Search

Emotion recognition is an important part of human-computer interaction and human communication information is multi-model. Despite advancements in emotion recognition models, certain challenges persist. The first problem (Problem 1) pertains to the predominant focus in existing research on mining the interaction information between modes and the context information in the dialogue process but neglects to mine the role information between multi-model states and context information in the dialogue process. The second problem (Problem 2) is in the context information of the dialogue where the information is not completely transmitted in a temporal structure. Aiming at these two problems, we propose a multi-model fusion dialogue graph attention network (MM DialogueGAT). To solve the problem 1, the bidirectional GRU mechanism is used to extract the information from each model. In the multi-model information fusion problem, different model configurations and different combinations use the cross-model multi-head attention mechanism to establish a multi-head attention layer. Text, video and audio information are used as the main and auxiliary modes for information fusion. To solve the problem 2, in the temporal context information extraction problem, the GAT graph structure is used to capture the context information in the mode. The results show that our model achieves good results using the IMEOCAP datasets.

Frequent merging, overtaking, and lane-changing behaviors at freeway on-ramp areas usually cause traffic bottlenecks with low efficiency and significant safety concerns. Fortunately, the development of connected and autonomous vehicles (CAVs) presents a promising technology for improving merging efficiency and safety. Current research on CAV merging strategies mainly focuses on single-lane scenarios and relies heavily on oversimplified simulation tests. To this end, this study proposes a merging strategy framework for the merging CAV in complex multi-lane mixed traffic scenarios, taking into account the potential interference caused by other vehicles’ lane-changing behaviors. First, a merging gap selection method assists the merging CAV in choosing a more suitable merging gap. Then, a lateral speed control strategy provides the CAV with highly efficient lateral guidance by giving optimal lateral speed control parameters. Furthermore, a pre-merging safety preparation method controls the CAV to adjust its speed longitudinally to avoid potential conflicts. Finally, the merging execution part is proposed to guarantee the CAV an effective, safe, and comfortable merging experience. The proposed model is tested in merging scenarios extracted from the Delft freeway trajectory dataset. Results indicate that the proposed merging strategy can significantly improve the merging efficiency by 45%, while offering a safe and comfortable merging trajectory for CAVs in multi-lane mixed traffic scenarios.

In the service industry, some operations require preparatory measures and cannot deliver immediate and reliable services. Identifying potential clients in densely populated environments presents a significant challenge for improving service efficiency. Current methodologies used to predict pedestrian trajectories demonstrate suboptimal performance in service-oriented contexts. In response to these challenges, this paper introduces “POI-GAN”, a novel approach tailored to forecasting pedestrian trajectories in service-centric settings. POI-GAN devises an interest point model that conceptualizes service locations within the scene as distinct obstacles. These service locations are subsequently characterized employing a social force model. Additionally, the framework introduces a field of view angle model which filters interactions between dynamic and static objects in the scene to establish plausibility. Subsequently, a generative model is used to produce projected pedestrian trajectories for future time frames. Empirical evaluations highlight the effectiveness of POI-GAN in improving trajectory prediction, particularly in scenarios with multiple interest points in the scene. Notably, POI-GAN exhibits superior performance when measured against analogous methods, as evidenced by the improved Average Displacement Error (ADE) and Final Displacement Error (FDE) metrics. This innovative approach empowers service providers with the capacity to effectively discern potential customers within the scene, ultimately elevating the quality of service delivery.

Vehicle object detection using UAV images is a crucial undertaking in urban traffic management and the advancement of autonomous driving technologies. Conventional networks fail to achieve accuary in detecting vehicle objects from a drone’s perspective due to the significant variations in the size of the target items, unequal distribution of their positions in the image, and image degradation induced by the drone’s movement.In order to surmount this challenge, this research suggests an enhanced TOOD object detection model named Drone-TOOD. The first proposal is to create a lightweight network skeleton called CSPRegNet by merging CSPblock with Regblock. Secondly, we incorporate Regblock into CSPPAFPN to enhance CSPRegPAFPN and incorporate EVCblock at the upsampling location of deep features to capture corner area details and minimize the degradation of feature information. In addition, a efficient task decomposition attention module is also proposed to enhance the interaction ability of positioning and classification tasks. This task decomposition module can highlight the characteristics of a specific task while retaining the characteristics of another task, thereby improving detection capabilities. Experiments conducted on the Drone Vision Challenge Benchmark (VisDrone) demonstrate that the enhanced model can obtain superior performance compared to TOOD. The average precision (mAP) achieved by our approach is 64%, surpassing TOOD by 7.9%. The frames per second (FPS) stayed constant at 27.2. Drone-TOOD demonstrates superior performance compared to other lightweight models on the VisDrone-2021 dataset. In order to demonstrate the robustness of our approach, we additionally performed ablation experiments and conducted tests on the UAV Detection and Tracking Dataset (UAVDT), resulting in an achieved mean average precision (mAP) of 64.6%. Furthermore, Drone-TOOD possesses a total of parameter approximately 21.7 M.

