ECG biometric using 2D Deep Convolutional Neural Network
ECG biometric using 2D Deep Convolutional Neural Network
复制标题
DOI：
10.1109/icce50685.2021.9427616
复制
发表时间：
2021-01
期刊：
影响因子：
--
通讯作者：
Siddartha Thentu;Renato Cordeiro;Younghee Park;Nima Karimian
中科院分区：
文献类型：
--
作者：
Siddartha Thentu;Renato Cordeiro;Younghee Park;Nima Karimian
我们提出了一种新颖的多尺度连续小波变换特征方法，以准确获得微纹理和多尺度心电图特征，并演示它如何从最先进的深度卷积神经网络技术中受益。换句话说，我们使用流行的 CNN 架构（例如 InceptionV3、VGG16、VGG19、Inception ResNetV2、MobileNetV2 和 Xception）进行迁移学习，这些架构已在 ImageNet 上进行了训练。我们提出的心电图生物识别框架在 CEBDB 上实现了 99.96% 的平均识别率，在 290 名受试者的 PTB 数据集上实现了 99.47% 的平均识别率。我们还使用其他两个具有不同行为的公共心电图数据集来评估所提出算法的有效性。
We propose a novel multi-scale continuous wavelet transform feature method to accurately obtain micro-texture and multi-scale ECG characteristics and demonstrate how it could benefit from the state-of-the-art deep convolutional neural network techniques. In other words, we performed transfer learning with popular CNN architectures such as InceptionV3, VGG16, VGG19, Inception ResNetV2, MobileNetV2, and Xception which have been trained on the ImageNet. Our proposed ECG biometric framework achieves an average identification rate of 99.96% on CEBDB, 99.47% on PTB dataset with 290 subjects. We also evaluate the effectiveness of the proposed algorithm with the other two public ECG datasets with diverse behaviors.