Eeg Applied Research In Personal Identification And Fatigue Detection

The application of biometric-based personal identification is more and more widely used in our daily life. Compared with the common used biometric modalities such as fingerprint and face, electroencephalography (EEG) has characteristics of anti-counterfeit and anti-stress, which are more important practical significance in the study of the personal identification applications. Feature extraction is a key step in EEG-based personal identification. Previous feature extraction methods are mostly based on the assumption that EEG is a stationary signal, however, this assumption is not suitable in practice, and EEG is a typical non-stable, non-linear signal. Based on this, a feature extraction method is proposed based on wavelet and wavelet packet transform, to obtain a higher recognition rate in EEG-based identification systems, and to realize a self-adaptive feature extraction.Driver fatigue is one of the main causes of traffic accidents and many scholars have carried out researches on driver fatigue detection. There are mainly three detection methods, respectively, based on driving behavior, based on computer vision and based on physiological signals. In all of them, EEG-based fatigue detection is considered the most accurate and objective. Using a portable EEG acquisition instrument to collect EEG when individual was in awake or fatigue state, this paper studied the application of single-electrode EEG in fatigue detection. Fatigue detection method based on the Mahalanobis distance shows the individual EEG logarithmic power spectrum changes in fatigue and awakes state; and the other detection method based on the EEG spectrum distinction can effectively detect the state of fatigue, pointing out the necessity for individual fatigue detection.The content of this thesis is divided into two parts:EEG-based personal identification using wavelet and wavelet packet and EEG-based fatigue detection.The main tasks are as follows:1. Compared with commonly used feature extraction and combining with time-domain and frequency-domain information, three feature extraction methods were proposed, respectively, including wavelet transform, wavelet packet transform and the best wavelet packet basis.2. In the study of feature extraction based on wavelet packet transform, the band range and feature combinations were adopted to obtain a high recognition rate. However, this method has poor adaptability to the different sets of data. An adaptive feature extraction method was proposed and obtained high recognition rate, which is based on best wavelet packet basis and use LDB algorithm as band distinguishing measure.3. EEG-based fatigue detection experiment was introduced and the subjective criteria of the degree of fatigue were proposed. In the case of not removing the noise, this paper realized two fatigue detections, including fatigue detection based on Mahalanobis distance and fatigue detection method based on spectrum distinguish.Fatigue detection methods based on the Mahalanobis distance is an unsupervised detection method, without considering impact on the test results due to EEG changes with factors such as diet and circadian. The fatigue detection method based on spectrum using Welch algorithm to estimate EEG power spectrum, indicated the necessity for analysis of the individual and changed fatigue detection problem into a two-class problem. Experimental results show the effectiveness of the two methods.