ABSTRACT
This paper proposes a novel time-frequency feature fusion method to recognise patients' behaviours based on the Frequency Modulated Continuous Wave (FMCW) radar system, which can locate patients as well as recognise their current actions and thus is expected to solve the shortage of medical staff caused by the novel coronavirus pneumonia (COVID-19). To recognise the patient's behaviour, the FMCW radar is utilised to acquire point clouds reflected by the human body, and the micro-Doppler spectrogram is generated by human motion. Then features are extracted and fused from the time-domain information of point clouds and the frequency-domain information of the micro-Doppler spectrogram respectively. According to the fused features, the patient's behaviour is recognised by a Bayesian optimisation random forest algorithm, where the role of Bayesian optimisation is to select the best hyper-parameters for the random forest, i.e. the number of random forest decision trees, the depth of leaves, and the number of features. The experimental results show that an average accuracy of 99.3% can be achieved by using the time-frequency fusion with the Bayesian optimisation random forest model to recognise six actions.
ABSTRACT
With the introduction of mobile-wearable gadgets, the embedding of health-tracking capabilities has advanced dramatically. The Covid-19 years have accelerated research toward wearable sensor-based health monitoring. One of the most important applications in health monitoring is human activity identification. Daily behaviors like walking, sitting and jogging, as well as crucial activities such as falling forward and backward, provide a barrier in HAR (human action recognition) since they are semantically comparable. Previously developed deep learning algorithms have addressed some of these issues. However, these algorithms are hindered by a lack of training data. To cope with nonuniform samples in the human activities data, which can lead to overfitting results, this work introduces the synthetic minority oversampling approach. This article proposes a unique configuration of stacked convolutional neural network (CNN)-AR-DenseNet. With Bayesian optimization, the parameters of AR-DenseNet are optimally optimized. When compared to state-of-the-art stacked CNN network methods, the classification accuracy improved by up to 3.22%, with a substantial improvement of 5.8% over the standard CNN algorithm. © 2022 John Wiley & Sons, Ltd.
ABSTRACT
Artificial Intelligence (AI) and Internet of Things (IoT) offer immense potential to transform conventional healthcare systems. The IoT and AI enabled smart systems can play a key role in driving the future of smart healthcare. Remote monitoring of critical and non-critical patients is one such field which can leverage the benefits of IoT and machine learning techniques. While some work has been done in developing paradigms to establish effective and reliable communications, there is still great potential to utilize optimized IoT network and machine learning technique to improve the overall performance of the communication systems, thus enabling fool-proof systems. This study develops a novel IoT framework to offer ultra-reliable low latency communications to monitor post-surgery patients. The work considers both critical and non-critical patients and is balanced between these to offer optimal performance for the desired outcomes. In addition, machine learning based regression analysis of patients' sensory data is performed to obtain highly accurate predictions of the patients' sensory data (patients' vitals), which enables highly accurate virtual observers to predict the data in case of communication failures. The performance analysis of the proposed IoT based vital signs monitoring system for the post-surgery patients offers reduced delay and packet loss in comparison to IEEE low latency deterministic networks. The gradient boosting regression analysis also gives a highly accurate prediction for slow as well as rapidly varying sensors for vital sign monitoring.