14-09-2017, 12:34 PM
The new algorithm has the property that the most appropriate wavelet filter is selected from a library of wavelet filters according to a proposed switching criterion. This algorithm produces a system that can adapt to the time variation of a signal. The results of the simulation are demonstrated for several audio signals to show the superior performance of the new method.
When wavelet non-linear functions are used in the neurons of a neural network we can estimate the wavelet-scale and translation parameters and the relative importance (weight) of each optimal base function with respect to the approximation of a given function (signal) in the minimum mean square error direction or with respect to the classification of a function (signal) with a minimum average classification error. This estimation can be considered as adaptive (dependent or signal dependent) sampling. In this article we provide theoretical evidence to demonstrate that such an adaptive sampling scheme constitutes a framework that implies that the neural network based estimation technique allows us to reconstruct the input signal of the adaptive wavelets so that the reconstruction is numerically stable. We apply the proposed representation and classification architectures for coding and classification of biological signals such as the electrocardiogram (ECG). Experimental details of the ECG encoder and ECG waveform classification such as P, QRS and T are provided. The experimental results presented in this paper are obtained by applying the proposed methodologies to the ECG (American Heart Association) standard database. The results indicate that a compression ratio of approximately two times better than current techniques can be obtained and an average classification accuracy of 94% can be obtained for the abnormal classification of P, QRS and T.
When wavelet non-linear functions are used in the neurons of a neural network we can estimate the wavelet-scale and translation parameters and the relative importance (weight) of each optimal base function with respect to the approximation of a given function (signal) in the minimum mean square error direction or with respect to the classification of a function (signal) with a minimum average classification error. This estimation can be considered as adaptive (dependent or signal dependent) sampling. In this article we provide theoretical evidence to demonstrate that such an adaptive sampling scheme constitutes a framework that implies that the neural network based estimation technique allows us to reconstruct the input signal of the adaptive wavelets so that the reconstruction is numerically stable. We apply the proposed representation and classification architectures for coding and classification of biological signals such as the electrocardiogram (ECG). Experimental details of the ECG encoder and ECG waveform classification such as P, QRS and T are provided. The experimental results presented in this paper are obtained by applying the proposed methodologies to the ECG (American Heart Association) standard database. The results indicate that a compression ratio of approximately two times better than current techniques can be obtained and an average classification accuracy of 94% can be obtained for the abnormal classification of P, QRS and T.