster system. Comparing the execution time, the parallel acceleration ratio and theparallel efficiency, we will find the key path algorithm based on task duplication will be fasterthan the common key path algorithm, and the CPU utilization of the cluster system node hasgot fully improved.