| In recent years,mobile devices have become widely popular,and they are playing an increasingly crucial role in people’s lives.At the same time,due to the growing severity of users’ concerns about mobile devices,the leakage of large amounts of personal information stored in mobile devices could lead to many irreparable losses.Identity authentication systems are the first line of defense in protecting users’ personal privacy.Traditional identity authentication systems include fingerprint recognition,facial recognition,and voice recognition,but these mature technologies have been seriously threatened in recent years.It is an urgent problem to find a solution that can both ensure that users’ identity features are not easily stolen and not increase sensor costs.This dissertation designs an identity authentication system based on human vibration features,which proposes two algorithms: a static identity authentication system based on low sampling rate and a dynamic identity authentication algorithm based on Kalman filtering.The former provides a simple and effective identity authentication scheme for low sampling rate static systems,while the latter provides high accuracy and robustness for dynamic identity authentication scenarios.In the static identity authentication algorithm based on low sampling rate,this dissertation addresses the problem of low sampling rate using a oversampling algorithm.The data is preprocessed through methods such as regularization,segmentation,and filtering to segment it into individual sample segments.MFCC algorithm,PCA,and FA algorithms are used for feature extraction of the vibration signals,and the resulting feature matrix is input into a GBDT classifier for training.Compared with traditional vibration-based identity authentication algorithms,this dissertation proposes a method that combines MFCC features with statistical features,effectively achieving false rejection rate and false acceptance rate of 13.3% and 8.5%,respectively.The influence of the classifier,sample length,and grip method on the identity authentication performance is discussed in this section’s experiments.In addition,for the possible arm motion during the user authentication process,this dissertation proposes a dynamic identity authentication algorithm based on Kalman filtering.The data is processed by oversampling to improve the sampling rate,and then normalized,framed and filtered based on the Kalman filtering algorithm to remove the influence of arm motion on the accelerometer.The framed data is then subjected to MFCC feature extraction and time-domain feature extraction using a Mel filter bank and LSTM network,respectively,and the resulting feature matrix is input into a CNN for training.Compared with traditional vibration-based identity authentication algorithms,this algorithm can complete authentication in dynamic scenarios by introducing Kalman filtering algorithm,LSTM network,and CNN,with false rejection rate and false acceptance rate of only 11% and 4.5%,respectively.In this section’s experiments,this algorithm maintains an accuracy rate of over 87.4% even in the face of ”sweaty palms interference”,and both Android and i OS phones have good migration characteristics. |
