基于ZYNQ平台的SVM分类器设计与实现  

Design and Implementation of SVM Classifier Based on ZYNQ Platform

作　　者：鲍温霞 黄敏 肖仲喆 BAO Wenxia;HUANG Min;XIAO Zhongzhe(School of Optoelectronic Science and Engineering,Soochow University,Suzhou Jiangsu 215006,China)

机构地区：[1]苏州大学光电科学与工程学院,江苏苏州215006

出　　处：《电子器件》2024年第6期1478-1484,共7页Chinese Journal of Electron Devices

摘　　要：研究基于Xilinx HLS高层次综合工具的SVM分类器设计,并在ZYNQ 7020平台上搭建水声信号特征分类系统对设计的SVM IP核进行测试。首先用500组水声信号特征在MATLAB中训练SVM分类网络,然后用C语言编写SVM网络分类算法,经HLS综合生成IP核。实验结果表明:所设计的基于ZYNQ平台的SVM分类器能够实现对水声信号特征值矩阵的分类,对一段水声信号特征进行有无目标分类的平均用时为6.86μs,仅为在MATLAB上运行SVM算法的1.1%,分类准确率可达99.31%。同时资源占用量少,仅为ZYNQ 7020中FPGA总资源量的10%(4347个LUT)。A new design of SVM classifier based on Xilinx HLS high-level synthesis tool is studied,and an underwater acoustic signal feature classification system is built on ZYNQ 7020 platform to test the designed SVM IP core.First,the SVM classification network is trained by using 500 groups of acoustic signal features on MATLAB.Then,the SVM classification algorithm is programmed in C language,and the IP core is generated by HLS synthesis.The experimental results show that the hardware acceleration system of SVM based on ZYNQ platform can classify the matrix of the underwater acoustic signal features,and the average time for classifying a section of underwater acoustic signal features with or without targets is 6.86μs,which is only 1.1%of the time taken for the SVM algorithm run on MATLAB,while the classification accuracy can reach 99.31%.The resource consumption is only 10%of the total FPGA resources in ZYNQ 7020(4347 LUTs).

关 键 词：支持向量机 IP核 ZYNQ 高层次综合 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]