基于RISC-V架构的异常检测系统设计

Design of anomaly detection system based on RISC-V architecture

本文针对在金属的生产和制造过程中的金属图片表面异常,设计了一种基于第五代精简指令集计算机的片上系统。运用结构精简、准确率高的生成对抗网络算法实现了对异常金属图片数据集的分类;在Xilinx Nexys4 DDR2 FPGA开发板上通过设计的硬件片上系统生成比特流文件,实现了整个端到端的识别系统,并结合已有的数据测试集验证了识别系统的实用性。实验结果表明,在异常金属图片数据集中,输入图片分辨率为64×64,在板上系统内实现了高达99.9%的识别准确率,为工业生产制造与异常检测提供了质量和效率保证。

Aiming at the abnormal surface of the metal pictures in the metal production and manufacturing process, a SOC based on the fifth-generation simplified instruction system RISC-V is designed. The detection and classification of anomaly metal images data sets are realized by using a simplified and highly accurate generative adversarial network algorithm. The whole end-to-end identification system is verified and implemented on Xilinx Nexys4 DDR2 FPGA development board. The experimental results show that in the abnormal metal images data set, the input images resolution is 64×64, and the recognition accuracy rate achieved in the on-board system can reach 99.9%, which provides the guarantee of the quality and efficiency for industrial production, manufacturing and anomaly detection.