Published papers

Vehicular ad hoc networks (VANETs) are integral to Intelligent Transportation Systems (ITS), serving as a crucial platform for deploying applications. The overall performance of VANETs is highly dependent on the vehicles, emphasizing the importance of effective fault (vehicle crash or exit) detection methods. Traditional fault detection faces challenges due to the high mobility of vehicles and prevalent software, hardware, and communication link faults, necessitating a robust solution for timely and accurate fault detection. This paper introduces ET-FD, a novel fault detection method based on ensemble learning designed to address the unique challenges in VANETs. ET-FD satisfies stringent accuracy requirements and aims to reduce system overhead significantly. We present experimental results demonstrating the superior performance of ET-FD in terms of fault detection accuracy and false positive rates, alongside comparisons with other classification methods that highlight its reduced training time. This confirms the potential of ET-FD as an efficient and reliable solution for fault detection in VANET environments.

Recent advancements in wireless communication technologies have made Wi-Fi signals indispensable in both personal and professional settings. The utilization of these signals for Human Activity Recognition (HAR) has emerged as a cutting-edge technology. By leveraging the fluctuations in Wi-Fi signals for HAR, this approach offers enhanced privacy compared to traditional visual surveillance methods. The essence of this technique lies in detecting subtle changes when Wi-Fi signals interact with the human body, which are then captured and interpreted by advanced algorithms. This article initially provides an overview of the key methodologies in HAR and the evolution of non-contact sensing, introducing sensor-based recognition, computer vision, and Wi-Fi signal based approaches, respectively. It then explores tools for Wi-Fi-based HAR signal collection and lists several high-quality datasets. Subsequently, the article reviews various sensing tasks enabled by Wi-Fi signal recognition, highlighting the application of deep learning networks in Wi-Fi signal detection. Experimental results are then presented that assess the capabilities of different networks. The findings indicate significant variability in the generalization capacities of neural networks and notable differences in test accuracy for various motion analyses.

Person identification is a critical task in applications such as security and surveillance, requiring reliable systems that perform robustly under diverse conditions. This study evaluates the Vision Transformer (ViT) and ResNet34 models across three modalities—RGB, thermal, and depth—using datasets collected with infrared array sensors and LiDAR sensors in controlled scenarios and varying resolutions (16 × 12 to 640 × 480) to explore their effectiveness in person identification. Preprocessing techniques, including YOLO-based cropping, were employed to improve subject isolation. Results show a similar identification performance between the three modalities, in particular in high resolution (i.e., 640 × 480), with RGB image classification reaching 100.0%, depth images reaching 99.54% and thermal images reaching 97.93%. However, upon deeper investigation, thermal images show more robustness and generalizability by maintaining focus on subject-specific features even at low resolutions. In contrast, RGB data performs well at high resolutions but exhibits reliance on background features as resolution decreases. Depth data shows significant degradation at lower resolutions, suffering from scattered attention and artifacts. These findings highlight the importance of modality selection, with thermal imaging emerging as the most reliable. Future work will explore multi-modal integration, advanced preprocessing, and hybrid architectures to enhance model adaptability and address current limitations. This study highlights the potential of thermal imaging and the need for modality-specific strategies in designing robust person identification systems.

The development of a cost-effective digital biomarker for detecting dementia is highly needed. While numerous studies have explored dementia detection through speech and natural language analysis, only a few studies have focused on dementia detection using face video recordings, and more in-depth research is needed. In this paper, we propose a method for detecting dementia and mild cognitive impairment (MCI), a pre-dementia stage, by utilizing four types of facial expression features extracted from recorded videos of participants. These features include Action Units, emotion categories, Valence-Arousal, and face embeddings. From the above features obtained from each video frame, various statistical information was extracted and used as features, and predictions were performed using a decision tree-based model. Our method was evaluated using face video recordings during conversations. The method achieved an area under the receiver operating characteristic curve (AUC) of 0.933 for dementia detection and 0.889 for MCI detection. Statistical analysis of facial expression features revealed that participants with dementia had fewer positive emotions, more negative emotions, and lower valence and arousal than healthy participants. These results indicate that the proposed method could serve as an explainable screening tool for the early detection of dementia and MCI.

