Feature Extraction, Classification And Identification Of Anesthesia Breathing Sound Signal

The breathing sounds are collectively referred to as the sound of the humanrespiratory system with external ventilation process, contain a large number ofpathological and physiological information. During surgery, monitoring of breathingsounds in patients under anesthesia, it means to predict the patients’ condition earlierthan other surveillance. According to the characteristics of the anesthesia breathingsound signal, digital signal processing methods are used to try to study the extractionmethod from a different angle, these studies help to establish a software which couldbe used on the breath test instrument and provide a new basis for doctors todetermine the condition of patients to avoid respiratory failure during surgery.First of all, HHT, wavelet transform and neural network is used in study ofanesthesia breathing sound, then the basic theory of HHT, wavelet and neuralnetwork is described.Secondly, the envelope of anesthesia breathing sound signal is extracted by HHT.The envelope is relatively smooth and has a self-adaptive by this method. There is apositive correlation between tidal volume and envelope of signal. Compared withlesion and no lesion of breathing signals, the energy and statistics of envelope areextracted for vectors of neural network, recognition rate is93.3%. It is effective ofthe method.Finally, because that single feature can not describe the whole information ofanesthesia breathing sound, wavelet packet algorithm is used to extract the energyfeature. Inspiratory and expiratory is decomposed of four-layer by wavelet packettransform, after selection of optimal basis, the equivalence relation between waveletpacket coefficient and energy of anesthesia breathing sound signal is established.Wavelet packet coefficient energy of each band is considered as features to constructthe normalized eigenvectors. The result is identified by BP neural network,recognition rate is95%.