错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

FPGA-Based Implementation of Convolutional Neural Networks Architecture for Detection of Dysphonia

  • Jyoti Mishra,
  • R. K. Sharma

摘要

With the advancement of modern technology, utilizing deep learning in bio-medical applications has become predominant. Deep learning-based classifiers are coming into the picture these days as an important automatic disease detection method. This study shows the implementation of the convolutional neural network (CNN) model for dysphonia disease detection on Xilinx zynq-7000 FPGA board using hardware description language, i.e., Verilog. Dysphonia refers to voice disorder, which causes degradation of voice quality slowly with time. Some patients even lose their voices for a certain period. Earlier detection of any disease is important to take necessary precautions. This our work aims to identify the disease at its early stages to reduce its severity caused by it. The CNN algorithm has been designed, trained, and validated on MATLAB with a dysphonia dataset which has been taken from the Saarbruecken voice database. For this study, 88 dysphonic and 72 healthy voice samples of males and females are considered. The MATLAB pre-trained model is further used to extract weights and biases to implement the CNN model on hardware. The FPGA implementation has improved the performance and speed of image classification when compared with the CNN model on MATLAB. We obtained training and validation accuracy of 83.12% and 76.89%, respectively, using MATLAB deep network designer, while in the case of FPGA, implementation better efficiency and accuracy are obtained.