BROWSE
Detail
Conference Paper
MBRC 기법을 적용한 AST 모델 기반 호흡음 분류 성능 분석
Cited - time in 
Share 
Share
- Issue Date
- 2024-06
- Citation
- 대한전자공학회 학술 대회 (하계) 2024, pp.1728-1732
- Publisher
- 대한전자공학회
- Language
- Korean
- Type
- Conference Paper
- Abstract
- In this paper, we propose a modified BRC technique to enhance the performance of respiratory sound classification using deep neural networks. When applied to the ICBHI dataset resampled to 16 kHz, the AST model utilizing the original BRC technique fails to improve classification performance and actually degrades it. This is because the clipping boundaries, which remove high frequency regions with minimal energy from the two dimensional acoustic features, occur over a wide frequency range, leading to inadequate training of the AST model. By addressing this issue with the MBRC technique, the AST model achieves a score of 60.98%, corresponding to a 2.47% performance improvement over the model without MBRC.
- KSP Keywords
- Classification Performance, Deep neural network(DNN), High frequency(HF), Minimal energy, Respiratory Sound(RS), Sound Classification, Wide Frequency Range, acoustic features, frequency regions, neural network(NN), performance improvement