ARMA Model Order Determination Using Edge Detection: A New Perspective

Abstract

An edge detection-based approach to estimate the order of an autoregressive moving average (ARMA) model process is presented. The proposed method performs edge detection to select the ARMA order by extracting the outlines of a data covariance matrix derived from the observed data sequence. The method is based on the minimum eigenvalue (MEV) criterion developed by Liang et al., IEEE Trans. Signal Process., 41(10): 3003-3009, 1993. The algorithm transforms the MEV covariance matrix into an image by normalizing and resizing the original covariance matrix. Then, a search is performed to locate changes in the intensity function, i.e., pixels where the brightness changes abruptly. Examples are presented to demonstrate the performance of this algorithm.