Classification of Lung Sounds Using CNN-Attention

EasyChair Preprint 4356

Abstract

Respiratory disease is a kind of disease which causes a high mortality rate, whether in rural areas or in cities. It is necessary to detect respiratory diseases in advance, and with the rapid development of deep learning in recent years, it will become a new trend of disease detection to analyze and calculate the respiratory sound through the computational power of computer. Attention mechanism has recently been proposed to deal with sequence problems (such as audio, text, etc.),It is claimed to be a kind of network that can replace RNN, LSTM and GRU. By imitating human visual attention mechanism, it can learn a weight distribution of data features, and then apply this weight distribution to the original features, so that the task mainly focuses on some key features, neglects unimportant features, and improves task efficiency.On the basis of the ICBHI benchmark data set, we propose a network, which combines CNN and attention mechanism to classify audio data and distinguish abnormal breath sounds from normal respiratory sounds. And we have carried on the experiment to test the robustness of the model.

Keyphrases: Attention Mechanism, CNN, deep learning

@booklet{EasyChair:4356,
  author    = {Chenghan Li and Huaichang Du and Bing Zhu},
  title     = {Classification of Lung Sounds Using CNN-Attention},
  howpublished = {EasyChair Preprint 4356},
  year      = {EasyChair, 2020}}