Linguistic-Fuzzy Classifier for Discrimination and Confidence Value Estimation

摘要

This report describes a new method for assigning an event to a particular class. An event is described by some attributes (e.g., size, shape, and intensity and their changes). These attributes have a distribution. Fuzzy membership functions provide a means for quantifying the importance of an attribute based on its value and distribution. With proper selection of attributes, we can calculate the probability that an event belongs to a particular class by selecting appropriate membership functions. We applied this to visible and IR camera data generated to support the DSI Program. The goal of the program is to investigate the possibility of using disparate sensors to serve as a chemical and biological early warning system and integrate them into the CB command and control network. Detecting when CB munitions are deployed requires developing algorithms that differentiate between the detonation of conventional and CB munitions. This report describes how we applied this new classification method to video signals generated from the visible cameras used during DSI field test. The report provides examples of how to use this method to estimate class confidence and will also show how the confidence values were used to discriminate between CB and conventional munitions detonations.