Canine Biometric Identification Using ECG Signals and CNN-LSTM Neural Networks

As global pet acceptance increases, the market size for pet ownership grows. Consequently, registering pets is becoming increasingly crucial, with some nations mandating it by law. Animal biometrics is a subject of ongoing research, spanning inscriptions, iris recognition, and facial recognition, with a growing number of companies partaking. However, biometric methods mostly rely on image recognition, which can result in degraded performance depending on the captured angle and external environment. To address this issue, we conducted a study to design and evaluate the performance of a deep learning-based dog identity recognition system that utilizes electrocardiogram (ECG) that is harder to forge than existing methods and does not require additional image processing. To evaluate performance, we utilized two dog ECG databases and conducted biometric recognition experiments with data collected from differing measurement environments from these integrated databases. Input signals for recognition were generated through both R-peak based and blind signal segmentation methods. For the purpose of dog identification, we developed and employed a 1D CNN-LSTM model as a classifier. Additionally, three DNN-based classifiers were developed to compare their performance with that of the proposed model. To evaluate performance, the confusion matrix was used in conjunction with metrics such as accuracy, equal error rate (EER), receiver operating characteristic (ROC) curve, and precision recall (PR) curve. The proposed model demonstrated up to 98.7% accuracy in the biometrics of a separate database of 16 subjects, and as high as 96.3% accuracy in the biometrics of an integrated dataset of 33 subjects. The suggested approach exhibited a 93.1% accuracy rate when employing the blind segmentation method, eliminating the need for supplementary signal processing to derive input signals.