The employment of Massive MIMO technology in 5G communication results in a large number of RF channels in the base station. Therefore, the base station will suffer from the huge burden of equipping discrete bandpass filters under the traditional filtering strategy. An alternative filtering strategy called spatial filtering characterizes direct covering frequency selective surface (FSS) on multi-channel antenna array of base station and is expected to greatly relax the need for discrete bandpass filters. An aperture-coupled-cross-resonator (ACCR) FSS based spatial filtering scheme is investigated in this study. A low-profile spatial filtering patch array (LPSFPA) with thickness of $0.07\lambda _{0}$ and operating around 3.6GHz is constructed by placing an asymmetrical ACCR FSS tightly upon a patch antenna array with a half-period lateral shift. A good out-of-band suppression at both band-edges and enhanced in-band impedance bandwidth are achieved when the main beam is scanned with an elevation angle range of 0° to 30° and arbitrary azimuth angle. To verify the design, a finite LPSFPA sample whose main beam can be directed to $\theta _{\mathrm {b}}= 0^{\mathrm {\circ }}$ , $15^{\mathrm {\circ }}$ and $30^{\mathrm {\circ }}$ in E plane by different phasing and power-dividing networks is fabricated and tested. The measured results show that the fabricated LPSFPA sample is superior to the ordinary radiation patch array without the ACCR FSS in terms of both the impedance bandwidth and out-of-band suppression for various main beam directions. For the normal-beam case, the measured −10dB impedance bandwidth is from 3.6-3.8 GHz; the measured out-of-band suppression level of peak realized gain is more than 19 dB in the lower band-edge from 3.2-3.5 GHz and 17 dB in the upper band-edge from 3.9-4.2 GHz. The proposed spatial filtering scheme has no negative influence on the radiation performance of the radiation patch array.

Blood samples are easily damaged in traditional bloodstain detection and identification. In complex scenes with interfering objects, bloodstain identification may be inaccurate, with low detection rates and false-positive results. In order to meet these challenges, we propose a bloodstain detection and identification method based on hyperspectral imaging and mixed convolutional neural networks, which enables fast and efficient non-destructive identification of bloodstains. In this study, we apply visible/near-infrared reflectance hyperspectral imaging in the 380–1000 nm spectral region to analyze the shape, structure, and biochemical characteristics of bloodstains. Hyperspectral images of bloodstains on different substrates and six bloodstain analogs are experimentally obtained. The acquired spectral pixels are pre-processed by Principal Component Analysis (PCA). For bloodstains and different bloodstain analogs, regions of interest are selected from each substance to obtain pixels, which are further used in convolutional neural network (CNN) modeling. After the mixed CNN modeling is completed, pixels are selected from the hyperspectral images as a test set for bloodstains and bloodstain analogs. Finally, the bloodstain recognition ability of the mixed 2D-3D CNN model is evaluated by analyzing the kappa coefficient and classification accuracy. The experimental results show that the accuracy of the constructed CNN bloodstain identification model reaches 95.4%. Compared with other methods, the bloodstain identification method proposed in this study has higher efficiency and accuracy in complex scenes. The results of this study will provide a reference for the future development of the bloodstain online detection system.

