摘要
本文描述了一款采用内嵌DSP芯片及MCU搭建的AI推理设备的设计与实现。设计中详细论述了利用MCU的USB接口,遵循RNDIS协议从上位机获取图像帧,并通过MCU的SPI接口与DSP进行指令和数据交互的方法。同时介绍了在DSP上运行卷积神经网络算法,使用MobileNet-SSD网络模型进行物体识别,并通过USB接口返回算法结果到上位机的过程。
This paper describes the design and implementation of an AI reasoning device based on embedded DSP and MCU chips.In the design,the MCU’s USB interface is used to transmit video frame from the host according to RNDIS protocol,the instructions and data are transmitted between MCU and DSP through SPI interface.Also,this paper introduces the process of running CNN algorithm on DSP,using MobileNet-SSD network model for object,and returning the result of algorithm to the host by USB.
作者
陈智
CHEN Zhi(Department of ISP,Fuzhou Rockchip Electronics Co.,Ltd.,Fuzhou,China,350003)
出处
《福建电脑》
2019年第6期29-32,共4页
Journal of Fujian Computer