Non-stationary signal feature characterization using adaptive dictionaries and non-negative matrix factorization

Our objective is to introduce a novel method of performing non-stationary signal analysis by means of enhanced training performance, suitable learning, and proper training dictionary elements in sparse representation techniques for real biomedical signal applications. Non-stationary signal characteristics pose severe challenges in terms of analysis and extraction of discriminant features. In addition, due to complexity of biomedical signals, the need for feature extraction algorithms that can localize to events of interest increases. To fulfil this objectives, we propose to use dictionary learning algorithms based on non-negative matrix factorization. This allows us to train the dictionary elements that lead to more robust classification performances. The proposed algorithm uses a time-frequency decomposition based on wavelet transform for non-stationary 1D biomedical signals. In this manuscript we aim to exploit non-stationary signal analysis through dictionary learning and study the discriminant features of these signals by means of sparse representation to design a robust algorithm in addition to higher classification performance.