Mobile edge computing (MEC) is considered a novel paradigm for computation-intensive and delay-sensitive tasks in fifth generation (5G) networks and beyond. However, its uncertainty, referred to as dynamic and randomness, from the mobile device, wireless channel, and edge network sides, results in high-dimensional, nonconvex, nonlinear, and NP-hard optimization problems. Thanks to the evolved reinforcement learning (RL), upon iteratively interacting with the dynamic and random environment, its trained agent can intelligently obtain the optimal policy in MEC. Furthermore, its evolved versions, such as deep reinforcement learning (DRL), can achieve higher convergence speed efficiency and learning accuracy based on the parametric approximation for the large-scale state-action space. This paper provides a comprehensive research review on RL-enabled MEC and offers insight for development in this area. More importantly, associated with free mobility, dynamic channels, and distributed services, the MEC challenges that can be solved by different kinds of RL algorithms are identified, followed by how they can be solved by RL solutions in diverse mobile applications. Finally, the open challenges are discussed to provide helpful guidance for future research in RL training and learning MEC.

In this paper, we apply a genetic algorithm (GA) for feature selection, wrapper approach, on wavelet scattering (WS) second-order coefficients to reduce the large frequency dimension (>500). The evaluation demonstrates the capability of GA to reduce the dimension space by approximately 30% while ensuring a minimum performance drop. Furthermore, the reduced WS directly impacts the training time of the convolutional neural network, by reducing the computational time by 20% to 32%. The paper extends its scopes to explore GA for feature selection on multiple timescales of WS: 46ms, 92ms, 185ms, and 371ms. Incorporating multiple timescales has improved classification performance (by ~ around 2.5%) as an acoustic representation usually contains information at different time scales. However, it can increase computational cost due to the larger frequency dimension of 1851. With the application of GA for feature selection, the frequency dimension is reduced by 50%, saving around 40% computational time, thus increasing the classification performance by 3% compared to a vanilla WS. Lastly, the entire implementations are evaluated using the Detection and Classification of Acoustic Scenes and Events (DCASE) 2020 dataset, and the proposed multiple timescales model achieves 73.32% of classification accuracy.

The key towards a low complexity model for convolution neural network is in controlling the number of parameters of the network and ensuring that the input representation is not extremely large. Hence, to tackle low complexity for acoustic scene classification (ASC), this paper proposed an enhanced wavelet scattering representation with a combination of mobile network modules and shuffling modules. While wavelet scattering comprises wavelet transform with multiple wavelet scales, the averaging operation to make the wavelet scattering invariant to translation limit the maximum timescale. Hence, wavelet scattering is affected by Heisenberg’s Uncertainty Principle. However, creating an input representation with multiple timescales does not meet the brief of low complexity modelling. Hence, we proposed a simple mixing of the first and second order with different timescales. The result is an input representation with nearly the same dimension as the usual wavelet scattering but with enhanced multiscale. To further leverage the ‘interleaved’ wavelet scattering, this paper presents sub-spectral shuffling inspired by shuffling modules that use stochasticity to improve the model’s generalization. Unlike channel shuffle that shuffles channel-wise and spatial shuffle that shuffles pixel-wise, sub-spectral shuffle aims at shuffling the feature maps frequency-wise with the concept of binning. Each bin is shuffled, so the high-frequency spectrum is shuffled to low-frequency spectrum position. As such, the model learns the general acoustic profile of a scene rather than memorizing what is happening at the low-frequency or high-frequency spectrum is erratic for ASC. In addition, this paper also studied temporal shuffling, which shuffles the feature maps temporal-wise, and evaluated sub-spectral shuffling, temporal shuffling, and channel shuffling individually. Our results demonstrated the superiority of sub-spectral shuffling and the modularity of shuffling modules. We then evaluate various combinations of the three shuffling modules on three acoustic scene classification datasets. Our best model combines the three shuffling modules and achieves 70.6% classification accuracy on DCASE 2021 Task 1a dataset, 82.15% on ESC-50 dataset, 81% on Urbansound8K, with ~65K parameters and a size of 126.6KB. In addition, the inclusion of shuffling modules has increased the model performance. Sub-spectral shuffling is especially useful in improving logloss, a metric used to determine the confidence level of the model.

Train network control management system (TCMS) is an important part of the High-speed rail train. Because of the TCMS's complex and redundant structure, long-term operation environment, etc., breakdowns inevitably in the long-time running. Based on the historical fault data of the TCMS accumulated during their online service, the working principles, failure modes, and effects analysis of TCMS are researched and the dynamic fault tree (DFT) model of TCMS failure is built. Then, the dynamic fault tree model is transformed into the Bayesian network (BN) model, which can model the reliability of such types of systems. Finally, combining DFT with BN is used for fault probability estimation and reliability assessment. The results present that increasing the reliability of key modules for the TCMS would be of great help to High-speed rail train engineers in the fault diagnosis field.

