Search

Highly automated vehicles (HAVs) are exposed to various kinds of internal or external data faults, which may lead to disengagements and collisions. Detecting critical faults and evaluating the fault tolerance of HAVs, especially in terms of motion planning, is of great importance before HAVs are deployed on a large scale. However, due to the curse of dimensionality, which is led by high-dimensional fault features, current studies on fault tolerance testing have been limited to evaluating the vulnerability of a single location on motion planners. In this article, we propose a fault injection (FI) tool based on the idea of adaptive design of experiments (ADOE) to expediate the fault tolerance test of motion planners. The framework utilizes surrogate models (SMs) to highlight high-probability spaces where critical faults may appear and therefore narrow down the search space to accelerate the search process. ADOE helps strategically update SMs through the search process. Results show that the proposed ADOE-based FI method has great potential in accelerating fault tolerance testing for motion planners. Out of four evaluated SMs, support vector regression performs the best. We further train a classifier based on the multilayer perceptron (MLP) to judge whether a given fault in an unexamined scenario is critical or not. The F1 score of the MLP-based classifier reaches 90%. [ABSTRACT FROM AUTHOR]

This Power quality problem is a vital issue in current-day power system with the increased penetration of distributed generation (DG) systems which greatly affects the stability and reliability of the system. This study presents a comprehensive analysis of detection and classification strategies for monitoring power quality (PQ) disturbances. Specifically, the focus is on PQ disturbances caused as a result of load switching and variations in environmental factors, such as changes in solar irradiation and wind speed. Twelve different types of PQ disturbances under various operating scenarios are considered for the detection and classification study. Twelve different statistical features are extracted by passing the disturbances through robust H-infinity filter in order to minimize the data size and computational complexities. The statistical properties of PQ disturbances are initially categorized using the Fast Generalized Fuzzy-C means (FGFCM) clustering approach. Further, the disturbances are correctly classified by pattern classification approach like support vector machines (SVM) in order to distinctly classify them. The validity of the classification analysis is reinforced by the inclusion of experimental data gathered from a prototype wind energy system and photovoltaic (PV) system setup. Again, Genetic algorithm (GA) is used for extracting optimal features and Particle swarm optimization (PSO) is applied for optimising the heuristic parameters of the SVM in order to improvise the classification accuracy under normal, noisy as well as increased penetration of wind and photovoltaic (PV) systems.

Meteorological identification and observation are crucial in production activities closely related to meteorology, such as agricultural production. Currently, some machine learning methods applied in meteorological identification exhibit low richness of semantic features and weak transferability in pre-trained models, leading to insufficient feature extraction capabilities. Additionally, these models often have relatively simple classification layers, and they tend to train as a holistic model. To address the aforementioned shortcomings, this paper proposes a mechanism of fused training, constructing a meteorological identification model based on the enhanced fusion of EVA02 and Linear Support Vector Classification (LinearSVC). The model fine-tunes the pre-trained EVA02 backbone network, fully stimulating the qualitative transformation of transfer learning’s high-level semantic abstractions and meteorological data representations. This enhances feature extraction capabilities. Additionally, the model integrates the Nystroem method with the LinearSVC algorithm for classification, further improving classification accuracy and the robustness of the model on small datasets. Through simulation experiments, the model achieves F1-scores on the public datasets MWD and WEAPD that surpass the current state-of-the-art methods by 0.75% and 9.89%, respectively, demonstrating the effectiveness of the proposed method.

Graphs classification is a relevant problem that arises in many disciplines. Using graphs directly instead of vectorization allows exploiting the intrinsic representations of the data. Support Vector Machines (SVM) is a supervised learning method based on the use of graph kernel functions used for this task. One of the problems of SVM, as the number of samples increases, is the cost of storing and solving the optimization problem related to SVM. In this work, we propose a method capable of finding a small representative subset of the whole graph data set such that an approximate solution of the SVM optimization problem can be obtained in a fraction of the time, and without significantly degrading the classification prediction error. The method is based on the use of Locality-Sensitive Hashing for projecting over the Hilbert spaces defined by appropriate graph kernels that measure similarity between the graphs. A description of the algorithm, as well as numerical results using two graph kernels (Simple Product and Random Walk) on simulated and real life data sets are presented. The numerical experiments compare the training times and the classification error between the SVM obtained with our smart sampling approach, and the SVM obtained over the complete data set or over a random sub-sample. The results offer evidence of the advantages of our proposal for solving large scale graph classification problems when using SVM.

Magnitude is one of the fundamental parameters of earthquake and also one of the essential information for earthquake early warning (EEW). Rapidly and accurately determining the high-or low-magnitude event is important for mitigating the damage of earthquake hazard. To address the problem of predicting whether an earthquake event is high magnitude ($M \ge5.5$) or low magnitude ($M < 5.5$), this letter proposes a machine learning magnitude classification framework (MCFrame) using single station data, which consists of feature extraction module and magnitude classifier, and we train the feature extraction module of MCFrame to learn the characteristics of high-magnitude ($M \ge5.5$) and low-magnitude ($M < 5.5$) earthquake events, using strong-motion records collected from the Japanese Kyoshin network (K-NET) seismic network. Then, the extracted features are used as an input of magnitude classifier to classify earthquake magnitudes. Meanwhile, we analyze the impact of three different magnitude classifiers on the performance of MCFrame, which include the deep neural network (DNN), support vector machine (SVM), and random forest (RF). We show that within 5 s after the P-wave arrival, for these three different classifiers, the magnitude classification accuracy of MCFrame is close, and the MCFrame proposed in this work has a better performance than baseline models for magnitude classification. Additionally, for the MCFrame, the accuracy of high-magnitude events ($M \ge5.5$) is more than 90%, and the accuracy of low-magnitude events ($M < 5.5$) is more than 99%. These results indicate that the MCFrame proposed in this work is significant for EEW. [ABSTRACT FROM AUTHOR]

Convolutional neural networks have demonstrated remarkable capability in extracting deep semantic features from images, leading to significant advancements in various image processing tasks. This success has also opened up new possibilities for change detection (CD) in remote sensing applications. But unlike the conventional image recognition tasks, the performance of AI models in CD heavily relies on the method used to fuse the features from two different phases of the image. The existing deep-learning-based methods for CD typically fuse features of bitemporal images using difference or concatenation techniques. However, these approaches often fail tails to prioritize potential change areas adequately and neglect the rich contextual information essential for discerning subtle changes, potentially leading to slower convergence speed and reduced accuracy. To tackle this challenge, we propose a novel feature fusion approach called feature-difference attention-based feature fusion CD network. This method aims to enhance feature fusion by incorporating a feature-difference attention-based feature fusion module, enabling a more focused analysis of change areas. Additionally, a deep-supervised attention module is implemented to leverage the deep surveillance module for cascading refinement of change areas. Furthermore, an atrous spatial pyramid pooling fast is employed to efficiently acquire multiscale object information. The proposed method is evaluated on two publicly available datasets, namely the WHU-CD and LEVIR-CD datasets. Compared with the state-of-the-art CD methods, the proposed method outperforms in all metrics, with an intersection over union of 92.49% and 85.56%, respectively. [ABSTRACT FROM AUTHOR]