In recent years, with the rapid development of 5G networks, the road traffic network composed of vehicles with different energy sources has become more and more complex, and the problems of environmental pollution and road congestion have also become increasingly serious. Electric vehicles are favored by people due to their environmental protection and energy-saving characteristics. However, improper charging dispatching will cause excess energy in charging stations, affecting the power grid and road traffic, such as energy shortages and lower traffic throughput. Therefore, how to design a reasonable charging strategy that can maximize the user's charging satisfaction and consume the energy of the charging station as much as possible becomes a challenge. Meanwhile, this strategy should consider power economic dispatch to reduce power generation costs and polluting gas emissions. With the support of 5G's high-bandwidth and low-latency characteristics, this paper designs an intelligent charging model which indirectly reflects the charging satisfaction through the time cost, energy consumption cost, charging cost, and the user's range anxiety, while consuming the remaining energy of the charging station as much as possible. Due to the uncertainty of wind and photovoltaic power generation, this paper proposes a two-stage economic dispatch model to improve the accuracy of power dispatch and reduce power generation costs and carbon emissions. Due to the highly variable traffic environment and energy demand, we employ proximal policy optimization-based deep reinforcement learning algorithms to realize electric vehicle charging dispatching and charging station power dispatching. Numerical results show the efficiency of our proposed strategy for electric vehicle charging in terms of the convergence speed.

Digital twin technology can be used in the internet of energy (IoE) systems to perform physical objects optimization and decision-making by updating relevant models in real-time. Model update is performed using real-time data generated by multiple devices contained in an IoE system. Since volume of the real-time data generated by IoE system is big, it is difficult for digital twin models to map the real-time data into virtual space. The difficulty of data mapping is because of high data volume and high resource consumption. In this paper, an adaptive data mapping mechanism using deep reinforcement learning (DRL) technique is devised. The proposed mechanism can dynamically adjust mapping data selection according to the environmental state and user requirements. By dynamically adjusting mapping data selection, the data and resource consumption used for mapping can be effectively reduced. The experimental results validate that the proposed mechanism can perform dynamic mapping data selection with reduced mapping data and low resource consumption.

The wake target has garnered increasing attention due to its length, which can be up to ten times that of the ship, and its inclusion of critical navigation information such as heading and speed. However, deep learning methods used in synthetic aperture radar (SAR) image wake detection tasks are limited to analyzing the features of the image itself, overlooking the characteristics of ship wakes in the frequency domain. This letter proposes a network called pseudo-siamese and multidomain feature fusion network (PSMDNet) that is composed of two parallel feature extraction branches. The feature extraction in the frequency domain uses the frequency channel attention network (FcaNet) as the backbone, incorporating an adjacent scale space attention module (ASSAM) to fuse high-level features into low-level features. The time domain uses the residual network (ResNet) as the backbone, incorporating a bidirectional feature channel module (BFCM) to enhance the representation of low-level spatial information. These two parallel branches extract the time- and frequency-domain features from the image to better capture the wake feature information. The proposed ASSAM module calculates weighted coding with context information, thereby selectively aggregating the unique linear spatial features of the wake into the low-level feature map. Verification experiments were conducted on the SAR-WAKE dataset, and the results demonstrate that the proposed method excels in detection accuracy compared with other algorithms, achieving excellent results of 92.71%. Particularly noteworthy is that the positioning and visualization of wake vertex and Kelvin arms are realized by the loss function designed for the wake.

Radiofrequency signal identification (RSI) provides a critical security solution for device authentication in the Internet of Things (IoT), characterized by extensive interconnections and interactions among numerous entities. By analyzing received radiofrequency signals, device-specific features are extracted at the receiver and used for identification. In a dynamic and ever-changing communication environment, where some devices not visible during the training process may appear during testing, a robust RSI method must not only identify devices encountered during training but also reject those that were not. In this article, we propose an open-set RSI method based on hypersphere manifold embedding. This approach leverages hypersphere projection for radiofrequency signal feature extraction on a hypersphere manifold, thereby avoiding the need to optimize intradevice variation in the radial direction. Additionally, we introduce an open-set identification approach based on generalized Pareto distribution, which does not rely on any radiofrequency signals from unknown devices. Extensive experimental results demonstrate that the proposed method achieves state-of-the-art identification performance.

Simultaneous Wireless Information and Power Transfer (SWIPT) is a unified approach to transfer power and information by exploiting the properties of radio signals. This paper considers a multi-antenna base station, multiple energy harvesting receivers (EHRs), and legitimate receivers in the presence of multiple multi-antenna eavesdroppers. A resource allocation problem of maximizing the harvested energy available at the EHR side in the presence of secrecy capacity and quality of service constraints is formulated for analysis. The transmit power, secrecy, and total transmit power requirements are used to formulate the constraints. We then propose neural network methods that use the channel state information data and constraint characteristics to obtain novel activation functions that can be used to predict beamforming vectors. Simulation results show that the proposed activation functions predict beamforming vectors to achieve better harvested energy and secrecy capacity.