To obtain independent navigation results for autonomous underwater vehicles (AUVs) and construct high-resolution consistent seabed maps, a particle filter-based bathymetric simultaneous localization and mapping (BSLAM) method with the mean trajectory map representation is proposed. To reduce the computational consumption, particles only keep the current estimated position of the vehicle, while all historical states of the vehicle are stored in the mean trajectory map. Using this set-up, only the weights of the particles which closed to the mean trajectory map are calculated with newly collected bathymetric data. A hierarchical clustering procedure is also discussed to identify invalid loop closures. The performance of the proposed method is validated using both the simulated data and the field data collected from sea trails. The results demonstrate that the proposed method is 50% more accurate and 50% faster than a state-of-the-art particle filter-based BSLAM method, and it has similar accuracy but 30% faster compared with a graph-based BSLAM method.

A graph-based particle filter bathymetric simultaneous localization and mapping (BSLAM) method is proposed to solve the oscillation problem of the trajectories estimated by particles when using a low precise vehicle motion model and obtain accurate navigation results for autonomous underwater vehicles (AUVs). A graph-based trajectory update method is proposed to update the trajectories stored in particles before particle weighting to weaken the influence of the low precise odometer model on the particle trajectories. A particle weighting method based on submap matching is proposed to improve the robustness of the particle filter. Besides, a graph-based map generation method is proposed to solve the map selection problem of the particle filtering theory. The performance of the proposed method is demonstrated using a simulated dataset and a field dataset collected from a sea trial. The results show that the proposed method is more accurate and effective compared with a state-of-art particle filter BSLAM method.

The lack of strongly labeled data can limit the potential of a Sound Event Detection (SED) system trained using deep learning approaches. To address this issue, this paper proposes a novel method to approximate strong labels for the weakly labeled data using Nonnegative Matrix Factorization (NMF) in a supervised manner. Using a combinative transfer learning and semi-supervised learning framework, two different Convolutional Neural Networks (CNN) are trained using synthetic data, approximated strongly labeled data, and unlabeled data where one model will produce the audio tags. In contrast, the other will produce the frame-level prediction. The proposed methodology is then evaluated on three different subsets of the Detection and Classification of Acoustic Scenes and Events (DCASE) 2020 dataset: validation dataset, challenge evaluation dataset, and public YouTube evaluation dataset. Based on the results, our proposed methodology outperforms the baseline system by a minimum of 7% across these three different data subsets. In addition, our proposed method also outperforms the top 3 submissions from the DCASE 2019 challenge task 4 on the validation and public YouTube evaluation datasets. Our system performance is also competitive against the top submission in DCASE 2020 challenge task 4 on the challenge evaluation data. A post-challenge analysis was also performed using the validation dataset, which revealed the causes of the performance difference between our system and the top submission of the DCASE 2020 challenge task 4. The leading causes that we observed are 1) detection threshold tuning method and 2) augmentation techniques used. We observed that our system could perform better than the first place submission by 1.5% by changing our detection threshold tuning method. In addition, the post-challenge analysis also revealed that our system is more robust than the top submission in DCASE 2020 challenge task 4 on long-duration audio clips, where we outperformed them by 37%.

The main merit of grid-connected adjustable-speed pumped storage unit (ASPSU) is the capability to control the pumping power flexibly, which contributes to active power balance in power system operation. Physics-based dynamic modeling of ASPSU in pumping mode under power regulation is the foundation of the quantitative analysis on its transient performance, which has been studied inadequately. This study aims to derive a dynamic model of ASPSU in pumping mode which embodies reversible pump-turbine characteristics, and to conduct eigen analysis on ASPSU in pumping mode which facilitates a better understanding of its intrinsic dynamics. Firstly, the dynamic model of reversible pump-turbine is revealed to depict its regulation properties that the pumping power is adjusted by rotational speed rather than guide vane opening. Then, the state-space representation of ASPSU in pumping mode is established to be utilized with small disturbances (including indicial and ramp tests) and validated by comparing simulated results with actual records of a commissioned ASPSU in Japan. It is also effective for compensating the wind power fluctuations based on a timescale of seconds in simulation. Finally, the eigenvalues, damping ratios, and participation factors of the state variables are investigated based on the small-signal-stability model of ASPSU in pumping mode under power regulation. The results of eigen analysis indicate that the small-signal behavior of the system is characterized by electromechanical mode and electromagnetic mode, and the latter is dominant. Moreover, the damping ratio of electromagnetic mode which is sensitive to system stability needs to be increased.

