Improvement of ECG based Personal Identification Performance in Different Bathtub Water Temperature by CNN

This paper aims at exploring the variety of Electrocardiogram(ECG) interval and amplitude during different bathtub water temperature and eliminating their influence on personal identification with ECG. There are 10 subjects in the experiment, each subject collects 2 ECG recordings, each recording is at least 220 s. One recording is collected at 38±0.5 °C bathtub water temperature and the other recording is collected at 42±0.5 °C bathtub water temperature. All the raw ECG are removed baseline drift and normalized, then the R peaks are detected and all the R-R interval(RRI) and amplitude are calculated. Through statistical analysis method, we find that the median of RRI in low bathtub water temperature is bigger than in high bathtub water temperature for all subjects, and compared with low bathtub water temperature, the variety of R peaks amplitude has 3 situations in high bathtub water temperature: increase, decrease and unchanged. Then all the QRS complex are segmented and are taken as training data and test data. During the training stage, there are 3340 training datasets, 1670 training datasets are from low bathing water temperature and the other 1670 training datasets are from high bathing water temperature. In the testing stage, first we use 410 testing data which are from low bathtub water temperature to test the trained model, the best and robust identification rate is 87.07%, when we use the other 410 testing data which are from high bathtub water temperature to test the trained model, the best and robust identification rate is 87.32%. To the best of our knowledge, this is the first time to explore the variety of ECG interval and amplitude during different bathing water temperature. However, further improvements are still needed during different bathing environment.