Classification of Abnormal and Normal ECG beat Based on Deep Learning Techniques

Electrocardiogram signals are classified as abnormal or normal ECG signal. Both the MIT-BIH Arrhythmia (BIHA) Database and the MIT-BIH Noise Stress Test Database (NSTDB) are used for the model's training and testing phases. Experimental For classification of arrhythmia from ECG, a variety of tools and techniques are available. The most important ones are methods like time series analysis, spectrogram analysis, scalogram analysis etc. Every technology has their own advantages and limitations that must be considered when using it in ECG signa. This study's main goal is to create a computational model that, given an electrocardiogram (ECG) signal can reliably categorise bits into normal and abnormal class. In this study, the proposed method is implemented in order to categorise an electrocardiogram (ECG) signal that has a noise levels between -6 to 24 decibel(dB) by the use of superlet transform, VGG18, and K-Nearest Neighbour (KNN) classifier. In this regard, the input signal is first pre-processed by the application of the superlet transform (SLT), and then it is turned into a 2-D temporal frequency spectrogram. Because a bandpass filter is already utilised internally by superlet, there is no need to make use of any additional filters in order to exclude the signal with a low SNR. Following that, the 2-D temporal frequency spectrogram is fed into the transfer learning model (VGG16) to extract the features. The features that were gathered were subjected to classification via the K-Nearest Neighbour (KNN) classifier. The results of the classification indicate that the accuracy of the proposed model in a noisy environment is 99.46%.