Recently, significant progress has been made in the field of face detection. However, despite the demand for its high accuracy and recall rate, the efficiency of face detection algorithm is another key factor in evaluating its performance, which puts forward serious challenges to current models. To boost up the efficiency of face recognition, in this paper, we propose a lightweight rapid framework, called LRNet, which has fewer convolutional layers and a higher efficiency. In particular, our framework is consist of two major modules. One is the Feature Map Fast Shrink Module (FMFSM), which leads to a fast reduction in the size of feature maps and time consumed for detection by using Two Information Flow Block (TIFB). Another module, namely Variable Scale Face Detection Module (VSFDM), is consist of the Retinal Receptive Field Block (RRFB) and designed to prevent a single or composite feature map from undertaking too much tasks. In addition, we propose a new anchor strategy that considers not only the density of anchors with different scales but also the position and central symmetry of the features. Our proposed LRNet achieves high accuracy and efficiency on the challenging FDDB dataset for face detection. When the number of false positives is 2000, its True Positive Rate (TPR) under discrete and continuous scores can achieve 0.951 and 0.725 respectively. When running on GTX 1080Ti, given images with a resolution of $1024\times 1024$ , the average time consumed for detection is merely 8.88 ms.

Interference exploitation has been considered an effective method to improve secrecy in underlaid device-to-device (D2D) communication networks. This paper studied interference exploitation in unauthorized underlaid D2D communication networks, where a cellular user (CU) transmits uplink information to a base station (BS) and a pair of suspicious D2D users (DUs) transmit the suspicious information at the same time. We used the beamforming of the full-duplex (FD) CU to simultaneously achieve the legitimate uplink transmission of the CU and wireless surveillance on the suspicious DUs. Considering the different priority levels of the tasks, we divided the real-time quality of service (QoS) requirements of the CU into three cases and formulated the problem in each case. We derived beamforming expressions for each problem in the perfect self-interference cancellation (PSIC) and then proposed targeted algorithms in the imperfect self-interference cancellation (ISIC). Finally, simulation results are provided to validate the effectiveness and the advantages of our proposed schemes in situations with different QoS requirements.

In the field of correlation filter object tracking, the traditional template-update method easily causes template drift, so it performs poorly in complex scenes. To enhance the robustness of the template, a novel incremental multi-template update strategy is proposed in this paper. We find that reliability varies among all historical filters and that highly reliable filters are key to achieving accurate tracking. The incremental multi-template update strategy combines the local maximum-reliability filter template with the historical filter template incrementally, which is obviously different from the traditional update method. We apply this strategy to two trackers with superior performance. The experimental results of three test benchmarks, including the VOT2016, OTB100 and UAV123 datasets, show that the performance of our trackers is superior to that of the state-of-the-art trackers.

At present, there are many challenges in the field of pedestrian attribute recognition, such as small targets of some attributes, imbalanced samples, and low recognition accuracy of complex samples. In view of the above problems, we improved the model in two perspectives: 1) We proposed Feature Pyramid Attention Model (FPAM). In order to solve the problem that attributes are distributed in different locations in the pedestrian image, FPAM adopted the attention mechanism on the basis of ResNet-50, by which the model’s attention could be focused on key areas of the image. As for the difficulty in small targets attributes, we adopted feature pyramid integration strategy; 2) We proposed Multi Label Focal Loss (MLFL). Referring to Binary Cross Entropy Loss Function (CE) and Weight Binary Cross Entropy Loss Function (WCE), we added the weight parameters of samples which are difficult to classify to improve the recognition accuracy, and the rate of convergence was increased. Results show that our proposed method achieves 84.83% mA, 79.37% Accuracy, 87.47% Precision, 86.09% Recall, and 86.77% F1 on PETA dataset.

