A novel heart sound-based identity recognition method

Identity recognition has been further studied based on heart sound due to its characteristics of being unique and difficult to replicate. However, recognition accuracy suffers from different statuses, especially pre-exercise and post-exercise. To address this issue, a novel adaptive segmentation method-based feature extraction method is proposed for automatically extracting a 36-dimensional feature in the time and frequency domains. Moreover, to evaluate the redundancy and efficiency of the extracted features, the feature contribution is first analyzed, and the features used as the input of the randomly selected random forest algorithm are then tested for sitting and post-exercise. The performance evaluation tested by 6373 cardiac cycles from 15 volunteers indicates that the variation coefficient of feature contribution is less than 5%, and the identification accuracy is 99.21%.