In this paper, we propose a deep Recurrent Neural Networks (RNNs) based on Gated Recurrent Unit (GRU) in a bidirectional manner (BGRU) for human identification from electrocardiogram (ECG) based biometrics, a classification task which aims to identify a subject from a given time-series sequential data. Despite having a major issue in traditional RNN networks which they learn representations from previous time sequences, bidirectional is designed to learn the representations from future time steps which enables for better understanding of context, and eliminate ambiguity. Moreover, GRU cell in RNNs deploys an update gate and a reset gate in a hidden state layer which is computationally efficient than a usual LSTM network due to the reduction of gates. The experimental results suggest that our proposed BGRU model, the combination of RNN with GRU cell unit in bidirectional manner, achieved a high classification accuracy of 98.55%. Various neural network architectures with different parameters are also evaluated for different approaches, including one-dimensional Convolutional Neural Network (1D-CNN), and traditional RNNs with LSTM and GRU for non-fiducial approach. The proposed models were evaluated with two publicly available datasets: ECG-ID Database (ECGID) and MIT-BIH Arrhythmia Database (MITDB). This paper is expected to demonstrate the feasibility and effectiveness of applying various deep learning approaches to biometric identification and also evaluate the effect of network performance on classification accuracy according to the changes in percentage of training dataset.

This paper conducts a comparative analysis of deep models in biometrics using scalogram of electrocardiogram (ECG). A scalogram is the absolute value of the continuous wavelet transform coefficients of a signal. Since biometrics using ECG signals are sensitive to noise, studies have been conducted by transforming signals into a frequency domain that is efficient for analyzing noisy signals. By transforming the signal from the time domain to the frequency domain using the wavelet, the 1-D signal becomes a 2-D matrix, and it could be analyzed at multiresolution. However, this process makes signal analysis morphologically complex. This means that existing simple classifiers could perform poorly. We investigate the possibility of using the scalogram of ECG as input to deep convolutional neural networks of deep learning, which exhibit optimal performance for the classification of morphological imagery. When training data is small or hardware is insufficient for training, transfer learning can be used with pretrained deep models to reduce learning time, and classify it well enough. In this paper, AlexNet, GoogLeNet, and ResNet are considered as deep models of convolutional neural network. The experiments are performed on two databases for performance evaluation. Physikalisch-Technische Bundesanstalt (PTB)-ECG is a well-known database, while Chosun University (CU)-ECG is directly built for this study using the developed ECG sensor. The ResNet was 0.73%—0.27% higher than AlexNet or GoogLeNet on PTB-ECG—and the ResNet was 0.94%—0.12% higher than AlexNet or GoogLeNet on CU-ECG.

Recently, the wearable internet of things (WIoT) has brought a new dimension of connectivity to wireless body area networks (WBANs). For instance, the WIoT provides valuable functions, such as collecting, analyzing, and transmitting data for real-time health monitoring of medical services. However, designing a clustering and routing protocol is challenging because WIoT attributes are subject to high interference in dense deployment, social mobility, and limited energy. In a WIoT-enabled WBAN, cooperative control is the fundamental function of clustering and routing. Cluster-based routing protocols are preferred for WBAN because of their scalability. However, one-hop neighbor-based clustering approaches do not ensure connectivity and reachability owing to node mobility. The existing two-hop-based clustering protocols are mostly centralized-based approaches, which are unsuitable for dynamic WBANs. In this paper, we propose a distributed energy-efficient two-hop-based clustering and routing protocol (DECR) targeting WIoT-enabled WBAN. In DECR, in the cluster formation phase, each node obtains the information of its neighbor nodes within the two-hop range. We utilize the modified grey-wolf optimization algorithm for the cluster head (CH) selection and routing optimization. Node connectivity and residual energy were jointly considered when determining the CH in each cluster. We also developed an analytical model to determine the optimal number of clusters by considering intra- and inter-cluster transmission distances to reduce the overall transmission distance and number of transmissions. Finally, we proposed a routing algorithm to ensure energy-efficient packet delivery from CH to sink. Our simulation outcomes revealed that the proposed DECR significantly outperforms the existing clustering and routing protocols in various performance metrics.

The risk of tampering exists for conventional user recognition methods based on biometrics such as face and fingerprint. Recently, research on user recognition using biometric signals such as electrocardiogram (ECG), electroencephalogram (EEG), and electromyogram (EMG) has been actively performed to overcome this issue. We herein propose a user recognition method applying a deep learning technique based on ensemble networks after transforming ECG signals into two-dimensional (2D) images. A preprocessing process for one-dimensional ECG signals is performed to remove noise or distortion; subsequently, they are projected onto a 2D image space and transformed into image data. For the proposed algorithm, we designed deep learning-based ensemble networks to improve the degraded performance arising from overfitting in a single network. Our experimental results demonstrate that the proposed ensemble networks exhibit an accuracy that is 1.7% higher than that of the single network. In particular, the performance of the ensemble networks is up to 13% higher compared to the single network that degrades the recognition rate by displaying similar features between classes.

The postmobile era will go beyond using individual smart devices and allow for user interaction by connecting various devices with sensing capabilities, such as smartphones, wearable devices, automobiles, and the Internet of Things. Wearable devices can continuously collect a variety of information on the users and their environment as the devices are worn in daily life. Because of this, real-time big data analysis technology is needed. This paper proposes a deep learning-based ensemble network model for improving the performance and overcoming the problems, which can occur on a single network. This model is designed so that the features produced by n number of single networks are combined and relearned. In addition, different parameter values are used on each single network, and the data used in the experiments are generated by the fiducial point method, which uses feature point detection, and the nonfiducial point method for periods of 1 sec and n sec. In the experiment results, in the case of fiducial point-based ECG signals, the ensemble network recognition performance shows a maximum of 0.8% higher accuracy than that of the single network. In the case of a 1 sec period nonfiducial point-based ECG signal, the ensemble network recognition performance is a minimum of 0.4% and a maximum of 1% higher than that of the single network. In the case of an n sec period, there is a maximum difference of 1.3%, and the proposed ensemble network shows better performance than the single network.

Taking care of individual pigs is important in the management of a group-housed pig farm. However, this is nearly impossible in a large-scale pig farm owing to the shortage of farm workers. Therefore, we propose an automatic monitoring method in this study to solve the management problem of a large-scale pig farm. Particularly, we aim to detect undergrown pigs in group-housed pig rooms by using deep-learning-based computer vision techniques. Because the typical deep learning techniques require a large computational overhead (i.e., Mask-R-CNN), fast and accurate detection of undergrown pigs on an IoT-based embedded device is very challenging. We first obtain the video monitoring data of grouphoused pigs by using a top-view camera that is installed in the pig room, and then detect each moving pig by combining image processing and deep learning techniques. Gaussian Mixture Model is used to detect moving frames and moving objects. In embedded device implementations, by applying deep learning (i.e., TinyYOLO3) only to a few frames with a large number of pixel changes, embedded GPUs can be used efficiently, satisfying the real-time requirement. As a subsequent step, we check the acceptable conditions of the posture and separability from each video frame of the continuous video stream. Finally, to compute the relative size of each pig quickly and accurately, we develop image processing steps to complement the result of deep learning with minimum computational overhead. Furthermore, by pipelining the CPU and GPU steps of a continuous video stream, we can hide the additional image processing time. Based on the experimental results obtained from an embedded device, we confirm that the proposed method can automatically detect undergrown pigs in real-time, by working as an early warning system without any human inspection or measurement of actual weight by a farm worker.