The Impact of Bathtub Water Temperature on Personal Identification with ECG Signal based on Convolutional Neural Network

This paper explores the impact of bathtub water temperature on personal identification with ECG signal using convolutional neural network. Ten volunteers' ECG records are collected with lead-III at low temperature of bathtub water $(38\pm0.2\ {{}^{\circ} \mathrm{C}}$ in average) and high temperature of bathtub water $(42\pm 0.5 {{}^{\circ}\mathrm{C}}$ in average) environments, respectively. Each record is about $5\pm 1$ minutes at low temperature and $4\pm1$ minutes at high temperature. In the data preprocessing stage, we denoise the original ECG signal and segment the QRS complex on a beat-by-beat basis. Then, we perform two interpolation calculations based on the QRS segmentation and convert the QRS complex into a binary image one by one. When we use the ECG signals which are collected at low temperature to train and test the CNN model, the identification rate is 82.67%. However, if we use the ECG signal collected at high temperature to test this trained CNN model, the identification rate is only 13.33%. Conversely, when we use the ECG signal collected at high temperature to train and test the CNN model, the identification rate is 85.50%. However, if we use the ECG signal collected at low temperature to test this trained CNN model, the identification rate is only 12.17%. Thus, we notice that the different bathtub water temperature has an important impact on the ECG signal patterns and it is feasible to perform personal identification by convolutional neural network with ECG signal collected during bathing at the same temperature.