Computer Engineering and Applications ›› 2021, Vol. 57 ›› Issue (23): 263-269.DOI: 10.3778/j.issn.1002-8331.2103-0006

• Engineering and Applications • Previous Articles     Next Articles

Lightweight Network Heart Sound Classifier Based on Hardware Acceleration

DU Yuzhang, PAN Jiahua, ZONG Rong, SU Wei, WANG Weilian   

  1. 1.School of Information Science and Engineering, Yunnan University, Kunming 650504, China
    2.Fuwai Yunnan Cardiovascular Hospital, Kunming 650102, China
  • Online:2021-12-01 Published:2021-12-02

基于硬件加速的轻量级网络心音分类器

杜煜章,潘家华,宗容,粟炜,王威廉   

  1. 1.云南大学 信息学院,昆明 650504
    2.云南省阜外心血管病医院,昆明 650102

Abstract:

In recent years, convolutional neural networks have been widely used in heart sound signal classification. In order to meet the requirements of low power consumption and mobility of the machine-assisted diagnosis system for congenital heart disease, a kind of heart sound classifier suitable for FPGA hardware platform based on the lightweight neural network MobileNet is put forward in this paper. The deep convolution, point-by-point convolution, and maximum pooling modules of the heart sound classifier are designed through HLS. To reduce network parameters and calculations for the heart sound classifier, the deep separable convolution is used. At the same time, the running speed of the classifier is improved by using multi-pixel, multi-channel parallelism, and fixed-point quantization. The experimental results show that in terms of computational efficiency, compared with the traditional network, the heart sound classifier achieves approximately 14 times faster than general-purpose CPU platforms.

Key words: hardware acceleration, FPGA, MobileNet, heart sound signal, high level synthesis

摘要:

近年来,卷积神经网络被广泛应用于心音信号分类。为满足先心病机器辅助诊断系统低功耗、可移动等方面需求,基于轻量级神经网络MobileNet,实现了一种适用于FPGA硬件平台的心音分类器。心音分类器的深度卷积、逐点卷积与最大池化等模块通过高层次综合进行设计。该心音分类器在利用深度可分离卷积减少网络参数与运算量的同时,通过多像素多通道并行及定点量化等方式,提升了分类器运行速度。经心音数据集实验结果表明,在计算效率方面,该心音分类器在FPGA上相较于在通用CPU上实现约14倍加速。

关键词: 硬件加速, FPGA, MobileNet, 心音信号, 高层次综合