Urban forests are a type of forest ecosystem characterized by a strong interconnection between human activities and the natural environment. They play a crucial role in storing above-ground carbon and regulating urban climates. Customizing low-altitude unmanned aerial vehicle (UAV) borne multispectral remote sensing (RS) systems with finely tuned sensors to accurately identify individual tree species can assist in mapping the spatial distribution of carbon stocks. The objective of this study is to investigate the robust correlation between specific spectral bands and the characteristics of urban forest species by quantifying the information encapsulated within those particular spectral bands, in order to facilitate the implementation of a low-altitude UAV-borne multispectral RS system. The proposed approach was compared with four major methods for reducing band dimensionality, namely analysis of variance, random forest, step discriminant analysis, and principal component analysis. The results of three classification models, namely support vector machine, multilayer perceptron, and boosting tree, demonstrate that the feature band combination extracted by our method consistently achieves the highest overall classification accuracy (ranging from 83.33% to 98.48%) across different types of urban forests, indicating its potential for fine identification of tree species. This research provides a fundamental basis for the dissemination and implementation of low-altitude UAV-borne multispectral RS technology in urban forestry management. [ABSTRACT FROM AUTHOR]

Human gait analysis aims to assess gait mechanics and to identify the deviations from “normal” gait patterns by using meaningful parameters extracted from gait data. As each parameter indicates different gait characteristics, a proper combination of key parameters is required to perform an overall gait assessment. Therefore, in this study, we introduced a simple gait index derived from the most important gait parameters (walking speed, maximum knee flexion angle, stride length, and stance-swing phase ratio) to quantify overall gait quality. We performed a systematic review to select the parameters and analyzed a gait dataset (120 healthy subjects) to develop the index and to determine the healthy range (0.50 – 0.67). To validate the parameter selection and to justify the defined index range, we applied a support vector machine algorithm to classify the dataset based on the selected parameters and achieved a high classification accuracy (∼95%). Also, we explored other published datasets that are in good agreement with the proposed index prediction, reinforcing the reliability and effectiveness of the developed gait index. The gait index can be used as a reference for preliminary assessment of human gait conditions and to quickly identify abnormal gait patterns and possible relation to health issues. [ABSTRACT FROM AUTHOR]

Railway point machines are the key equipment that controls the train route and affects the safety of train operation. Complex and harsh working environments lead to frequent failures, accounting for 40% of the total failures of the railway signaling system. Thus, it is an urgent task to present an intelligent fault diagnosis approach. Considering the easy acquisition and anti-interference characteristics of vibration signals, this article develops a vibration signal-based diagnosis approach. First, variational mode decomposition (VMD) is utilized for nonstationary vibration signal preprocessing, which is verified as a more effective tool than empirical mode decomposition. Then, to comprehensively characterize nonlinear fault characteristics, five kinds of entropy are extracted. To eliminate the redundant information of high-dimensional features, kernel principal component analysis is utilized for multientropy feature fusion. Experiment comparisons demonstrate the superiority of the proposed VMD preprocessing and multientropy fusion method. The diagnosis accuracies of normal-to-reverse and reverse-to-normal switching directions reach 96.57% and 99.43%, respectively, which provides theoretical support for onsite operation and maintenance staff. [ABSTRACT FROM AUTHOR]

During group interactions, we react and modulate our emotions and behaviour to the group through phenomena including emotion contagion and physiological synchrony. Previous work on emotion recognition through video/image has shown that group context information improves the classification performance. However, when using physiological data, literature mostly focuses on intrapersonal models that leave-out group information, while interpersonal models are unexplored. This paper introduces a new interpersonal Weighted Group Synchrony approach, which relies on Electrodermal Activity (EDA) and Heart-Rate Variability (HRV). We perform an analysis of synchrony metrics applied across diverse data representations (EDA and HRV morphology and features, recurrence plot, spectrogram), to identify which metrics and modalities better characterise physiological synchrony for emotion recognition. We explored two datasets (AMIGOS and K-EmoCon), covering different group sizes (4 vs dyad) and group-based activities (video-watching vs conversation). The experimental results show that integrating group information improves arousal and valence classification, across all datasets, with the exception of K-EmoCon on valence. The proposed method was able to attain mean M-F1 of $\approx$ ≈ 72.15% arousal and 81.16% valence for AMIGOS, and M-F1 of $\approx$ ≈ 52.63% arousal, 65.09% valence for K-EmoCon, surpassing previous work results for K-EmoCon on arousal, and providing a new baseline on AMIGOS for long-videos. [ABSTRACT FROM AUTHOR]

Aspect level sentiment classification aims to identify the sentiment polarity towards a particular aspect in a sentence. Previous attention-based methods generate an aspect-specific representation for each aspect and employ it to classify the sentiment polarity. However, normalized attention scores scatter over every word in the sentence, resulting in two issues. First, the attention may inherently introduce noise and downgrade the performance. Second, the opinion words may be “diluted” by other words, while the opinion feature should dominate for sentiment analysis. The issues become more severe in multi-aspect sentences. In this paper, we address the above two issues via hybrid regularizations, i.e., aspect-level and task-level regularizations. Concretely, the aspect-level regularizations constrain the attention weights to alleviate noise. Among them, orthogonal regularization is designed for multi-aspect sentences and sparse regularization is for single-aspect sentences. To extract sentiment-dominant features, task-level regularization is proposed by introducing an orthogonal auxiliary task, i.e., aspect category detection. This regularization can allocate task-oriented context information for specific downstream tasks. Extensive experimental results on three public datasets demonstrate the effectiveness of the proposed approach in both single-task and multi-task scenarios. [ABSTRACT FROM AUTHOR]

Affective computing strives to recognize a person's affective state (e.g., emotion, mood) based on what can be observed. However, electroencephalogram (EEG) and video technologies have not been widely adopted for daily life affect monitoring due to obtrusiveness and privacy concerns. Although the connection between affective states and biophysical data collected with unobtrusive wrist-worn wearables in lab settings has been established successfully, the number of studies for affect recognition in the wild is still limited, and current methods have not yet provided the accuracy necessary for robust applications. In this study, we propose a smart mood-aware emotion detection method. The proposed emotion recognition method extracts the most distinctive features from the physiological data and adds the output of the automated mood detection system as an input to improve performance. The effect of the division of self-report scales into emotion classes is also investigated. The proposed system obtained higher emotion recognition accuracies than most in-the-wild studies when we tested it with the daily life data collected from 14 participants for one week. [ABSTRACT FROM AUTHOR]

