This study characterized the absolute pressure measurement error and reliability of a new fully integrated (Kinetyx, SI) plantar-pressure measurement system (PPMS) versus an industry-standard PPMS (F-Scan, Tekscan) during an established benchtop testing protocol as well as via a research-grade, instrumented treadmill (Bertec) during a running protocol. Benchtop testing results showed that both SI and F-Scan had strong positive linearity (Pearson’s correlation coefficient, PCC = 0.86–0.97, PCC = 0.87–0.92; RMSE = 15.96 ± 9.49) and mean root mean squared error RMSE (9.17 ± 2.02) compared to the F-Scan on a progressive loading step test. The SI and F-Scan had comparable results for linearity and hysteresis on a sinusoidal loading test (PCC = 0.92–0.99; 5.04 ± 1.41; PCC = 0.94–0.99; 6.15 ± 1.39, respectively). SI had less mean RMSE (6.19 ± 1.38) than the F-Scan (8.66 ±2.31) on the sinusoidal test and less absolute error (4.08 ± 3.26) than the F-Scan (16.38 ± 12.43) on a static test. Both the SI and F-Scan had near-perfect between-day reliability interclass correlation coefficient, ICC = 0.97–1.00) to the F-Scan (ICC = 0.96–1.00). During running, the SI pressure output had a near-perfect linearity and low RMSE compared to the force measurement from the Bertec treadmill. However, the SI pressure output had a mean hysteresis of 7.67% with a 28.47% maximum hysteresis, which may have implications for the accurate quantification of kinetic gait measures during running. Full article

At present, some studies have combined federated learning with blockchain, so that participants can conduct federated learning tasks under decentralized conditions, sharing and aggregating model parameters. However, these schemes do not take into account the trusted supervision of federated learning and the case of malicious node attacks. This paper introduces the concept of a trusted computing sandbox to solve this problem. A federated learning multi-task scheduling mechanism based on a trusted computing sandbox is designed and a decentralized trusted computing sandbox composed of computing resources provided by each participant is constructed as a state channel. The training process of the model is carried out in the channel and the malicious behavior is supervised by the smart contract, ensuring the data privacy of the participant node and the reliability of the calculation during the training process. In addition, considering the resource heterogeneity of participant nodes, the deep reinforcement learning method was used in this paper to solve the resource scheduling optimization problem in the process of constructing the state channel. The proposed algorithm aims to minimize the completion time of the system and improve the efficiency of the system while meeting the requirements of tasks on service quality as much as possible. Experimental results show that the proposed algorithm has better performance than the traditional heuristic algorithm and meta-heuristic algorithm. Full article

Optical techniques are often inadequate in estimating bruise age since they are not sensitive to the depth of chromophores at the location of the bruise. To address this shortcoming, we used pulsed photothermal radiometry (PPTR) for depth profiling of bruises with two wavelengths, 532 nm (KTP laser) and 1064 nm (Nd:YAG laser). Six volunteers with eight bruises of exactly known and documented times of injury were enrolled in the study. A homogeneous part of the bruise was irradiated first with a 5 ms pulse at 532 nm and then with a 5 ms pulse at 1064 nm. The resulting transient surface temperature change was collected with a fast IR camera. The initial temperature–depth profiles were reconstructed by solving the ill-posed inverse problem using a custom reconstruction algorithm. The PPTR signals and reconstructed initial temperature profiles showed that the 532 nm wavelength probed the shallow skin layers revealing moderate changes during bruise development, while the 1064 nm wavelength provided additional information for severe bruises, in which swelling was present. Our two-wavelength approach has the potential for an improved estimation of the bruise age, especially if combined with modeling of bruise dynamics. Full article

With the rise of latency-sensitive and computationally intensive applications in mobile edge computing (MEC) environments, the computation offloading strategy has been widely studied to meet the low-latency demands of these applications. However, the uncertainty of various tasks and the time-varying conditions of wireless networks make it difficult for mobile devices to make efficient decisions. The existing methods also face the problems of long-delay decisions and user data privacy disclosures. In this paper, we present the FDRT, a federated learning and deep reinforcement learning-based method with two types of agents for computation offload, to minimize the system latency. FDRT uses a multi-agent collaborative computation offloading strategy, namely, DRT. DRT divides the offloading decision into whether to compute tasks locally and whether to offload tasks to MEC servers. The designed DDQN agent considers the task information, its own resources, and the network status conditions of mobile devices, and the designed D3QN agent considers these conditions of all MEC servers in the collaborative cloud-side end MEC system; both jointly learn the optimal decision. FDRT also applies federated learning to reduce communication overhead and optimize the model training of DRT by designing a new parameter aggregation method, while protecting user data privacy. The simulation results showed that DRT effectively reduced the average task execution delay by up to 50% compared with several baselines and state-of-the-art offloading strategies. FRDT also accelerates the convergence rate of multi-agent training and reduces the training time of DRT by 61.7%. Full article

