A novel method for feature extraction of crackles in lung sound

Crackles are an important kind abnormal lung sound for detection in lung sound analysis. Focused on characteristic morphology of crackles in time-domain, a novel time-domain processing method is proposed to extract features of crackles based on the newly rising theories of Fractional Hilbert Transform. After applying the transformation of Fractional Hilbert Transform with various fractional values, exclusive timedomain features are merged and can be used as validated detection features. Experiments show great application feasibilities for such kind of wave detections. Later we use correlation functions to construct an elementary detection system, and system simulation results support the effectiveness of our work. Discussions on detection errors are followed.