Desk workers spend long periods sitting with poor posture. This leads to back pain and musculoskeletal disorders. Current solutions to track posture use sensors or cameras that may make sitting uncomfortable and violate privacy. This letter presents a system that detects the leg position when sitting at a desk via 2-D light detection and ranging (LiDAR), and a support vector machine running on a Raspberry Pi 4. The system recognizes six common positions: legs slightly open with knees bent, legs crossed, one leg on the other leg’s knee, legs extended forward, backward, and legs parallel with an angle of ~90° between upper and lower legs. The results of an experiment involving 30 workers showed an accuracy of 99.2%. [ABSTRACT FROM AUTHOR]

According to the literature, the utilization of spatial features can significantly enhance the accuracy of hyperspectral image (HSI) classification. Fractal features are powerful measures of texture, representing the local complexity of an image. In HSI classification, textural features are typically extracted from dimensionally reduced data cubes, such as principal component analysis (PCA). However, the effectiveness of textures obtained from alternative feature extraction (FE) methods in improving classification accuracy has not been extensively investigated. This study introduces a new ensemble support vector machine classification system that combines spectral features derived from PCA, minimum noise fraction (MNF), linear discriminant analysis (LDA), and fractal features derived from these FE methods. The final results on two HSI datasets, namely, Indian Pines (IP) and Pavia University (PU), demonstrate that the proposed classification method achieves approximately 95.75% and 99.36% accuracies, outperforming several other spatial–spectral HSI classification methods. [ABSTRACT FROM AUTHOR]

In the dynamic game between jammers and radars, it is easily affected by the complex electromagnetic environment, resulting in the generation of asymmetric radar information. In this letter, a jamming effect evaluation method for asymmetric radar information is proposed from the perspective of a jammer. An online jamming effect evaluation model is constructed. An improved class-attribute contingency coefficient (ICACC) method is proposed. The feature attribute weight and inconsistency rate are used to constrain the division process of breakpoints, so that the discretization of radar data is more reasonable. At the same time, a basic belief assignment (BBA) correction method that combines probability similarity for asymmetric information is proposed. The output results of the support vector machine (SVM) are corrected to reduce the influence of incomplete radar sample data on the discriminant and evaluation results. Experiments show that the proposed method can achieve efficient jamming effect evaluation under asymmetric radar information conditions. [ABSTRACT FROM AUTHOR]

Classification of very high-resolution (VHR) aerial remote sensing (RS) images is a well-established research area in the RS community as it provides valuable spatial information for decision-making. Existing works on VHR aerial RS image classification produce an excellent classification performance; nevertheless, they have a limited capability to well-represent VHR RS images having complex and small objects, thereby leading to performance instability. As such, we propose a novel plug-and-play multiscale attention feature extraction block (MSAFEB) based on multiscale (MS) convolution at two levels with skip connection, producing discriminative/salient information at a deeper/finer level. The experimental study on two benchmark VHR aerial RS image datasets (AID and NWPU) demonstrates that our proposal achieves a stable/consistent performance (minimum standard deviation (SD) of 0.002) and competent overall classification performance (AID: 95.85% and NWPU: 94.09%). [ABSTRACT FROM AUTHOR]

In the visual object tracking via unmanned aerial vehicle (UAV), the correlation filtering (CF) is one of the mainstream methods for the reason that optimizing filters by circulant samples facilitates the calculation. The prevalent discriminative CF (DCF) frameworks credited to a ridge regression model followed by various regularizations focus on regressing to a fixed Gaussian label, which may easily induce overfitting. To these concerns, we integrate the hinge-squared loss (HSL) of structured support vector machine (SVM) with the CF model so as to attain a dynamical label which regresses the difference between target and background samples and thus enhances the robustness in tracking. Moreover, we assume weighted channels to augment the discrimination of HSL, thus indicating it has as good extensibility as the ridge regression in usual CF frameworks. The proposed method channel-weighted structured CF (CWSCF) has achieved excellent performance compared with other methods in mainstream UAV tracking benchmarks. [ABSTRACT FROM AUTHOR]

Micro-motion feature vectors of space targets are usually unevenly and multicluster distributed, which limits the performance of the traditional radar anomaly detection methods. To solve this problem, a novel detection method of radar space target abnormal motion method via local density peaks (LDPs) and micro-motion feature is proposed in this letter. First, two discriminative micro-motion features are extracted from the radar echoes to construct a 2-D feature space. Then the abnormal motion detector is derived by classifying the feature vectors into different clusters according to the LDPs and minimum spanning tree clustering (LDP-MST) and solving for the decision thresholds of each cluster with the LDPs, neighbors, and some preset false alarm rates. Electromagnetic simulation experiment results demonstrate that the detection rate of the proposed method is 2.49%, 5.26%, 9.63%, 15.37%, 27.99%, and 49.45% higher than six state-of-art methods, respectively, when the false alarm rate is 5%. [ABSTRACT FROM AUTHOR]

Supervised hyperspectral classification is one of the important applications in hyperspectral remote sensing. Currently, the enormous complexity of the hyperspectral data makes hyperspectral feature extraction a very challenging job. In this letter, we propose a feature extraction approach by developing a 2-D compact variational mode decomposition (2-D-C-VMD) for the supervised hyperspectral images (HSIs) classification. Specifically, the HSIs are first decomposed into a series of mode components using 2-D-C-VMD. Then for the subsequent supervised classification, a collection of feature samples are created by elaborating the collected pattern components. Finally, a support vector machine (SVM) is adopted as the classifier to validate the aforementioned feature extraction algorithm. According to experimental findings, this approach obtained a high level of classification accuracy with a compact feature extraction process and the time for feature extraction is also shortened. [ABSTRACT FROM AUTHOR]

Quad-pol synthetic aperture radar (SAR) is one of the most effective approaches for dry and wet snow identification data, but its applicability is limited by the high cost of quad-pol SAR data. Dual-pol SAR such as Sentinel-1 has larger spatial coverage, longer time sequences, and freely accessible data, but there is still a highly uncertainty in dual-pol SAR to distinguish dry and wet snow due to limited polarimetric information. In this study, a pixel neighborhood-based snow identification algorithm was developed and verified using dual-pol C-band SAR data in Northern Xinjiang, China. A total of six decomposed parameters were obtained to characterize the polarimetric information of dual-pol SAR data by modifying the H- $\alpha $ decomposition applicable to dual-pol SAR data. In the case of limited training samples, polarimetric features that were most sensitive to snow identification were selected as the optimal features for support vector machine (SVM), and the result derived from SVM was employed as the initial labels of Markov random field (MRF) model to separate dry and wet snow using iterative conditional mode (ICM). Then, the proposed algorithm, dual-pol SVM-MRF (DSVM-MRF), was validated and compared with previously published methods. The results show that the DSVM-MRF acquires the superior snow recognition with the overall accuracy (OA) and Kappa coefficient of 84.5% and 0.58%, respectively. [ABSTRACT FROM AUTHOR]

