Search

Recently, the interest in using wearable devices or the internet of things (IoT)-based biometric authentication, especially IoT-based electrocardiogram (ECG) has increased. ECG-based biometric authentication has received great attention as a next-generation promising technique and been implemented with various approaches to improve the authentication performance for the past few decades. However, ECG signals of a person may vary according to his/her physical states, or health conditions, possibly leading to authentication failure in some cases. Therefore, it is essential to design a robust method that handles the ECG subject variability for accurate authentication. In this Chapter, we proposed an efficient and robust authentication system based on ECG. In this study, we propose a novel deep learning fusion framework using the transfer learning concept where the deep features extracted from different models are combined into a single feature which are then fed to a custom classifier such as a support vector machine (SVM) for authentication. Cross-validation studies are used to assess the performance of the proposed authentication system using two public databases. Evaluation results show that the performance of our fusion model achieved an authentication accuracy of 99.4% with a high level of precision and recall. Finally, the results show that the proposed system is suitable for real-time applications.

Such a hot open issue in the area of computer vision is the classification of visual images especially in Internet of Things (IoT) and remote mid-band and high-band based connections. In this paper, we propose a robust and efficient taxonomy framework. The proposed model utilizes the well-known convolutional neural network composites to construct a robust Visual Image Classification Network (VICNet). The VICNet consists of three convolutional layers, four Relu/Leaky Relu activation layers, three max-pooling layers and only two fully connected layers for extracting expected input image features. To make the training process faster, we used non-saturating neurons with a very efficient Graphics Processing Unit (GPU) implementation for the convolution operation. To minimize over-fitting issue in the fully-connected layers, we utilized a recently-developed regularization approach “dropout” with a dropping probability of 50%. The proposed VICNet framework has a high potential capability in the recognition of test images. The experimental and simulations results proven the efficacy of the proposed model.