Human Identification System Research Using Electrocardiograms Based On Neural Network

The current era of rapid development of technology, the security of information to get more attention. Common identification technology exhibit their drawbacks, has been unable to meet the demand for security, with biometric technology is the product of this demand under this technique uses the body's own inherent characteristics to personal identity authentication, has not lost a unique, easy to collect and many other advantages. Currently, fingerprint recognition, voice recognition and face recognition technology has been applied. However, they all have their flaws, fingerprints easy to stay attached to the surface; the voice can imitate, face from the photo to get more. ECG signal is important to the body signals, often as a clinical diagnosis is based, can be used as an effective tool for authentication. ECG signal is the body's source signal, due to differences in the human body, it has unique characteristics, and not easy to imitate.ECG as identification technology is a relatively new identification technology, this paper has been on the reference information on ECG identification technology to achieve the extraction of ECG feature points, the feature weights and completed the most accurate analysis Excellent feature subset selection using GA to optimize the RBF network and experiments designed to detect the network performance. Mainly to do the following work:(1)Interference with the characteristics filtered ECG extraction. Initial ECG signal wavelet decomposition, abandoned 8-scale decomposition coefficient of eliminating low-frequency interference, and then reconstructed signal; set threshold after wavelet decomposition, elimination of high-frequency interference; atrous algorithm for ECG signal in accordance with decomposition. Setting an appropriate time window, extract the maximum value within this time window, the maximum value of the R-wave peak that is needed, and then determine the next R-wave peak to R-wave peak point as the reference point, set the search window before and after the R-wave peak position of each extract obtained 50 ms minimum value can be obtained at Q-wave and S-wave position; the start and end points can be used to extract the smallest error linear fitting method.(2) to determine the optimal feature subset. Describes the linear discriminant analysis method, analyze the ECG feature weights, constructed in accordance with various features ofthe orderly queues classifying contribution, based on the identification rate of the network to choose the best subset.(3) optimizing neural network classifier. The front extract a subset of RBF neural network as input. Using GA algorithm to optimize and compare the recognition rate and the RBF neural network classifier to determine the GA optimization neural network classifier.