LungRN+NL: An Improved Adventitious Lung Sound Classification Using Non-Local Block ResNet Neural Network with Mixup Data Augmentation

Yi Ma, Xinzi Xu, Yongfu Li


Performing an automated adventitious lung sound detection is a challenging task since the sound is susceptible to noises (heartbeat, motion artifacts, and audio sound) and there is subtle discrimination among different categories. An adventitious lung sound classification model, LungRN+NL, is proposed in this work, which has demonstrated a drastic improvement compared to our previous work and the state-of-the-art models. This new model has incorporated the non-local block in the ResNet architecture. To address the imbalance problem and to improve the robustness of the model, we have also incorporated the mixup method to augment the training dataset. Our model has been implemented and compared with the state-of-the-art works using the official ICBHI 2017 challenge dataset and their evaluation method. As a result, LungRN+NL has achieved a performance score of 52.26%, which is improved by 2.1–12.7% compared to the state-of-the-art models.


 DOI: 10.21437/Interspeech.2020-2487

Cite as: Ma, Y., Xu, X., Li, Y. (2020) LungRN+NL: An Improved Adventitious Lung Sound Classification Using Non-Local Block ResNet Neural Network with Mixup Data Augmentation. Proc. Interspeech 2020, 2902-2906, DOI: 10.21437/Interspeech.2020-2487.


@inproceedings{Ma2020,
  author={Yi Ma and Xinzi Xu and Yongfu Li},
  title={{LungRN+NL: An Improved Adventitious Lung Sound Classification Using Non-Local Block ResNet Neural Network with Mixup Data Augmentation}},
  year=2020,
  booktitle={Proc. Interspeech 2020},
  pages={2902--2906},
  doi={10.21437/Interspeech.2020-2487},
  url={http://dx.doi.org/10.21437/Interspeech.2020-2487}
}