Classification invariante en translation de signaux non stationnaires
This is the latest version of this eprint.
This paper addresses the question of the classification of on-stationary signals. We make the hypothesis that each signal includes a pattern, the shape of which is discriminant but the time location of which is random and unknown. We propose to build a graphical representation, based on continuous wavelet transform in order to eliminate the absolute time reference of the description. Then, we define a dot product between graphs and compare the results of SVM and k-nn.
Available Versions of this Item