Hyperspectral image (HSI) classification is a hot topic in the remote sensing community. However, it is challenging to fully use spatial–spectral information for HSI classification due to the high dimensionality of the data, high intraclass variability, and the limited availability of training samples. To deal with these issues, we propose a novel feature extraction method called nonlocal correntropy matrix (NLCM) representation in this letter. NLCM can characterize the spectral correlation and effectively extract discriminative features for HSI classification. We verify the effectiveness of the proposed method on two widely used datasets. The results show that NLCM performs better than the state-of-the-art methods, especially when the training set size is small. Furthermore, the experimental results also demonstrate that the proposed method outperforms compared methods significantly when the land covers are complex and with irregular distributions. [ABSTRACT FROM AUTHOR]

The convolutional sparse coding (CSC) can learn shift-invariant convolution kernels. In deep convolutional neural networks, it takes a lot of time to train the convolution kernels. In this letter, a deep two-stage CSC network (DTCSCNet) is proposed, which can be used to simultaneously extract spatial features and spectral features from hyperspectral image (HSI) without back propagation and fine-tuning process, thus saving a lot of time. Furthermore, to further improve the performance of the network, we incorporate multiscale information. After deep feature extraction using DTCSCNet, we further investigate the classification performance of different classifiers on the extracted features. Experimental results show that the proposed method can obtain better classification performance compared with some closely related HSI classification methods. [ABSTRACT FROM AUTHOR]

There is growing interest in affective computing for the representation and prediction of emotions along ordinal scales. However, the term ordinal emotion label has been used to refer to both absolute notions such as low or high arousal, as well as relation notions such as arousal is higher at one instance compared to another. In this paper, we introduce the terminology absolute and relative ordinal labels to make this distinction clear and investigate both with a view to integrate them and exploit their complementary nature. We propose a Markovian framework referred to as Dynamic Ordinal Markov Model (DOMM) that makes use of both absolute and relative ordinal information, to improve speech based ordinal emotion prediction. Finally, the proposed framework is validated on two speech corpora commonly used in affective computing, the RECOLA and the IEMOCAP databases, across a range of system configurations. The results consistently indicate that integrating relative ordinal information improves absolute ordinal emotion prediction. [ABSTRACT FROM AUTHOR]

Electromagnetic (EM) brakes are widely applied in many automation and control fields. Assembly quality defects in EM brakes can significantly reduce braking performance and pose potential safety risks. The primary challenge in assembly quality control is the ability to predict product quality and prevent potential failures. In view of the lack of quality information interaction and real-time feedback in the traditional quality prediction process, an intelligent predictive quality control model for the assembly process of EM brakes is proposed in this paper. First, we introduce an assembly quality control model based on digital twin (DT) technology. Second, a data-driven quality prediction algorithm is developed using least-squares support vector regression (LSSVR) and improved particle swarm optimisation (IPSO). To reduce the complexity of the prediction process, grey relational analysis (GRA) is used to analyse and extract key quality characteristics data. EM friction brakes are employed as an example to analyse the product assembly process, and the results provide valuable insights into the implementation of DT technology. Finally, the feasibility of the developed prediction model is verified with real assembly datasets. The results demonstrate the effectiveness and precision of the DT model and the GRA-IPSO-LSSVR method in predicting assembly quality.

Control systems of robotic prostheses should be designed to decode the users’ intent to start, stop, or change locomotion; and to select the suitable control strategy, accordingly. This paper describes a locomotion mode recognition algorithm based on adaptive Dynamic Movement Primitive models used as locomotion templates. The models take foot-ground contact information and thigh roll angle, measured by an inertial measurement unit, for generating continuous model variables to extract features for a set of Support Vector Machines. The proposed algorithm was tested offline on data acquired from 10 intact subjects and 1 subject with transtibial amputation, in ground-level walking and stair ascending/descending activities. Following subject-specific training, results on intact subjects showed that the algorithm can classify initiatory and steady-state steps with up to 100.00% median accuracy medially at 28.45% and 27.40% of the swing phase, respectively. While the transitory steps were classified with up to 87.30% median accuracy medially at 90.54% of the swing phase. Results with data of the transtibial amputee showed that the algorithm classified initiatory, steady-state, and transitory steps with up to 92.59%, 100%, and 93.10% median accuracies medially at 19.48%, 51.47%, and 93.33% of the swing phase, respectively. The results support the feasibility of this approach in robotic prosthesis control.

Devices capable of tracking the user’s gaze have become significantly more affordable over the past few years, thus broadening their application, including in-home and office computers and various customer service equipment. Although such devices have comparatively low operating frequencies and limited resolution, they are sufficient to supplement or replace classic input interfaces, such as the keyboard and mouse. The biometric application we researched verifies a user’s identity based on parameters acquired with a low-cost eye tracker. The use of the eye-tracking device in bank booths has many advantages, including the fact that eye trackers are contactless devices, which, especially in the light of the Covid pandemic, has increasing importance, in addition to providing a solution for confirming the liveness of the user. This paper describes an experiment in which 20 features extracted from eye movement data related mainly to saccades and fixations are used as a complementary biometric modality to authenticate clients at banking kiosks. Data were collected from 39 subjects while operating a banking system using engineered biometric kiosk prototypes. Authentication performance employing eye-movement tracking and parameterizing was compared for two classifiers: Support Vector Machine (SVM) and eXtreme Gradient Boosting (XGBoost). The results showed that the XGBoost-based classifier outperformed the SVM-based one regarding equal error rates (6.76% to 8% vs. 16.21 to 18.78%). Similar differences were obtained for true acceptance rates at different false acceptance rates (0.1 and 0.01), where the SVM-based classifier achieved a maximum of 81.08% and the XGBoost-based achieved 98.65%. Finally, prospects for the broader application of eye movement tracking as a biometric modality are discussed.

