Spectrum Feature Extraction and Signal Detection of Wheeze in Time-frequency domain

We propose a new algorithm to detect wheeze signals in respiratory sounds based on S transform time-frequency spectrum analysis. Wheezes are of sinusoidal morphological characteristics in the time domain, and its features are difficult to extract directly. This paper introduces S transform, which shows high time analysis resolution in high frequency field and high frequency analysis resolution in low frequency field, to extract features of wheezes in the time-frequency domain. Respiratory sounds were transformed to time-frequency domain by S transform, and then, two-dimension spectrum image features, which are corresponding to wheeze signals, were extracted. Thus, wheeze signal detection has been realized. Experiments show that the algorithm do well in the case of training and detection for each subject, with a sensitivity value as high as 100% for detection and position prediction value higher than 98%. However, the method failed to extract global features of wheeze from different subjects, which requires future exploratory research.