Contribution aux techniques de débruitage basées EMD

Abstract

In, this thesis we have proposed new tools for denoising no-stationary signals based on the Empirical Modes Decomposition EMD and its variants. First, the proposed contributions were applied to synthetic data and then to real ECG signals. Similarly, they are compared to other original approaches that exist in the literature and the most popular of the state of the art. Then, we objectively evaluated the performance of our contributions by comparing the results to other classical approaches using the signal-to-noise ratio, SNR and the mean squared error MSE. Our work is presented in two parts which are: EMD-based denoising parametric methods, and EMD-based denoising non-parametric methods. In addition, We have proposed a new criterion for the reconstruction of the signal, this criterion is based on the determination of the index, which corresponds to the maximum value of the SNR after denoising. We have shown that our approaches are very effective for very noisy signals. Finally, we have proposed denoising methods based on the exploitation of the EEMD algorithm. We have also shown that EEMD-based denoising results exceed EMD-based denoising results. However, EEMD based denoising algorithms require a higher computation time than EMD based denoising algorithms. In this work, it has been demonstrated that the proposed algorithms give better results compared to those of the literature. Our contributions have been illustrated on real ECG signals, and can also be applied to other applications in various fields such as mechanical fault diagnosis or image processing