Automatic classification of childhood asthma and pneumonia based on cough sound analysis

Cough is an important basis for the diagnosis of respiratory diseases, and it has important research significance to predict and classify the types of coughs in childhood diseases. Aiming at the difficulty in distinguishing between asthma and pneumonia in common pediatric diseases, a method based on cough sound analysis is proposed to realize the automatic classification of pneumonia and asthma. In order to improve the classification accuracy of cough sounds, in feature extraction, the Mel frequency cepstrum coefficient and short-term energy are combined and applied to the cough sound classification system, so as to achieve the prevention of children's respiratory diseases. The cough sound fragments suffering from asthma and pneumonia are preprocessed by pre-emphasis, framing and other operations, the 24-dimensional mel frequency cepstral coefficient and short-term energy mixing feature parameters are extracted and applied to the support vector machine classifier. In the same experimental environment, using specificity, sensitivity, and accuracy as measurement indicators, the classification performance of the feature parameter extraction method and the traditional feature extraction method are compared. The experimental results of multiple sets of cough audio samples show that the use of this mixed feature parameter improves the performance of various classifications with less complexity, and the classification accuracy rate reaches 90%, which has high feasibility and effectiveness.