Smart contracts based on blockchain are widely used in finance, management, Internet of Things, healthcare, and other fields. However, with the rapid development of smart contracts, the corresponding security vulnerability attack cases occur frequently. Existing Ethereum smart contract vulnerability detection tools based on static analysis techniques rely too much on expert rules, for this reason, this paper proposes an Ethereum smart contract vulnerability detection method SCSVM based on support vector machine technology. A representation of smart contracts is constructed based on the word-to-vector technique, the features of Ethereum smart contracts are extracted based on the support vector machine technique, and these features are combined to identify vulnerabilities. Experiments on Smartbugs and Smartbugs-wild show that SCSVM is significantly effective. It achieves a detection accuracy of 87.51%, outperforming five typical static analysis vulnerability detection tools in terms of F1-score. To alleviate the problems of deep learning methods over-relying on large-scale data to train models and collecting a large number of smart contract attack samples in a short period, this paper proposes a basic learner-meta-learner framework, SCLMF. solc-based acquisition of the bytecode of Ethereum smart contract Solidity, on which smart contract representations are constructed via Python and the use of SCLMF for vulnerability detection. The experiments on WScrawlD show that SCLMF has a certain detection effect. Also, to further verify the effectiveness of SCLMF, experiments were conducted on Omniglot, and the detection accuracy was 96.7% and 98.5% under 5-way 1-shot and 5-way 5-shot conditions, respectively, which exceeded Memory-Augmented Neural Networks and CONVOLUTIONAL SIAMESE NETS. In summary, the experiments proved the effectiveness of SCSVM and SCLMF in Ethereum smart contract vulnerability detection.

The importance of care robots is growing owing to the increasing numbers of the elderly and the shortage of caregivers. For robots to automatically assist the movements of the elderly users, the state of their pose should be estimated. Hence, we proposed a method for estimating the user’s state using candidate positions of the center of gravity of the robot user using a few sensors. In this method, sensor outputs in all states and movements of the user are collected in advance, and a state estimation and abnormality detection model created by machine learning using the relationship between the outputs and state is used as training data. However, it is difficult to collect training data pertaining to elderly people, before use, for this purpose. Therefore, a database of human actions is created for use as the training data. Rich databases of such actions performed sans robots already exist, but not where care robots are used. Conventional methods for simulating human body movements to estimate human states require actual data and expert coordination and do not address detailed state estimation for physical support. Therefore, this study proposed a method for generating training data using a human link model, which enables the creation of a model that can estimate the state of a robot user without requiring the elderly user to stand up and perform other robot-assisted actions before use. Training data is generated using a link model and candidate centers of gravity, a simple method by which the state of the care robot user can be estimated, and physical support for standing, walking, and sitting can be provided. The effectiveness of the state estimation using the link model generated training data is verified off-line using sensor data independently obtained from the actual movements of the robot and user and also through experiments using an actual care robot. The results validated the state estimation since the time error was sufficiently short (0.35–1.85s), and the experiment confirmed that the robot could realize assistive actions with 90% accuracy.

In this paper, we study the classification problem by estimating the conditional probability function of the given data. Different from the traditional expected risk estimation theory on empirical data, we calculate the probability via Fredholm equation, this leads to estimate the distribution of the data. Based on the Fredholm equation, a new expected risk estimation theory by estimating the cumulative distribution function is presented. The main characteristic of the new expected risk estimation is to measure the risk on the distribution of the input space. The corresponding empirical risk estimation is also presented, and an $\varepsilon $ -insensitive $L_{1}$ cumulative support vector machines ( $\varepsilon -L_{1}$ VSVM) is proposed by introducing an insensitive loss. It is worth mentioning that the classification model and the classification evaluation indicator based on the new mechanism are different from the traditional one. Experimental results show the effectiveness of the proposed $\varepsilon -L_{1}$ VSVM and the corresponding cumulative distribution function indicator on validity and interpretability of small data classification.

A substantial body of evidence on traffic accidents indicates that distracted driving is a major cause of accidents. To investigate the relationship between driver performance in both normal and distracted states and to discriminate which state a driver is in based on driving performance, a simulator was used to simulate drivers driving in an urban road environment, and a distraction task was designed to motivate drivers to enter a cognitive distraction state. Data were collected from 80 drivers in each of the two driving states, and a database was built. Deep neural network (DNN) is a multilayer perceptron structure, relative to the defects of other classical algorithms with unclear classification thresholds. DNN can be utilized to process high-dimensional data through its neural node node-linked structure. Combining the Gray Wolf algorithm (GWO) for the initial weights and thresholds of the DNN network as a whole, a four-layer network structure is built to predict the test samples, while support vector machines (SVM) and random forest as controls. The results show that the accuracy of DNN for predicting test samples is 95.13%, that of SVM is 77.56%, and that of random forest is 72.32%. F1 score of the DNN model is higher than that of SVM, and the detection effect is better and can be applied to detect the driving status of drivers.

Autism is a prevalent neurodevelopmental disorder characterized by impairments in social and communicative behaviors. Possible connections between autism and facial expression recognition have recently been studied in the literature. However, most works are based on facial images or short videos. Few works aim at Autism Diagnostic Observation Schedule (ADOS) videos due to their complexity (e.g., interaction between interviewer and interviewee) and length (e.g., usually last for hours). In this paper, we attempt to fill this gap by developing a novel discriminative few shot learning method to analyze hour-long video data and exploring the fusion of facial dynamics for the trait classification of ASD. Leveraging well-established computer vision tools from spatio-temporal feature extraction and marginal fisher analysis to few-shot learning and scene-level fusion, we have constructed a three-category system to classify an individual into Autism, Autism Spectrum, and Non-Spectrum. For the first time, we have shown that certain interview scenes carry more discriminative information for ASD trait classification than others. Experimental results are reported to demonstrate the potential of the proposed automatic ASD trait classification system (achieving 91.72% accuracy on the Caltech ADOS video dataset) and the benefits of few-shot learning and scene-level fusion strategy by extensive ablation studies. [ABSTRACT FROM AUTHOR]

Device heterogeneity can cause a detrimental impact on the classification of healthcare data. In this work, we propose the Maximum Difference-based Heterogeneity Mitigation (MDHM) method to address device heterogeneity. Mitigating heterogeneity increases the reliability of using multiple devices from different manufacturers for measuring a particular physiological signal. Further, we propose an attention-based bilevel GRU (Gated Recurrent Unit) model, abbreviated as AT2GRU, to classify multi-modal healthcare time-series data for human emotion recognition. The physiological signals of Electroencephalogram (EEG) and Electrocardiogram (ECG) for twenty-three persons are leveraged from the DREAMER dataset for emotion recognition. Also, from the DEAP dataset, the biosignals namely EEG, Galvanic Skin Response (GSR), Respiration Amplitude (RA), Skin Temperature (ST), Blood Volume (BV), Electromyogram (EMG) and Electrooculogram (EOG) of thirty-two persons are used for emotion recognition. The EEG and the other biosignals are denoised by the wavelet filters for enhancing the model's classification accuracy. A multi-class classification is carried out considering valence, arousal, and dominance for each person in the datasets. The classification accuracy is validated against the self-assessment obtained from the respective person after watching a movie/video. The proposed AT2GRU model surpasses the other sequential models namely Long Short Term Memory (LSTM) and GRU in performance. [ABSTRACT FROM AUTHOR]

