| With its unique advantages,millimeter wave radar has been widely used in the fields of vital sign detection and identity authentication.However,current mmwave radar based vital sign detection algorithms are not efficient enough in coping with the effects of various types of noise and harmonics.At the same time,it is difficult to guarantee the high reliability of algorithm prediction information in some complex scenarios such as long-distance and low signal-to-noise ratio.In addition,the current algorithms that use vital sign signals for identity authentication have difficulty in ensuring the robustness of the algorithms when dealing with low quality of input data,similar features,low signalto-noise ratio,and limited radar resolution.In this thesis,millimeter wave radar based vital sign detection and identity authentication are investigated and the following work is done.(1)To address challenge that the current vital signs detection algorithm is difficult to balance noise elimination and harmonic suppression,this thesis proposes an improved vital sign detection algorithm for radar signals.In the personnel localization phase,the periodicity of the vital signs signal and the non-periodicity of the noise are analyzed,and the average cancellation method is designed to eliminate the effects of multipath effects in the environment,static object clutter and the DC components generated after Fourier transform;In the phase extraction stage,phase unwrapping,wavelet transform and interpolation smoothing are used to extract high-quality phase signals;In the signal separation stage,the combination of ensemble empirical modal decomposition and improved independent component analysis ensures the noise immunity performance while suppressing the masking of the heartbeat frequency by the respiratory harmonic frequency,and obtains a more accurate vital sign signal.The experimental results show that the algorithm designed in this thesis has a higher accuracy of vital sign extraction compared with some commonly used vital sign detection algorithms.(2)To address the situation that the accuracy of single feature input for authentication is not high,this thesis proposes an authentication algorithm based on multi-feature fusion.The algorithm first extracts the respiration waveform,heartbeat waveform and wavelet spectrum features from the radar echo signal,and inputs these features into different convolutional layers of the convolutional neural network.After convolutional feature extraction,the features extracted from the respiration waveform,heartbeat waveform and wavelet spectrum are fused;Then,the network is trained and the best classification model is saved to classify the test set to verify the recognition accuracy of the algorithm;Finally,open-set recognition experiments are designed to verify the generalization ability of the algorithm when encountering unknown data.The experimental results show that the multi-feature fusion algorithm proposed in this thesis perform well in identity authentication,and the algorithm proposed in this thesis can maintain a high recognition accuracy for known users in different number of people scenarios compared with existing algorithms,and maintain its reliability when encountering unknown illegal user attacks. |
