Automatic infrasound signal detection using the Hough transform

Abstract

The Hough transform is a mathematical device that allows the retrieval of parametric curve information from binary-pixelated data in the presence of noise. This slope-intercept transform maps each point in the image space S into a straight line in parameter space P and has the very useful property that all points in S that lie along the same straight-line map to the same number of straight lines in P with a common intersection point. Thus with a suitable counting procedure, the problem of extended straight-line detection in noisy pixelated data becomes one of local peak finding, a problem that may be substantially more tractable. In this study, an algorithm that utilizes the Hough transform for the detection of signals in International Monitoring System style infrasonic array data by seeking periods of constant backazimuth that are associated with coherent acoustic signals is described. A system of synthetic signal implants is used to assess the performance of the detection algorithm by generating a set of pseudo Receiver Operator Characteristic curves. A feature of the detection algorithm is the ability to accommodate full three-dimensional array geometry.