Emotion understanding represents a core aspect of human communication. Our social behaviours are closely linked to expressing our emotions and understanding others’ emotional and mental states through social signals. The majority of the existing work proceeds by extracting meaningful features from each modality and applying fusion techniques either at a feature level or decision level. However, these techniques are incapable of translating the constant talk and feedback between different modalities. Such constant talk is particularly important in continuous emotion recognition, where one modality can predict, enhance and complement the other. This article proposes three multisensory integration models, based on different pathways of multisensory integration in the brain; that is, integration by convergence, early cross-modal enhancement, and integration through neural synchrony. The proposed models are designed and implemented using third-generation neural networks, Spiking Neural Networks (SNN). The models are evaluated using widely adopted, third-party datasets and compared to state-of-the-art multimodal fusion techniques, such as early, late and deep learning fusion. Evaluation results show that the three proposed models have achieved comparable results to the state-of-the-art supervised learning techniques. More importantly, this article demonstrates plausible ways to translate constant talk between modalities during the training phase, which also brings advantages in generalisation and robustness to noise. [ABSTRACT FROM AUTHOR]

Emotion plays an important role in human cognition and behavior. How to recognize emotions based on physiological signals has attracted increasing research interests worldwide up to date. Both traditional eastern medicine and modern western medicine have confirmed the existence of relationship between human emotionality and heart activity. However, in practice, emotion recognition only using Electrocardiogram (ECG) signals seems quite challenging, not only due to the severe noise interferences and the serious data variations, but also because of the ambiguous relationship between emotional states and ECG data. Such difficulty can even be compared to finding needles in a haystack. As an innovative endeavor to deal with the issue of only-ECG-based emotion recognition, this paper has proposed a novel solution from the perspective of weak signal classification. The proposed solution extracts the static-dynamic representation under the principle of Yin-Yang balance from the heartbeat data, and then utilizes the set-based collaborative measurement upon the thought of data coopetition to classify these features for recognizing emotions. Experimental results have demonstrated the effectiveness, efficiency and adaptiveness of the solution for uncovering the potential relationship between emotions and ECG. Thus, this proposal has also illuminated a promising research direction for the general problem of weak signal classification. [ABSTRACT FROM AUTHOR]

Generally, discontinuous pulsewidth modulation (DPWM) strategy used in single Vienna rectifier will result in large current ripple and neutral-point fluctuation. When applied in interleaved rectifier, the issue of neutral-point voltage still exists. Besides, the input current around zero-crossing point is distorted due to the characteristics of unidirectional topology. Hence, a novel hybrid DPWM (H-DPWM) strategy for parallel Vienna rectifier is proposed in this article to reduce the distortion, current ripple, and neutral-point fluctuation. In this article, the characteristics of current ripple and neutral-point fluctuation with traditional DPWM (T-DPWM) are clarified at first. Then, the quantitative analyses on current ripple and neutral-point fluctuation with T-DPWM, classical space vector modulation, and optimized DPWM (O-DPWM) are presented in detail. To solve the issue of current distortion, the H-DPWM is proposed based on O-DPWM. Furthermore, the switching losses of these modulations are evaluated. At last, the analyses and performance of the proposed H-DPWM are verified by the experiments. [ABSTRACT FROM AUTHOR]

Multiphase drives (MPD) in general and six-phase in particular have been gaining popularity in many industries, which calls for a proper multiphase inverter (MPI) design. For voltage source inverter (VSI)-fed drives, the knowledge of voltage and current stresses on the dc-link are imperative for input capacitor sizing. To this end, the voltage and current stresses on the dc-link capacitor in two-level six-phase VSIs are examined thoroughly in this article for two configurations of load/winding spatial distribution: symmetric and asymmetric. First, the harmonic spectrum of the input dc current of each inverter is analyzed in detail by benchmarking them against the conventional three-phase VSI, to precisely establish the dc-capacitor requirement reduction in six-phase counterparts. Second, analytical formulae for the dc-link capacitor voltage ripples are derived for both configurations. Third, simple formulae for dc-capacitor sizing for six-phase VSIs with different load configurations are provided. The accuracy of the derived formulae is verified by simulation and experimental testing at various power factors and modulation depth. It is found that six-phase VSI supplying symmetric and asymmetric loads reaps 10% and 7% lower dc-link current ripples, respectively. Hence, six-phase symmetric loads yield the smallest capacitor size. [ABSTRACT FROM AUTHOR]

The coupling between the rectifier and inverter makes the modulation of the indirect matrix converter (IMC) relatively complicated. A novel modulation method is proposed in this letter, where a space vector modulation without a zero-voltage vector, which is only responsible for adjusting the phase of the output voltage vector, is used for the inverter, and a novel delta-sigma method (ΔΣM) to modulate the input current vector phase and dc-link voltage is adopted by the rectifier. Duty cycles of the inverter are obtained by looking up the table of reference output voltages and updated at a 5-kHz carrier cycle, while switching states of the rectifier change once per cycle at a 40-kHz sampling frequency. By decoupling the rectifier and inverter, modulation of the IMC is greatly simplified. Methods that optimize the selection of rectifier switching states under various output/input voltage transfer ratios are also discussed. Simulations and experimental results are presented to validate the proposed method. [ABSTRACT FROM AUTHOR]

In this article, we propose a monitoring procedure based on a multilayer neural network with multivalued neurons (MLMVN) capable of preventing catastrophic failures of dc–dc converters. The neural classifier allows both the detection of any malfunction and its localization. Thanks to the low computational complexity, the proposed method operates online, estimating the deviations of the passive components from their nominal values: this allows control strategies to be promptly adopted and operation of the dc–dc converter to be kept in high efficiency and reliability conditions. Since measuring the voltage and current on each component increases the complexity of the system, a testability analysis is proposed with the aim of identifying the minimum number of measurements needed to distinguish the classes of failure. To make the testability phase easier and more intuitive, a graphical representation is proposed. As a case study, prognostic analysis has been applied to prevent catastrophic failures in a synchronous Zeta converter. Several fault conditions have been analyzed through simulations and experimental tests. The obtained results confirm the ability of the proposed method to prevent failures and, also, show that the application of MLMVN results in a better performance than classic solutions available in the literature, such as support vector machine. [ABSTRACT FROM AUTHOR]

