Design of airborne ultra-lightweight convolutional neural network accelerator

Abstract

The huge weight parameters and complex network layer structure of convolutional neural network make its computational complexity too high, and the required computing resources and storage resources also increase rapidly with the increase of network layers, so it is difficult to deploy in airborne embedded computing systems with strict requirements on resources and power consumption, which restricts the development of airborne embedded computing systems towards high intelligence. Aiming at the demand of ultra-lightweight intelligent computing in the resource-constrained airborne embedded computing system, a set of optimization and acceleration strategy of convolutional neural network model is proposed. After ultra-lightweight processing of the algorithm model, a convolutional neural network accelerator is built by combining acceleration operators, and the function verification of network model reasoning process is carried out based on FPGA. The results show that the established accelerator can significantly reduce the occupancy rate of hardware resources and obtain a good algorithm speedup ratio, which is of important significance for the design of airborne embedded intelligent computing system.

Keywords

• embedded computing system
• convolutional neural network
• ultra-lightweight
• hardware accelerator
• fpga implementation