In heterogeneousWireless Rechargeable Sensor Networks (WRSNs) comprising both fixed and mobile nodes, existing energy replenishment strategies face severe challenges in cost, efficiency, and scalability. Fixed sensor clusters require efficient area-based charging to maintain network density, while dynamic sensors with unpredictable trajectories demand highly responsive on-demand service. Conventional single-modal or simplistic hybrid schemes cannot simultaneously satisfy these distinct and often conflicting operational requirements, leading to resource mismatch and critically limiting the overall network lifetime and quality of service. This paper introduces an Adaptive Co-operative Dual-Mode Charging Scheduling (AC-DCS) scheme, which integrates two novel algorithms. For fixed sensor clusters, the Portable-Charger Deployment Planning (PCDP) algorithm optimizes “one-to-many” charging using portable directional modules; for dynamic sensors, the Deep Adaptive Rescue Selector (DARS), a lightweight reinforcement learning model, provides ”one-to-one” on-demand service. Extensive simulations validate that in a 500-node network, AC-DCS improves the sensor survival rate by over 23%, improves mobile energy efficiency by approximately 65%, and reduces charging latency by at least 33% compared to state-of-the-art benchmarks, offering a robust solution for long-term large-scale IoT deployments.

Wireless Rechargeable Sensor Networks (WRSNs) extend the operational lifetime of battery-powered sensor deployments through Wireless Power Transfer (WPT). Traditional approaches often address node clustering, charging strategies, and charger routing independently, frequently failing to adapt to real-time demand variations and network survivability. This paper proposes a two-phase Sector-based Charging with Resilient Scheduling (SCRS) scheme designed for WRSNs utilizing fixed directional chargers. In Phase I, the SCRS scheme provides initial energy to mitigate the risk of early sensor outages. Phase II employs a quantile-based strategy to dynamically allocate charging time, thereby tightly constraining energy overflow. By optimizing charging schedules based on real-time demands, SCRS enhances energy utilization and improves network survivability. Simulation results demonstrate that SCRS reduces charging time by up to 62.96%, decreases energy overflow by up to 88.89%, and achieves near-zero sensor outage durations on average across broad configurations. Furthermore, the proposed scheme completely eliminates outages under specific high-agility configurations (e.g., utilizing the Q1 setting). Crucially, the validation context of this study incorporates rich stochastic radio models, including Nakagami-mfading and log-normal shadowing, to closely emulate realistic multipath and obstructed environments. Consequently, while the analytical bounds and comparative advantages of the SCRS scheme are robust within these environments, transitioning these findings to highly mobile real-world IoT deployments will necessitate further empirical validation in physical testbeds.

Atrial Fibrillation (AF) screening from face videos has become popular with the trend of telemedicine and telehealth in recent years. In this study, the largest facial image database for camera-based AF detection is proposed. There are 657 participants from two clinical sites and each of them is recorded for about 10 minutes of video data, which can be further processed as over 10 000 segments around 30 seconds, where the duration setting is referred to the guideline of AF diagnosis. It is also worth noting that, 2 979 segments are segment-wise labeled, that is, every rhythm is independently labeled with AF or not. Besides, all labels are confirmed by the cardiologist manually. Various environments, talking, facial expressions, and head movements are involved in data collection, which meets the situations in practical usage. Specific to camera-based AF screening, a novel CNN-based architecture equipped with an attention mechanism is proposed. It is capable of fusing heartbeat consistency, heart rate variability derived from remote photoplethysmography, and motion features simultaneously to reliable outputs. With the proposed model, the performance of intra-database evaluation comes up to 96.62% of sensitivity, 90.61% of specificity, and 0.96 of AUC. Furthermore, to check the capability of adaptation of the proposed method thoroughly, the cross-database evaluation is also conducted, and the performance also reaches about 90% on average with the AUCs being over 0.94 in both clinical sites.

Although deep convolutional networks have achieved great performance in face recognition tasks, the challenge of domain discrepancy still exists in real world applications. Lack of domain coverage of training data (source domain) makes the learned models degenerate in a testing scenario (target domain). In face recognition tasks, classes in two domains are usually different, so classical domain adaptation approaches, assuming there are shared classes in domains, may not be reasonable solutions for this problem. In this paper, self-supervised learning is adopted to learn a better embedding space where the subjects in target domain are more distinguishable. The learning goal is maximizing the similarity between the embeddings of each image and its mirror in both domains. The experiments show its competitive results compared with prior works. To know the reason why it can achieve such performance, we further discuss how this approach affects the learning of embeddings.

