Intelligent Recognition: Components of the Short-Time Fourier Transform Vs. Conventional Approaches

A new feature representation approach is generalized and used for Gaussian recognition. The generalized approach consists of using two new recognition features-the real and imaginary Fourier components-taking into account the covariance between features. The generalization approach improves the recognition effectiveness. An advanced time-frequency technique, the short-time Fourier transform, is considered. The covariance and the correlation coefficient between the proposed features are obtained for the first time for arbitrary stationary signals. The recognition effectiveness between the generalized approach and the Hartley, cosine and Power Spectral Density (PSD) approaches is compared. It is shown that the Hartley, cosine and PSD approaches are not optimal. Use of the generalized approach provides an essential increase in effectiveness in comparison with the Hartley, cosine and PSD approaches. Application of the generalized approach is considered for vibration diagnostics of object damping and fatigue. The application results agree with the theoretical results.