Automatic classification of time-frequency plots applied to the center-of-pressure rotational components

Time-frequency plots are widely applied to the non-stationary analysis of signals. These plots may be difficult to interpret, particularly when large data sets have to be considered. The aim of this work is to propose an automatic procedure of feature selection and clustering to be applied to time-frequency plots. We focus on the application of this procedure to plots obtained from a non-stationary analysis of the center-of-pressure signals acquired in upright bipedal stance. From a data set of 168 time-frequency plots we obtained 5 different clusters, each characterized by a few distinctive features. We were able to interpret the results of the clustering relating them to the physiological mechanisms underlying postural sway.