An Approach for Feature Interpolation Between Classes to Simplify Calibration of Sensor Systems for Odour Monitoring

Abstract

An approach for the augmentation of n-dimensional data was implemented and tested using artificial and real gas sensor data. With this method, a new (super)class is generated from two or more smaller classes by interpolating between these initial classes. It can be useful, for example, to assign a larger part of the feature space to a specific group of substances when performing classification tasks. Interpolation between two normally distributed classes generates very good results. For the interpolation between more than two classes or between inhomogeneous classes, several approaches were tested.