The Research Of ECG Signal Compression Algorithm Based On Compressed Sensing

The widely use of ECG monitor produce a large amount of ECG data needed to be compressed. The emerging compressed sensing (CS) theory is a new kind of signal acquisition and compression paradigm. This new theory provides a new direction for the compression of ECG signal. This thesis studies the ECG signal compression algorithm based on compressed sensing.The sparsity of the signal is the fundamental to realize the compressed sensing. According to the characteristics of ECG signal, this thesis mainly studies the sparsity of the ECG signal and how to train its sparse dictionary. K-SVD dictionary training algorithm is the most widely used dictionary training algorithm. ECG’s sparse dictionary is trained using the K-SVD algorithm. Through the experiments of sparse representation error contrast in the same sparsity level and sparsity contrast in the same sparse representation error conditions, it is showed that the ECG signal has better sparsity under the trained dictionary than the fixed dictionary of DCT. Then through analyzing the deficiencies of the trained dictionary, a new dictionary training algorithm is proposed based on ECG signal feature waveform matching technology. The experiment of sparse representation error contrast in the same sparsity level shows that the dictionary trained by the proposed algorithm can sparse representation of ECG signal better and can meet the ECG’s quasi-periodic very well.Analyzing the compression experiments of the training dictionary base on the ECG signal feature waveform matching technology, the results show that the atoms of the dictionary trained by the ECG signal feature waveform matching technology are the typical template of ECG signal feature waveform, meeting the requirements of medical applications. At the same time, loss compression is converted to lossless compression. At last, all the ECG samples of MLII lead of MIT-BIH arrhythmia database are compressed using the proposed method, achieving the averaging compression ratio of 3.51.