MWRS-BFSC + CNN2D

Sik-Ho Tsang

·

Follow

3 min read

·

May 18, 2024

--

Listen

Share

Research on bark-frequency spectral coefficients heart sound classification algorithm based on multiple window time-frequency reassignment
MWRS-BFSC + CNN2D, by Yunnan University, Kunming Medical University, and Fuwai Cardiovascular Hospital of Yunnan Province
2024 JBH (Sik-Ho Tsang @ Medium)

Phonocardiogram (PCG)/Heart Sound Classification
2013 … 2023 … [CTENN] [Bispectrum + ViT]
==== My Other Paper Readings Are Also Over Here ====

• Bark-Frequency Spectral Coefficient (BFSC) with multiple window time-frequency reassignment (MWRS) is used as features input into CNN2D for heart sound classification.

Outline

1. MWRS-BFSC + CNN2D
2. Results

1. BFSC + CNN2D

1.1. Preprocessing

• Due to no accurate heart sound segmentation algorithm, random segmentation is used.
• The segment has 256 sampling points, which is 51.2ms long, the stride used is 128 sampling points.

1.2. MWRS-BFSC

• The signal is pre-emphasized. Then Short-time Fourier Transform (STFT), S, is extracted.

• The corresponding energy spectrogram, E, is:

• BFSC is sensitive to noise. To increase the resolution, time-frequency spectogram with reassignment, RS, is used:

• where:

• Multiple window function is then used.

• where k=4 and dk is just used for averaging.

1.3. CNN2D

• CNN2D, CNN1D, GRU, and LSTM are tried as above.

2. Results

CNN2D is found to be the best as above.

MWRS-BFSC is found to be the best.

k=4 is the best.

MWRS-BFSC + CNN (CNN2D) is the best.