Graphics Processing Methods Based on Deep Learning in the Context of Big Data

Abstract

The purpose of this article is to conduct research on graphics processing (GP) methods based on DL in the context of big data (BD). First, this article studies the characteristics of current BD and investigates related applications. Then based on DL, this article proposes a method of processing graphics content. This method uses three different DL network structures to extract graphic features. In this paper, the OpenMP parallel model is used to optimize the graph search algorithm in parallel on the CPU platform, and the algorithm is optimized in parallel by using the principle of program locality, reducing synchronization overhead and load balancing. Thirdly, in view of the irregularity of the memory access of the graph search algorithm, the FPGA algorithm hardware accelerator is customized using BD technology. Finally, it introduces the development tools of DL algorithm hardware acceleration, and compares them with traditional processing methods. Comparative experiments show that the image processing method based on DL proposed in this paper has an average recognition rate of about 10% higher than traditional processing methods, and has better results, providing an important reference for the development of image processing.