In elderly monitoring or indoor intrusion detection, the recognition of human activity is a key task. Owing to several strengths of Wi-Fi-based devices, including their non-contact and privacy protection, these devices have been widely applied in the area of smart homes. By the deep learning technique, numerous Wi-Fi-based activity recognition methods can realize satisfied recognitions, however, these methods may fail to recognize the activities of an unknown person without the learning process. In this study, using channel state information (CSI) data, a novel cross-person activity recognition (CPAR) method is proposed by a deep learning approach with generalization capability. Combining one of the state-of-the-art deep neural networks (DNNs) used in activity recognition, i.e., attention-based bi-directional long short-term memory (ABLSTM), the snapshot ensemble is the first to be adopted to train several base-classifiers for enhancing the generalization and practicability of recognition. Second, to discriminate the extracted features, metric learning is further introduced by using the center loss, obtaining snapshot ensemble-used ABLSTM with center loss (SE-ABLSTM-C). In the experiments of CPAR, the proposed SE-ABLSTM-C method markedly improved the recognition accuracies to an application level, for seven categories of activities.

High-altitude platform stations (HAPSs) that provide direct communication services to smartphones on the ground are garnering significant attention as innovative mobile communication platforms suitable for ultrawide coverage areas. The deployment of a multi-cell configuration is essential for enhancing communication capacity over large geographical regions. However, designing and optimizing multi-cell configurations proves challenging due to the numerous antenna parameters involved, such as beamwidth and beam direction. HAPSs also hold promise in serving as reliable networks during disasters, as they can quickly restore coverage when terrestrial base stations (BSs) are in failure, allowing users to maintain connectivity via HAPSs using the same frequency band as terrestrial networks. Despite this advantage, HAPSs can potentially cause interference in terrestrial networks during such scenarios, necessitating careful considerations to avoid disrupting existing terrestrial networks. However, the optimization of cell configurations in the coexistence of HAPSs and terrestrial networks remains largely unexplored. To address this gap, this paper proposes a design scheme that leverages a genetic algorithm (GA) to optimize antenna parameters for each cell, thereby maximizing both user coverage and area coverage while ensuring coexistence with terrestrial networks. We optimize antenna parameters to increase user coverage and area coverage as a multi-objective optimization problem, resulting in a Pareto front that illustrates the trade-off choices. Additionally, the proposed method protects terrestrial networks by setting the initial beam direction of each cell to avoid interference with terrestrial BSs. Consequently, the GA can initiate optimization with candidate combinations that adhere to the interference constraints of terrestrial networks, thus expediting the optimization process and yielding improved coverage. Simulation results demonstrate that the proposed scheme outperforms conventional schemes which solely control beam direction, leading to superior coverage. While the results show trade-off between user coverage and area coverage, it is also clarified that considering both types of coverage show the potential to achieve significant improvements in both aspects simultaneously. Moreover, the results highlight the efficacy of the initial beam direction setting in enhancing coverage rates and significantly reducing the number of combinations required for the GA to converge.

Contactless Wi-Fi sensing, using channel state information (CSI) fingerprints, plays a pivotal role in communication, smart healthcare, and industrial automation. Deep learning has revolutionized the efficiency of non-contact sensing technology. Owing to its robust feature extraction capabilities and the interconnectedness of diverse sensing tasks, methods that address multiple tasks at once, like joint activity recognition and indoor localization (JARIL), have gained prominence. The primary goal of JARIL is to improve performance while reducing computational demands. Nevertheless, there remains substantial potential for enhancing its effectiveness through additional refinement and optimization measures. To address this, we introduce a regularized multi-label learning (RML) framework specifically designed for JARIL. This framework combines a parameter-efficient backbone network based on multi-scale separable convolution with residual connections, and a regularization training strategy. The latter strategy boosts performance by linearly combining two distinct CSI samples with their labels, creating new training instances in the training process. Simulation results show that the proposed method boasts a recognition accuracy of 91.73% and a localization precision of 99.64%. This marks an improvement of 4.32% and 3.60% respectively, in comparison to the prior ResNet1D+-based JARIL method. The codes can be downloaded from https://github.com/BeechburgPieStar/JARIL.

