An interactive feature extraction program

Automatic signal analysis procedures generally involve extracting a set of features from a signal and making decisions based on those features. For many aperiodic and otherwise difficult signals, the choice of the optimum feature set and related parameters is an arduous task. This paper describes an interactive program used for experimenting with a feature set based on modified Haar functions. Differences between the Haar features and the Fast Fourier transform coefficients are discussed. Three examples are presented concerning the use of the interactive program on different signals; a test signal, a textile yarn-mass measurement signal, and electroencephalograms (EEGs). Finally, the decision process for an Automatic Sleep Scoring Program is discussed. The scoring accuracy of 88% for this computer system compares favorably to the accuracies between human scorers.