A Deep Learning Prediction Process Based on Low-power Heterogeneous Multi Core Architecture

With the rapid development of machine learning both in theory and practice in the past decade. And recently, it is widely used in applications and cloud services. As the emerging field of machine learning, deep learning shows excellent ability in solving complex learning problems. In this paper, we designed a deep learning prediction process based on low-power heterogeneous multi core architecture. Firstly, the fundamental principle of image recognition method based on deep learning reviewed as the basis of the research. Secondly, a set of key algorithm design to parallel access and process image for object detection based on Parallella multi core platform was proposed to improve the detection speed and the computational resource efficiency on single node. Thirdly, Rockchip RK3288 SoC with 4 Arm Cortex-A17 cores hardware platform, Xilinx Zynq and Adapteva Epiphany combined heterogeneous multi core hardware platform was introduced. Some key designs based on Parallella board's architecture to achieve image recognition was proposed to improve the recognition speed and the computational resource efficiency. Finally, The experimental results that based on Parallella board indicate that the proposed image recognition system can achieve nearly 14.8 times speedup than dual-core Arm which was integrated in Parallella board with similar accuracy and achieve 8.6 times speedup than RK3288 board which has the newest series of high-performance Arm core CPU as the control included 4 Arm Cortex-A17 cores.