Owing to the increasing popularity of smart agriculture in recent years, it is necessary to develop a single sensor that can measure several soil properties, particularly the soil water content and matric potential. Therefore, in this study, we developed a sensor that can simultaneously measure soil water content (θ), electrical conductivity (σ_b), temperature, and matric potential (ψ). The proposed sensor can determine θ and σ_b using time domain transmissiometry and can determine ψ based on the capacitance of the accompanying ceramic plate. A series of laboratory and field tests were conducted to evaluate the performance of the sensor. The sensor output values were correlated with the soil properties, and the temperature dependence of the sensor outputs was evaluated. Additionally, field tests were conducted to measure transient soil conditions over a long period. The results show that the developed sensor can measure each soil property with acceptable accuracy. Moreover, the root-mean-square errors of the sensor and reference values were 1.7 for the dielectric constant (which is equivalent to θ), 62 mS m⁻¹ for σ_b, and 0.05–0.88 for log ψ. The temperature dependence was not a problem, except when ψ was below −100 kPa. The sensor can be used for long-term measurements in agricultural fields and exhibited sufficient lifetime and performance. We believe that the developed sensor can contribute to smart agriculture and research on heat and mass transfer in soil. Full article

Intervals of low-quality photoplethysmogram (PPG) signals might lead to significant inaccuracies in estimation of pulse arrival time (PAT) during polysomnography (PSG) studies. While PSG is considered to be a “gold standard” test for diagnosing obstructive sleep apnea (OSA), it also enables tracking apnea-related nocturnal blood pressure fluctuations correlated with PAT. Since the electrocardiogram (ECG) is recorded synchronously with the PPG during PSG, it makes sense to use the ECG signal for PPG signal-quality assessment. (1) Objective: to develop a PPG signal-quality assessment algorithm for robust PAT estimation, and investigate the influence of signal quality on PAT during various sleep stages and events such as OSA. (2) Approach: the proposed algorithm uses R and T waves from the ECG to determine approximate locations of PPG pulse onsets. The MESA database of 2055 PSG recordings was used for this study. (3) Results: the proportions of high-quality PPG were significantly lower in apnea-related oxygen desaturation (matched-pairs $r_c$ = 0.88 and $r_c$ = 0.97, compared to OSA and hypopnea, respectively, when p < 0.001) and arousal ($r_c$ = 0.93 and $r_c$ = 0.98, when p < 0.001) than in apnea events. The significantly large effect size of interquartile ranges of PAT distributions was between low- and high-quality PPG (p < 0.001, $r_c$ = 0.98), and regular and irregular pulse waves (p < 0.001, $r_c$ = 0.74), whereas a lower quality of the PPG signal was found to be associated with a higher interquartile range of PAT across all subjects. Suggested PPG signal quality-based PAT evaluation reduced deviations (e.g., $r_c$ = 0.97, $r_c$ = 0.97, $r_c$ = 0.99 in hypopnea, oxygen desaturation, and arousal stages, respectively, when p < 0.001) and allowed obtaining statistically larger differences between different sleep stages and events. (4) Significance: the implemented algorithm has the potential to increase the robustness of PAT estimation in PSG studies related to nocturnal blood pressure monitoring. Full article

The power demand from gas turbines in electrical grids is becoming more dynamic due to the rising demand for power generation from renewable energy sources. Therefore, including the transient data in the fault diagnostic process is important when the steady-state data are limited and if some component faults are more observable in the transient condition than in the steady-state condition. This study analyses the transient behaviour of a three-shaft industrial gas turbine engine in clean and degraded conditions with consideration of the secondary air system and variable inlet guide vane effects. Different gas path faults are simulated to demonstrate how magnified the transient measurement deviations are compared with the steady-state measurement deviations. The results show that some of the key measurement deviations are considerably higher in the transient mode than in the steady state. This confirms the importance of considering transient measurements for early fault detection and more accurate diagnostic solutions. Full article

Deep learning methods have achieved outstanding results in many image processing and computer vision tasks, such as image segmentation. However, they usually do not consider spatial dependencies among pixels/voxels in the image. To obtain better results, some methods have been proposed to apply classic spatial regularization, such as total variation, into deep learning models. However, for some challenging images, especially those with fine structures and low contrast, classical regularizations are not suitable. We derived a new regularization to improve the connectivity of segmentation results and make it applicable to deep learning. Our experimental results show that for both deep learning methods and unsupervised methods, the proposed method can improve performance by increasing connectivity and dealing with low contrast and, therefore, enhance segmentation results. Full article

