Data Analysis Using Scale-space Filtering and Bayesian Probabilistic Reasoning

Abstract This paper describes a program for the analysis of output curves from a differential thermal analyzer (DTA). The program first extracts probabilistic qualitative features from a DTA curve of a soil sample, and then uses Bayesian probabilistic reasoning to infer what minerals are present in the soil. It consists of a qualifier module and a classifier module. The qualifier employs a simple and efficient extension of scale-space filtering DTA data. Ordinarily when filtering operations are not highly accurate, points can vanish from contours in the scale-space image. To handle the problem of vanishing points, our algorithm uses perceptual organization heuristics to group the points into lines. It then groups these lines into contours by using additional heuristics. Probabilities are associated with these contours using domain-specific correlations. A Bayes tree classifier processes probabilistic features to infer the presence of different minerals in the soil. We show experimentally that using domain-specific correlations to infer qualitative features, this algorithm outperforms a domain-independent algorithm that does not.