Pattern recognition of non-stationary time series with finite length

Statistical learning and recognition methods were used to extract the characteristics of size series measurements of cocoon filaments that are non-stationary in terms of mean and auto-covariance, by using the time varying parameter auto-regressive (TVPAR) model. After the system was taught to recognize the size data, the system correctly recognized the size of series of cocoon filaments as much as 96.95% of the time for a single series and 98.72% of the time for the mean of two series. The correct recognition rate was higher after suitable filtering. The theory and method can be used to analyze other types of non-stationary finite length time series.