Electric Vehicle (EV) charging demand and charging station availability forecasting is one of the challenges in the intelligent transportation system. With accurate EV station availability prediction, suitable charging behaviors can be scheduled in advance to relieve range anxiety. Many existing deep learning methods have been proposed to address this issue; however, due to the complex road network structure and complex external factors, such as points of interest (POIs) and weather effects, many commonly used algorithms can only extract the historical usage information and do not consider the comprehensive influence of external factors. To enhance the prediction accuracy and interpretability, the Attribute-Augmented Spatiotemporal Graph Informer (AST-GIN) structure is proposed in this study by combining the Graph Convolutional Network (GCN) layer and the Informer layer to extract both the external and internal spatiotemporal dependence of relevant transportation data. The external factors are modeled as dynamic attributes by the attributeaugmented encoder for training. The AST-GIN model was tested on the data collected in Dundee City, and the experimental results showed the effectiveness of our model considering external factors’ influence on various horizon settings compared with other baselines. Full article

Internet of Drones (IoD), designed to coordinate the access of unmanned aerial vehicles (UAVs), is a specific application of the Internet of Things (IoT). Drones are used to control airspace and offer services such as rescue, traffic surveillance, environmental monitoring, delivery and so on. However, IoD continues to suffer from privacy and security issues. Firstly, messages are transmitted over public channels in IoD environments, which compromises data security. Further, sensitive data can also be extracted from stolen mobile devices of remote users. Moreover, drones are susceptible to physical capture and manipulation by adversaries, which are called drone capture attacks. Thus, the development of a secure and lightweight authentication scheme is essential to overcoming these security vulnerabilities, even on resource-constrained drones. In 2021, Akram et al. proposed a secure and lightweight user–drone authentication scheme for drone networks. However, we discovered that Akram et al.’s scheme is susceptible to user and drone impersonation, verification table leakage, and denial of service (DoS) attacks. Furthermore, their scheme cannot provide perfect forward secrecy. To overcome the aforementioned security vulnerabilities, we propose a secure mutual authentication and key agreement scheme between user and drone pairs. The proposed scheme utilizes physical unclonable function (PUF) to give drones uniqueness and resistance against drone stolen attacks. Moreover, the proposed scheme uses a fuzzy extractor to utilize the biometrics of users as secret parameters. We analyze the security of the proposed scheme using informal security analysis, Burrows–Abadi–Needham (BAN) logic, a Real-or-Random (RoR) model, and Automated Verification of Internet Security Protocols and Applications (AVISPA) simulation. We also compared the security features and performance of the proposed scheme and the existing related schemes. Therefore, we demonstrate that the proposed scheme is suitable for IoD environments that can provide users with secure and convenient wireless communications. Full article

The safety assessment of cyber-physical systems (CPSs) requires tremendous effort, as the complexity of cyber-physical systems is increasing. A well-known approach for the safety assessment of CPSs is fault injection (FI). The goal of fault injection is to find a catastrophic fault that can cause the system to fail by injecting faults into it. These catastrophic faults are less likely to occur, and finding them requires tremendous labor and cost. In this study, we propose a reinforcement learning (RL)-based method to automatically configure faults in the system under test and to find catastrophic faults in the early stage of system development at the model level. The proposed method provides a guideline to utilize high-level domain knowledge about a system model for constructing the reinforcement learning agent and fault injection setup. In this study, we used the system (safety) specification to shape the reward function in the reinforcement learning agent. The reinforcement learning agent dynamically interacted with the model under test to identify catastrophic faults. We compared the proposed method with random-based fault injection in two case studies using MATLAB/Simulink. Our proposed method outperformed random-based fault injection in terms of the severity and number of faults found. Full article