Facial action unit detection (FAUD) is aimed to detect subtle facial motions, known as action units (AUs), induced by the traction or towing of facial muscles. In recent years, FAUD has become an attractive task since these slight changes on face reveal some cues of emotions which can then be utilized to infer people’s affective. Many prior works are developed based on physical features of AUs, such as regional occurrence and temporal continuity, but few of them concern the negative impact of individual differences, causing an unrobust result in variant subjects and environments. To deal with the problem, a contrastive feature learning method is proposed to make a convolutional neural network (CNN) learn to extract the contrastive feature which is the difference between the features of a neutral face image and an AU-occurred face image. In this way, the individual information will be mitigated so that it becomes easier to detect AUs according to the features. A great disparity between the number of positive and negative samples, named as data imbalance, makes a serious interference in most detection problems, including FAUD. A lower occurrence rate of an AU leads a more severe data imbalance problem, which results in a biased result for the AU detection. Therefore, the class-weighted loss is proposed to change the weight-ratio between the positive and negative samples so as to encourage the learning of the positive samples and ease that of negative samples. Two widely used databases, BP4D and DISFA, are adopted to be the benchmark of the performance testing. In the experiment, it shows that the proposed method performs well both in accuracy and speed comparing with the state-of-the-art approaches.

To handle the customer distribution in the certain areas, crowd counting is necessary for such applications, which is a labor-intensive work for human. Therefore, an automatic crowd counting system is in great demand, but it is still a challenging problem since the human heads and bodies are usually highly overlapping in crowd images. In this paper, a counting-by-regression framework is employed. The human head is modeled as a Guassian distribution. With a crowd density map estimator, the head count can be obtained by integrating over the density map. Most existing approaches only apply density map regression for training a density map estimator, but it is hard to find a suitable training parameters to train a good one; actually, the head count is overestimated easily. To mitigate this problem, counting regression is combined with density map regression. A deeper and lighter fully convolutional network (FCN) is designed to be a crowd density map estimator. The input and output size of the FCN are the same. After training by the proposed method, our model is more competitive comparing with others. The parameter quantity of the model is the lowest, and it needs the least inference time.

To successfully increase athletes' or exercisers' fitness and endurance, the factors of physiological signal, emotion, or the level of fatigue should be considered during the training program. Many clinical decision support systems can assist to monitor the exercisers by some wearable devices. And, the questionnaire should also be taken into account to produce a report. Such process is cumbersome, and the results are not objective. Furthermore, one may feel uncomfortable when wearing the devices during the training program. In this research, the Rating of Perceived Exertion (RPE) is expected to be estimated automatically without any wearable devices and questionnaires. A camera based heart rate detection algorithm and a fatigue expression feature extractor are fused to estimate the RPE value. The results show that our heart rate detection algorithm can be competitive to the wearable devices, and the trend of the detected heart rate is correlated to RPE. Moreover, the fatigue feature can help reduce the error of the estimation.

A Tent-map chaotic Newton-Raphson optimization based neural network predictive control (TCNR-NPC) is developed to apply to the long-delay permanent magnet synchronous motor (PMSM) system in this paper. Due to a nonlinear model utilized in the predictive controller, nonlinear optimization methods turn into an important issue. To overcome the shortcoming of the conventional nonlinear programming on the initial condition sensitivity and maintain the accuracy of optimal solution, chaos optimization algorithm (COA) and Newton-Raphson (NR) are combined. With the comparison of COA and NR based optimization methods, our approach, the Tent-map chaotic Newton-Raphson (TCNR) optimization, is easier to reach the global optimum, thus, it would be employed in neural network predictive control. It is found that TCNR-NPC has a better performance than those of GPC, modified GPC, adaptive extended PSO based NPC, and PSO based PI controllers in real experiments.

Recent advances of digital microfluidic biochips (DMFBs) have revolutionized the traditional laboratory procedures. By providing the droplet-based system, DMFB can perform real-time biological analysis and safety-critical biomedical applications. However, different droplets being transported and manipulated on the DMFB may introduce the contamination problem caused by liquid residue between different biomolecules. To overcome this problem, a wash droplet is introduced to clean the contaminations on the surface of the microfluidic array. However, current scheduling of wash droplet does not restrict the extra used cells and execution time of bioassay, thereby degrading the reliability and fault-tolerance significantly. In this paper, we propose a contamination aware droplet routing algorithm for DMFBs. To reduce the routing complexity and the used cells, we first construct preferred routing tracks by analyzing the global moving vector of droplets to guide the droplet routing. To cope with contaminations within one subproblem, we first apply ak-shortest path routing technique to minimize the contaminated spots. Then, to take advantage of multiple wash droplets, we adopt a minimum cost circulation (MCC) algorithm for optimal wash-droplet routing to simultaneously minimize used cells and the cleaning time. Since the droplet routing problem consists of several subproblems, a look-ahead prediction technique is further used to determine the contaminations between successive subproblems. After that, we can simultaneously clean both contaminations within one subproblem and those between successive subproblems by using the MCC-based algorithm to reduce the execution time and the used cells significantly. Based on four widely used bioassays, our algorithm reduces the used cells and the execution time significantly compared with the state-of-the-art algorithm.