One major problem limiting the practicality of a brain-computer interface (BCI) is the need for large amount of labeled data to calibrate its classification model. Although the effectiveness of transfer learning (TL) for conquering this problem has been evidenced by many studies, a highly recognized approach has not yet been established. In this paper, we propose a Euclidean alignment (EA)-based intra- and inter-subject common spatial pattern (EA-IISCSP) algorithm for estimating four spatial filters, which aim at exploiting intra- and inter-subject similarities and variability to enhance the robustness of feature signals. Based on the algorithm, a TL-based classification framework was developed for enhancing the performance of motor imagery (MI) BCIs, in which the feature vector extracted by each filter is dimensionally reduced by linear discriminant analysis (LDA) and a support vector machine (SVM) is used for classification. The performance of the proposed algorithm was evaluated on two MI data sets and compared with that of three state-of-the-art TL algorithms. Experimental results showed that the proposed algorithm significantly outperforms these competing algorithms for training trials per class from 15 to 50 and can reduce the amount of training data while maintaining an acceptable accuracy, thus facilitating the practical application of MI-based BCIs. [ABSTRACT FROM AUTHOR]

For patients who need lower-limb kinetism rehabilitation training, this paper proposes an effective data-driven approach seeking the design of 1-degree-of-freedom (DOF) six-bar rehab mechanism through gait prediction by body parameters. First, gait trajectories from 79 healthy volunteers are collected along with their body parameters. Then, the normalized gait samples are clustered and regressed into a limited number of representative trajectories with K-means algorithm, and the cluster index is recorded as the label for each trajectory. Next, a genetic-algorithm-optimized support vector machine method is adopted to establish a classifier for the trajectories, obtaining the correspondence between body parameters and cluster labels of gait trajectories. As a result, once a group of body parameters are input into the classifier, the suitable gait trajectory can be predicted for the specific patient. A GA-BFGS algorithm is developed for 1-DOF six-bar mechanism synthesis and a GUI design software is presented that shows how the data-driven design process is realized. The novelty of this paper is using clustering and prediction technique to accomplish the patient-mechanism matching, so that simple, low-priced 1-DOF mechanisms could be adopted for large number of various patients without expensive customized design for each individual. In the end, a gait rehab device design example is provided, and a prototype device driven by a constant speed motor is presented, which illustrates the feasibility of the proposed method. [ABSTRACT FROM AUTHOR]

Motor imagery-based brain-computer interfaces (MI-BCIs) features are generally extracted from a wide fixed frequency band and time window of EEG signal. The performance suffers from individual differences in corresponding time to MI tasks. In order to solve the problem, in this study, we propose a novel method named Riemannian sparse optimization and Dempster-Shafer fusion of multi-time-frequency patterns (RSODSF) to enhance the decoding efficiency. First, we effectively combine the Riemannian geometry of the spatial covariance matrix with sparse optimization to extract more robust and distinct features. Second, the Dempster-Shafer theory is introduced and used to fuse each time window after sparse optimization of Riemannian features. Besides, the probabilistic values of the support vector machine (SVM) are obtained and transformed to effectively fuse multiple classifiers to leverage potential soft information of multiple trained SVM. The open-access BCI Competition IV dataset IIa and Competition III dataset IIIa are employed to evaluate the performance of the proposed RSODSF. It achieves higher average accuracy (89.7% and 96.8%) than state-of-the-art methods. The improvement over the common spatial patterns (SFBCSP) are respectively 9.9% and 12.4% (p < 0.01, paired t-test). These results show that our proposed RSODSF method is a promising candidate for the performance improvement of MI-BCI. [ABSTRACT FROM AUTHOR]

This article proposes a space vector modulated model predictive control (SVM $^{2}$ PC) for voltage source converters. It overcomes limitations of standard finite control set model predictive controller, ensuring fast dynamic response, fixed switching frequency, and low computational burden. It combines control and modulation in a convex optimization problem with affine inequality constraints to minimize the tracking errors of the output variables. First, the sector of the space vector diagram where the unconstrained solution lies is identified. Then, using the Karush–Kuhn–Tucker conditions, feasible duty cycles are computed for both linear and overmodulation regions. The proposed SVM $^{2}$ PC is extended to multilevel converters. Hardware in the loop results are presented for three different case studies: 1) Grid-forming inverter with LC filter; 2) two-level grid-following inverter with LCL filter; and 3) grid-following neutral point clamped inverter with LCL filter. Experimental results demonstrate that the proposed SVM $^{2}$ PC provides good performance, fast response, and low computational burden when compared with previously reported alternatives. [ABSTRACT FROM AUTHOR]

Target recognition in SAR images was widely studied over the years. Most of these works were usually based on the assumption that the targets in the test set belong to a limited set of classes. In the practical scenarios, it is common to encounter various kinds of new targets. It is therefore more meaningful to study target recognition in open-world environments. In these scenes, it is needed to reject the unknown classes while maintain the classification performance on known classes. In the past years, few works were devoted to open set target recognition. Though the detection performance on unknown targets can be improved to a certain extent in the preceding works, most detection schemes are independent of a pretrained feature extractor, leading to potential open space risks. Besides, the model architectures are complicated, resulting in huge computational cost. To solve these problems, a family of new methods for open set target recognition is proposed. Targets indistinguishable from known classes are constructed by random sampling combination strategy. They are further sent into the classifier for feature learning. The original open-world environment is then transformed into a closed-world environment containing the unknown class. Moreover, the special implication of generated unknown targets is highlighted and used to realize unknown detection. Extensive experimental results on the MSTAR benchmark dataset illustrate the effectiveness of the proposed methods. [ABSTRACT FROM AUTHOR]

Forecasting the future average speed fluctuation is conducive to timely perceiving the degree of urban road congestion. Nevertheless, a new challenge has arisen in accurately predicting the spot speed on an urban road network area composed by multiple road segments. To surmount this, an intelligent method for spot speed estimation, namely, the fusion deep learning model with sequence-to-sequence structure (Seq2Seq-FDL), is proposed. It is a sequence-to-sequence learning structure that covers the encoding context vector. The unidirectional convolutional long short-term memory (ConvLSTM) and bidirectional ConvLSTM networks are used to capture traffic short-term and periodic spatiotemporal synchronization characteristics, respectively. Finally, the prediction results are computed by linear feature fusion. Furthermore, the black-box effect of the proposed model is explained by the least absolute shrinkage and selection operator regression model based on Granger causality, and the rationality of the model is verified with an example of spot speed in a road network of Guangzhou. The experimental results were shown to indicate that the traffic spatiotemporal characteristics significantly contributed to the prediction effect of the Seq2Seq-FDL model. The prediction accuracy was superior to a convolutional neural network-connected LSTM network and a ConvLSTM network with or without a bidirectional LSTM; the proposed model presented good applicability in the multiple-time-step prediction of spot speed for multiple road segments. [ABSTRACT FROM AUTHOR]

