People counting is one of the hottest issues in sensing applications. Impulse radio ultrawideband radar has been extensively adopted to count people because it provides a device-free solution without illumination and privacy concerns. However, current solutions have limited performances in congested environments due to signal superpositions and obstructions. In this letter, a hybrid feature extraction method based on the curvelet transform and the distance bin is proposed. First, 2-D radar matrix features are extracted at multiple scales and multiple angles by applying the curvelet transform. Then, the distance bin concept is introduced by dividing each row of the matrix into several bins along the propagating distance to select features. A radar signal data set is constructed for three density scenarios, including people randomly walking in a constrained area at densities of three and four persons per square meter and people in a queue with an average between-person distance of 10 cm. The number of people in the data set scenarios varies from 0 to 20. Four classifiers--a decision tree, an AdaBoost classifier, a random forest, and a neural network--are compared to validate the hybrid features. The random forest achieves the highest accuracy of above 97% in the three density scenarios. To further investigate the reliability of the hybrid features, they are compared with three other features: cluster features, activity features, and features extracted by a convolutional neural network. The comparison results reveal that the proposed hybrid features are stable, and their performance is substantially more effective than that of the others.

In the demand for protecting the increasing aged groups from heart attacks, the improvement of the mobile electrocardiogram (ECG) monitoring systems becomes significant. The limitations of the arrhythmia classification in these systems are the lack of ability to cope with motion state and the low accuracy in new users' data. This paper proposes a system which applies the impulse radio ultra wideband radar data as additional information to assist the arrhythmia classification of ECG recordings in the slight motion state. Besides, this proposed system employs a cascade convolutional neural network to achieve an integrated analysis of ECG recordings and radar data. The experiments are implemented in the Caffe platform and the result reaches an accuracy of 88.89% in the slight motion state. It turns out that this proposed system keeps a stable accuracy of classification for normal and abnormal heartbeats in the slight motion state.

Further applications of impulse radio ultra-wideband radar in mobile health are hindered by the difficulty in extracting such vital signals as heartbeats from moving targets. Although the empirical mode decomposition based method is applied in recovering waveforms of heartbeats and estimating heart rates, the instantaneous heart rate is not achievable. This paper proposes a Heartbeat Estimation And Recovery (HEAR) approach to expand the application to mobile scenarios and extract instantaneous heartbeats. Firstly, the HEAR approach acquires vital signals by mapping maximum echo amplitudes to the fast time delay and compensating large body movements. Secondly, HEAR adopts the variational nonlinear chirp mode decomposition in extracting instantaneous frequencies of heartbeats. Thirdly, HEAR extends the clutter removal method based on the wavelet decomposition with a two-parameter exponential threshold. Compared to heart rates simultaneously collected by electrocardiograms (ECG), HEAR achieves a minimum error rate 4.6% in moving state and 2.25% in resting state. The Bland–Altman analysis verifies the consistency of beat-to-beat intervals in ECG and extracted heartbeat signals with the mean deviation smaller than 0.1 s. It indicates that HEAR is practical in offering clinical diagnoses such as the heart rate variability analysis in mobile monitoring.

To enhance electrocardiogram (ECG) monitoring systems in personalized detections, deep neural networks (DNNs) are applied to overcome individual differences by periodical retraining. As introduced previously [4], DNNs relieve individual differences by fusing ECG with impulse radio ultra-wide band (IR-UWB) radar. However, such DNN-based ECG monitoring system tends to overfit into personal small datasets and is difficult to generalize to newly collected unlabeled data. This paper proposes a self-adjustable domain adaptation (SADA) strategy to prevent from overfitting and exploit unlabeled data. Firstly, this paper enlarges the database of ECG and radar data with actual records acquired from 28 testers and expanded by the data augmentation. Secondly, to utilize unlabeled data, SADA combines self organizing maps with the transfer learning in predicting labels. Thirdly, SADA integrates the one-class classification with domain adaptation algorithms to reduce overfitting. Based on our enlarged database and standard databases, a large dataset of 73200 records and a small one of 1849 records are built up to verify our proposal. Results show SADA's effectiveness in predicting labels and increments in the sensitivity of DNNs by 14.4% compared with existing domain adaptation algorithms.

Pruning is effective to reduce neural networks’ parameters and accelerate inferences, facilitating deep learning in resource-limited scenarios. This paper proposes an asynchronous pruning method for multi-branch networks on the basis of our previous work on channel coresets constructions, to achieve module-level pruning. Firstly, this paper accelerates coreset based pruning by batch sampling with a sampling probability decided on our-designed importance function. Secondly, this paper gives asynchronous pruning solutions with an in-place distillation of feature maps for deployment on multi-branch networks such as ResNet and SqueezeNet. Thirdly, this paper provides an extension to neuron pruning by grouping weights as channels. During tests on sensitivity of different layers to channel pruning, our method outperforms comparison schemes on object detection networks, indicating advantages of data-independent channel selections in maintaining precision. As shown in tests of asynchronous pruning solutions on multi-branch classification networks, our method further decreases FLOPs with a small accuracy decline on ResNet and acquires a small accuracy increment on SqueezeNet. In tests on neuron pruning, our method achieves an accuracy comparable to existing coreset based pruning methods by two solutions of precision recovery.