Research on Classification of Electrocardiogram Signals Based on Improved Frequency Sliced Wavelet Transform

Aiming at the problems such as poor adaptability of basis function, high noise sensitivity and feature redundancy in ECG classification, this paper proposes an adaptive feature extraction framework based on improved Frequency Slice Wavelet Transform (FSWT). It includes dynamic band adjustment model and morphological adaptive kernel function, which can be used to preprocess electrocardiograph (ECG) image data to make data have time-frequency characteristics and remove noise at the same time. The experiment was based on the MIT-BIH database for data preprocessing, and the model combined with convolutional neural network and recurrent neural network was trained. The traditional image data preprocessing methods such as image translation, rotation and cropping were compared with the methods adopted in this experiment. Finally, it is concluded that the improved frequency slice wavelet transform can improve the accuracy and anti-interference ability of the recognition of ECG signals based on convolutional neural network, and improve its robustness, which is suitable for mobile wearable ECG detection and other medical applications.