Malware traffic classification (MTC) is one of the important research topics in the field of cyber security. Existing MTC methods based on deep learning have been developed based on the assumption of enough high-quality samples and powerful computing resources. However, both are hard to obtain in real applications especially in availability of IoT. In this paper, we propose a few-shot MTC (FS-MTC) method combining knowledge transfer and neural architecture search (i.e. NAS-based FS-MTC) with limited training samples as well as acceptable computational resources, in order to mitigate the identified challenges. Specifically, our proposed method first converts the raw network traffic into traffic images through data pre-processing to serve as input data for the neural network. Second, we use neural architecture search to adaptively search for the effective feature extraction model on the source domain (including Edge-IIoTset, Bot-IoT, and benign USTC-TFC2016). Third, the searched model is pre-trained on source task to achieve the generic feature representation of malware traffic. Finally, we only use few-shot malware traffic samples to fine-tune the pre-trained model to quickly adapt to new types of MTC tasks in realistic network environments. The experimental results show that the proposed NAS-based FS-MTC method has great scalability and classification performance in different FS-MTC tasks, including 5-way K-shot USTC-TFC2016 dataset and 10-way K-shot CIC-IoT dataset. Compared with state-of-the-art methods in the field of malware classification, the proposed NAS-based FS-MTC has higher classification accuracy. Especially in the 1-shot case of the USTC-TFC2016 dataset, its average accuracy is as high as 86.91%.

Ensuring network security, effective malware detection (MD) is of paramount importance. Traditional methods often struggle to accurately learn and process the characteristics of network traffic data, and must balance rapid processing with retaining memory for previously encountered malware categories as new ones emerge. To tackle these challenges, we propose a cutting-edge approach using self-paced class incremental learning (SPCIL). This method harnesses network traffic data for enhanced class incremental learning (CIL). A pivotal technique in deep learning, CIL facilitates the integration of new malware classes while preserving recognition of prior categories. The unique loss function in our SPCIL-driven MD combines sparse pairwise loss with sparse loss, striking an optimal balance between model simplicity and accuracy. Experimental results reveal that SPCIL proficiently identifies both existing and emerging malware classes, adeptly addressing catastrophic forgetting. In comparison to other incremental learning approaches, SPCIL stands out in performance and efficiency. It operates with a minimal model parameter count (8.35 million) and in increments of 2, 4, and 5, achieves impressive accuracy rates of 89.61%, 94.74%, and 97.21% respectively, underscoring its effectiveness and operational efficiency.

People Identification is a critical aspect in developing modern vehicles, aimed at enhancing safety and comfort levels. Most traditional methods of people identification in vehicles use RGB images or videos. In this study, we introduce a novel methodology for identifying individuals in private car scenarios, utilizing 3D Light Detection and Ranging (LiDAR) technology, generative image inpainting based on Contextual Attention, and the YOLOv5 model. Initially, we gather data utilizing a 3D-LiDAR instrument and subsequently convert the acquired depth data into depth images. Following this, the depth images are annotated manually to indicate the positions and identifiers of various individuals occupying distinct seats. This annotated data serves as the training material for the YOLOv5 model, facilitating the recognition and categorization of subjects. However, given that individuals seated in the back often have parts of their bodies occluded by the front seats and the passengers in them, we employ generative image inpainting techniques to reveal the occluded portions. This step significantly enhances the precision in detecting and identifying individuals situated in the back seats. We implemented our strategy on a restricted group of four participants, conducting training and testing phases within identical environments. Prior to the inpainting process, the classification's F1 score stood at 66.5%. After inpainting, we observed a notable surge in the F1 score for the rear-seat passengers increased by 17.1%.

In this article, we leverage the outage probability knowledge map to characterize the connection between unmanned aerial vehicles (UAVs) and cellular networks. The outage probability knowledge map is a database that simulates the connection between UAV and the cellular network during real hovers, which helps to enhance the UAV’s awareness of the environment and reduce the connection interruption under complex real-time channel state information. We assume that the UAV roughly samples from the actual radio environment of the airspace in advance, and calculates the outage probability of the sampled points. After that, the UAV reconstructs the actual outage knowledge map, and flies in the airspace to learn the optimal UAV trajectory planning policy based on the reconstructed map. The optimization objective is to minimize the flight energy cost of the UAV performing tasks. In this article, we propose a deep image prior based radio map reconstruction (DIPRMR) method to reconstruct the map, and then propose a deep reinforcement learning based trajectory optimization algorithm. The UAV that performs the task adjusts the flight trajectory based on the outage probability knowledge obtained from the reconstructed complete map. Simulation results show that the proposed online trajectory optimization scheme based on outage probability knowledge map can obtain great returns in terms of maintaining connectivity, reducing task completion time and energy consumption.