| ECG signal is a kind of important physiological signals from human body, it contains a lot of information on the state of the heart pathology and ECG signal reflects the heart and cardiovascular structure and its physiological and pathological information. The analysis of ECG signal is of great significance to the diagnosis of diseases of the cardiovascular system, its accuracy and reliability will directly affect the effects of clinical diagnostic and prognostic evaluation of heart patients. The traditional methods of ECG recognition are clinical auscultation by doctors, which obviously has certain subjectivity, instability and relatively poor accuracy. Normally, ECG signal acquired from human body is more or less influenced by power frequency interference, myoelectricity interference and baseline drift signal interference. Eliminating the noises of ECG signal effectively has an important significance to the detection of characteristic wave and pathological diagnosis.Non-negative Matrix Factorization(NMF), as a new feature separation method, is put forward by Lee and already under the background of the application of blind source separation in1999, and published in the journal Nature, and gradually developed into an effective method in signal processing and data analysis. NMF makes that different results have been obtained completely by adding non-negative matrix elements in the process of matrix decomposing, which realizes the nonlinear dimensionality. With the deepening of the research on blind source separation (BSS), the NMF has gradually become a data processing tool in the field of signal processing, biomedical engineering and image engineering research, which is favored by most scholars. This article proposes that non-negative matrix decomposition is applied to denoising of ECG signal, with a fast convergence rate and characters of sparse, non-negative and dimension reduction.In a study of basic NMF algorithm, NMF is one of the nonnegative constraints. In this way, by the decomposed signal data base, and is used to reconstruct the weight coefficient are negative. In this mode, only allow linear superposition calculation, this guarantee form whole "local" mode. Therefore, NMF is regarded as a method of partial feature extraction. However, NMF algorithm to get the "part" of sometimes is not as we expected, localization, and basic NMF method in some cases the recognition rate is not desirable.With the inspiration of the NMF original algorithm, local signal data to establish himself in research PNMF algorithm, its purpose is to limit access code vector by introducing a sparse sex local decomposition (matrix H) of real objects, and make basic component local sparse matrix (W), strengthen localization characteristics of base composition, make the algorithm is suitable for application of local feature is very important.In this paper, combining with the characteristics of the NMF algorithm and the characteristics of ECG signal for the first time put forward a new algorithm of NMF, PNMF. Combining with the MIT/BIH ECG in international standard database data and simulation of baseline drift, power frequency interference and myoelectricity interference noise synthesis including noise of ECG signal, and apply the proposed PNMF algorithm of blind source separation experiments, using the result of the separation of SNR (signal to noise thewire, SNR) evaluation parameters for quantitative evaluation, and compares the three different NMF algorithm, experimental results show that PNMF algorithm can effectively separated the ECG signal, this may provides certain reference basis for late accurate diagnosis. |
PDF Full Text Request