Research On Parallel Optimization Technology Of Invoice Date Extraction Based On .Net

The constant development of multicore processor hardware technology has opened new possibilities and opportunities for software performance improvement.However,traditional serial programs cannot exploit the computing power of multicore processors effectively.In order to utilize the advantage of multicore processors fully,it is very necessary to reconstruct original serial programs to parallel ones.According to real project requirements,this paper need study and implement the parallel optimization of the vehicle invoice date extraction program to take full advantage of multicore resource to accelerate the execution efficiency of the program.This paper firstly studies a new task-based parallel programming technology using the Microsoft.NET Framework 4.0,and summarizes three parallel programming models,which are task parallelism model,data parallelism model,and mixed parallelism model,supported by TPL(Task Parallel Library).Then through parallel analysis of the invoice date extraction program,we can find parallelizable modules of the serial program.And according to the decomposition manner of these modules,this paper determines the corresponding parallel transformation schemes.Finally,this paper utilizes the above parallel programming models to realize the parallel optimization of the original serial program.This paper mainly designs and implements the following four parallel optimization schemes:(1)specific to eight segmented numeric character images(invoice date is made up of eight numbers),this paper realizes the local data parallel optimization of subsequent operations of eight character images using data parallelism model;(2)regarding all invoice images as the parallel processing object,this paper utilizes data parallelism model to implement the overall data parallel optimization of invoice images recognition.Through this optimization scheme,each invoice image can be processed with others concurrently;(3)the pipeline mode implemented b,y mixed parallelism model is used to parallelize each processing stage of the invoice image recognition program on different images;(4)through the deep research on each processing stage of the program,this paper explores a parallel optimization method of load balanced pipeline which is specific to the program,and this method can be implemented by mixed parallelism model.Through performance analysis of the above four parallel optimization programs,we find the overall data parallel optimization and the load balanced pipeline mode can achieve an almost liner speedup.And in the eight-core environment,the speedup of these two parallel programs is about 7.2x.Therefore,we can apply these two schemes to the real application to meet user needs of functionality and performance.