Design and Implementation of 3D CNN Algorithm Based on FPGA

With the deepening of CNN algorithm network model, the effective implementation of 3D CNN algorithm has become a challenge. In this paper, a design architecture of 3D CNN algorithm based on software and hardware collaboration is proposed, and its feasibility is verified by simulation. A series of comparison experiments show that the proposed method reduces unit power consumption by 0.75um, and the computing time of 6-layer and 8-layer convolution is reduced by 8276ms and 161761ms, respectively, compared with CPU. Therefore, on the basis of making full use of resources and ensuring accuracy, the proposed method realizes the optimization of power consumption and running speed while ensuring accuracy.