Statistical Model for the Classification of the Wavelet Transforms of T-ray Pulses

This study applies auto regressive (AR) and auto regressive moving average (ARMA) modeling to wavelet decomposed terahertz pulsed signals to assist biomedical diagnosis and mail/packaging inspection. T-ray classification systems supply a wealth of information about test samples to make possible the discrimination of heterogeneous layers within an object. In this paper, the classification of normal human bone (NHB) osteoblasts against human osteosarcoma (HOS) cells and the identification of seven different powder samples are demonstrated. A correlation method and an improved Prony's method are investigated in the calculation of the AR and ARMA model parameters. These parameters are obtained for models from second to eighth orders and are subsequently used as feature vectors for classification. For pre-processing, wavelet de-noising methods including the SURE (Stein's unbiased estimate of risk) and heuristic SURE soft threshold shrinkage algorithms are employed to de-noise the normalized T-ray pulsed signals. A Mahalanobis distance classifier is used to perform the final classification. The error prediction covariance of AR/ARMA modeling and the classification accuracy are calculated and used as metrics for comparison