Cloud computing (CC) benefits and opportunities are among the fastest growing technologies in the computer industry. Cloud computing’s challenges include resource allocation, security, quality of service, availability, privacy, data management, performance compatibility, and fault tolerance. Fault tolerance (FT) refers to a system’s ability to continue performing its intended task in the presence of defects. Fault-tolerance challenges include heterogeneity and a lack of standards, the need for automation, cloud downtime reliability, consideration for recovery point objects, recovery time objects, and cloud workload. The proposed research includes machine learning (ML) algorithms such as naïve Bayes (NB), library support vector machine (LibSVM), multinomial logistic regression (MLR), sequential minimal optimization (SMO), K-nearest neighbor (KNN), and random forest (RF) as well as a fault-tolerance method known as delta-checkpointing to achieve higher accuracy, lesser fault prediction error, and reliability. Furthermore, the secondary data were collected from the homonymous, experimental high-performance computing (HPC) system at the Swiss Federal Institute of Technology (ETH), Zurich, and the primary data were generated using virtual machines (VMs) to select the best machine learning classifier. In this article, the secondary and primary data were divided into two split ratios of 80/20 and 70/30, respectively, and cross-validation (5-fold) was used to identify more accuracy and less prediction of faults in terms of true, false, repair, and failure of virtual machines. Secondary data results show that naïve Bayes performed exceptionally well on CPU-Mem mono and multi blocks, and sequential minimal optimization performed very well on HDD mono and multi blocks in terms of accuracy and fault prediction. In the case of greater accuracy and less fault prediction, primary data results revealed that random forest performed very well in terms of accuracy and fault prediction but not with good time complexity. Sequential minimal optimization has good time complexity with minor differences in random forest accuracy and fault prediction. We decided to modify sequential minimal optimization. Finally, the modified sequential minimal optimization (MSMO) algorithm with the fault-tolerance delta-checkpointing (D-CP) method is proposed to improve accuracy, fault prediction error, and reliability in cloud computing. Full article

Adapting intelligent context-aware systems (CAS) to future operating rooms (OR) aims to improve situational awareness and provide surgical decision support systems to medical teams. CAS analyzes data streams from available devices during surgery and communicates real-time knowledge to clinicians. Indeed, recent advances in computer vision and machine learning, particularly deep learning, paved the way for extensive research to develop CAS. In this work, a deep learning approach for analyzing laparoscopic videos for surgical phase recognition, tool classification, and weakly-supervised tool localization in laparoscopic videos was proposed. The ResNet-50 convolutional neural network (CNN) architecture was adapted by adding attention modules and fusing features from multiple stages to generate better-focused, generalized, and well-representative features. Then, a multi-map convolutional layer followed by tool-wise and spatial pooling operations was utilized to perform tool localization and generate tool presence confidences. Finally, the long short-term memory (LSTM) network was employed to model temporal information and perform tool classification and phase recognition. The proposed approach was evaluated on the Cholec80 dataset. The experimental results (i.e., 88.5% and 89.0% mean precision and recall for phase recognition, respectively, 95.6% mean average precision for tool presence detection, and a 70.1% F1-score for tool localization) demonstrated the ability of the model to learn discriminative features for all tasks. The performances revealed the importance of integrating attention modules and multi-stage feature fusion for more robust and precise detection of surgical phases and tools. Full article

The energy consumed in buildings constitutes more than half of the total electricity consumption and is highly correlated with the number of occupants; therefore, it is necessary to use occupancy information in energy consumption analysis. However, the number of occupants may not be accurate owing to measurement errors caused by various factors, such as the locations of sensors or cameras and the communication environment. In this study, occupancy was measured using an object recognition camera, the number of people was additionally collected by manual aggregation, measurement error in occupancy count was analyzed, and the true count was estimated using a deep learning model. The energy consumption based on occupancy was predicted using the measured and estimated values. To this end, deep learning was used to predict energy consumption after analyzing the correlation between occupancy and energy consumption. Results showed that the performance of occupancy estimation was 1.9 based on RMSE, which is a 71.1% improvement compared to the original occupancy sensing. The RMSE of predicted energy consumption based on the estimated occupancy was 56.0, which is a 5.2% improvement compared to the original energy estimation. Full article

The aim of this study was to determine a gait pattern, i.e., a subset of spatial and temporal parameters, through a supervised machine learning (ML) approach, which could be used to reliably distinguish Parkinson’s Disease (PD) patients with and without mild cognitive impairment (MCI). Thus, 80 PD patients underwent gait analysis and spatial–temporal parameters were acquired in three different conditions (normal gait, motor dual task and cognitive dual task). Statistical analysis was performed to investigate the data and, then, five ML algorithms and the wrapper method were implemented: Decision Tree (DT), Random Forest (RF), Naïve Bayes (NB), Support Vector Machine (SVM) and K-Nearest Neighbour (KNN). First, the algorithms for classifying PD patients with MCI were trained and validated on an internal dataset (sixty patients) and, then, the performance was tested by using an external dataset (twenty patients). Specificity, sensitivity, precision, accuracy and area under the receiver operating characteristic curve were calculated. SVM and RF showed the best performance and detected MCI with an accuracy of over 80.0%. The key features emerging from this study are stance phase, mean velocity, step length and cycle length; moreover, the major number of features selected by the wrapper belonged to the cognitive dual task, thus, supporting the close relationship between gait dysfunction and MCI in PD. Full article