High-quality corpora have been very scarce for the text emotion research. Existing corpora with multi-label emotion annotations have been either too small or too class-biased to properly support a supervised emotion learning. In this article, we propose a novel active learning method for efficiently instructing the human annotations for a less-biased and high-quality multi-label emotion corpus. Specifically, to compensate annotation for the minority-class examples, we propose a complementary sampling strategy based on unlabeled resources by measuring a probabilistic distance between the expected emotion label distribution in a temporary corpus and an uniform distribution. Qualitative evaluations are also given to the unlabeled examples, in which we evaluate the model uncertainties for multi-label emotion predictions, their syntactic representativeness for the other unlabeled examples, and their diverseness to the labeled examples, for a high-quality sampling. Through active learning, a supervised emotion classifier gets progressively improved by learning from these new examples. Experiment results suggest that by following these sampling strategies we can develop a corpus of high-quality examples with significantly relieved bias for emotion classes. Compared to the learning procedures based on traditional active learning algorithms, our learning procedure indicates the most efficient learning curve and estimates the best multi-label emotion predictions. [ABSTRACT FROM AUTHOR]

This work proposes a hybrid space vector modulation (HSVM) scheme for multiport hybrid converters (MHCs). Moreover, the impact of shifting the auxiliary currents from the main ones is proposed and investigated in order to enhance the converter efficiency. The proposed operational schemes have been implemented in a three-phase 5 kW MHC prototype. It is shown that the proposed HSVM scheme can improve the MHC efficiency by 0.3% at full load with respect to space vector modulation. At partial loads, the improvement is even more significant, reaching +0.7% at 30% of the rated power. A further 0.15% increase in efficiency at full power can be achieved by a 180 $^{\circ }$ phase shifting of the auxiliary currents with respect to the main terminal currents, reaching a peak efficiency of 98.5%. [ABSTRACT FROM AUTHOR]

Available data in machine learning applications is becoming increasingly complex, due to higher dimensionality and difficult classes. There exists a wide variety of approaches to measuring complexity of labeled data, according to class overlap, separability or boundary shapes, as well as group morphology. Many techniques can transform the data in order to find better features, but few focus on specifically reducing data complexity. Most data transformation methods mainly treat the dimensionality aspect, leaving aside the available information within class labels which can be useful when classes are somehow complex. This paper proposes an autoencoder-based approach to complexity reduction, using class labels in order to inform the loss function about the adequacy of the generated variables. This leads to three different new feature learners, Scorer, Skaler and Slicer. They are based on Fisher’s discriminant ratio, the Kullback-Leibler divergence and least-squares support vector machines, respectively. They can be applied as a preprocessing stage for a binary classification problem. A thorough experimentation across a collection of 27 datasets and a range of complexity and classification metrics shows that class-informed autoencoders perform better than 4 other popular unsupervised feature extraction techniques, especially when the final objective is using the data for a classification task. [ABSTRACT FROM AUTHOR]

Convolutional neural networks (CNNs) have achieved remarkable performance in driver drowsiness detection based on the extraction of deep features of drivers’ faces. However, the performance of driver drowsiness detection methods decreases sharply when complications, such as illumination changes in the cab, occlusions and shadows on the driver’s face, and variations in the driver’s head pose, occur. In addition, current driver drowsiness detection methods are not capable of distinguishing between driver states, such as talking versus yawning or blinking versus closing eyes. Therefore, technical challenges remain in driver drowsiness detection. In this article, we propose a novel and robust two-stream spatial–temporal graph convolutional network (2s-STGCN) for driver drowsiness detection to solve the above-mentioned challenges. To take advantage of the spatial and temporal features of the input data, we use a facial landmark detection method to extract the driver’s facial landmarks from real-time videos and then obtain the driver drowsiness detection result by 2s-STGCN. Unlike existing methods, our proposed method uses videos rather than consecutive video frames as processing units. This is the first effort to exploit these processing units in the field of driver drowsiness detection. Moreover, the two-stream framework not only models both the spatial and temporal features but also models both the first-order and second-order information simultaneously, thereby notably improving driver drowsiness detection. Extensive experiments have been performed on the yawn detection dataset (YawDD) and the National TsingHua University drowsy driver detection (NTHU-DDD) dataset. The experimental results validate the feasibility of the proposed method. This method achieves an average accuracy of 93.4% on the YawDD dataset and an average accuracy of 92.7% on the evaluation set of the NTHU-DDD dataset. [ABSTRACT FROM AUTHOR]

Detection of mild cognitive impairment (MCI) and dementia (DEM) is an important topic because, unless it is treated early, MCI can progress to DEM, which is an untreatable disease. This article proposes a timed-up-and-go (TUG) task features analysis and classification of MCI and DEM using inertial measurement units (IMUs) in wearable devices. Our goal is to create a generalized model that can be used for preclinical screening. As a result, rather than classifying only one subtype of DEM, such as Alzheimer’s disease (AD) or Parkinson’s disease (PD), we classify all subtypes as DEM. We also utilize feature selection methods on features from TUG tasks to optimize the MCI and DEM classification performance. From the results, our generalized model can outperform other works in normal control (NC)-MCI&DEM classification with an accuracy of 86.94% and sensitivity of 97.40%. For NC-DEM classification, the performance of our generalized model is slightly lower than that of specific-subtype models (e.g., NC versus AD). However, our generalized model can outperform the specific-subtype models when using a diverse variety of subtypes. It is a reasonable tradeoff, and it can be a good first step toward a future where the patient can preclinically self-screening the cognitive impairments using wearable devices in free-living environments. This could allow patients to notice the cognitive impairment early on and seek a comprehensive diagnosis from a doctor. [ABSTRACT FROM AUTHOR]

Series arc fault (SAF) is the main cause of the electrical fire and other safety accidents. In the industrial field, now, it is still a lack of effective solutions to accurately identify the SAF that occurred in the three-phase motor with the frequency converter load (TPMFCL) circuit. The transient SAF experiments and two-phase SAF experiments in the TPMFCL circuit were carried out. A novel SAF identification method based on variational mode decomposition (VMD) of a single-phase current signal and entropy-based features was proposed. First, the current signal is decomposed by VMD. The intrinsic mode function (IMF) components whose central frequency is within the frequency band between 0 and 650 Hz are obtained. Second, the sample entropy and the energy entropy of those IMFs are calculated and used as the fault features. Finally, the SAF is identified by using a support vector machine optimized by firefly algorithm (FA-SVM). Both the identification accuracy and calculation time of the method were systematically tested by changing several different test conditions. When the sampling frequency is 2 kHz and the sample length is four current cycles, the method can accurately and quickly identify the SAF that occurred in the TPMFCL circuit even under complicated harmonic power supply conditions. The conclusions can be applied to develop the industrial arc fault circuit interrupter (AFCI) used in the TPMFCL circuit. [ABSTRACT FROM AUTHOR]