The early fault of a cable is usually a incipient fault, which is difficult to find and can easily develop into an insulation breakdown fault. Thus, it is of great significance to quickly and accurately find an incipient cable fault. This paper presents an fault location method for cable insulation based on the electric quantities of both ends. Based on the equivalent circuit model of three-phase single-core cable distribution parameters, the calculation method of the voltage along the cable is derived. The equivalent model of cable insulation in resistive and capacitive parallel is proposed and applied to the early fault modeling of cable. On the basis of the calculation formula of the voltage along the cable and the electric quantity at both ends of the cable, the analytical expression of the cable fault location is obtained by solving the fault equation directly. Extensive simulations show that the algorithm, this method can locate the incipient fault accurately, and is not affected by the fault point, fault resistance and fault inception angle.

Person re-identification (re-ID) presents various applications in surveillance system, but most existing models are proposed under supervised framework. These methods require large amounts of annotated pedestrian data, which limits their scalability and flexibility in a new application scenario. Aiming to relax this limitation, this article exploits the attribute-invariant characteristics and domain correlations into cross-domain person re-ID, while most unsupervised methods only consider the identity features and ignore the different importance of each source image to the target domain. Specifically, this article proposes an Attribute Memory Transfer Network (AMTNet) with two major contributions of domain-balanced memory and attribute-invariant memory modules. The first domain balanced-memory integrates a domain correlation learning method to evaluate the importance of each source image to the target domain, which is involved into the transfer learning; The second attribute-invariant memory can transfer the source attribute knowledge into the target domain with preserving the identity information to conduct the re-ID process. Extensive evaluated experiments elaborate the superiority of AMTNet on two large datasets of Market-1501 and DukeMTMC-reID, compared with hand-crafted and deep learning feature-based methods.

This paper focuses on the dynamic modeling, controller design, and simulation verification of a new cable-driven underwater vehicle model system, which is proposed for future application on a tension leg platform. Compared with conventional underwater vehicles, the proposed cable-driven underwater vehicle model system has a unique structure and subjects to not only unknown underwater disturbances but also time-varying nonlinear cable tractions. To achieve accurate and robust trajectory tracking control, a universal high-order nonlinear dynamic model is established with multisource uncertainties, and an integrated estimation-based adaptive backstepping terminal sliding mode controller is designed. The proposed controller utilizes the integrated estimation method to handle the multisource unknown uncertainties, where recurrent radial basis function neural network and disturbance observer are employed for the estimations of uncertainties, and adaptive robust methods are utilized for compensations of estimation errors. The backstepping terminal sliding mode control method is utilized to improve the tracking performance and converging rate. In addition, the issue of “explosion of complexity” occurred in a traditional backstepping design is tackled by an adaptive control method, and the input saturation problem is solved by anti-windup compensator. Based on the Lyapunov analysis, all closed-loop signals are proved to be uniformly ultimately bounded. The numerical simulations show that the developed controller is not only robust against environmental disturbances and adaptive to unknown time-varying uncertainties but also able to steer the trajectory tracking errors along the prescribed transient, thus leading to effective cable-driven underwater vehicle model system control.

Image classification has attracted more and more attention. During the past decades, image classification has shown growing interest in representation-based classification methods, such as sparse representation-based classification and collaborative representation-based classification. However, the available representation-based methods still suffer from some problems. Especially, most methods only consider the shared representation of a test image. In this paper, we propose an elastic-net regularized regression algorithm (ENRR) for image classification. Specifically, our proposed method combines shared sparse representation with class specific collaborative representation when representing the test sample. Moreover, we extend the proposed ENRR to arbitrary kernel space to achieve better classification performance due to specificities and complexities of original images. The extensive experiments on face recognition datasets, handwritten recognition datasets, and remote sensing image datasets clearly demonstrate that the proposed ENRR outperforms several conventional methods in classification accuracy.

This paper presents an adaptive null-space-based behavioral (NSB) method to deal with the problems of saturation planning and lack of adaptability when the traditional NSB method is applied to the formation control of multiple unmanned surface vehicles (MUSVs). First, the NSB method is analyzed, and the matrix theory is introduced to propose a behavior priority theory determination method based on a vector graph. Second, consider the maneuverability of the unmanned surface vehicle (USV), variable coefficients with physical significances are introduced to redefine the behavioral motion model, making the speed limit solved in each working condition within the maneuvering range of the USV and effectively improving the formation ability of MUSVs. Third, a logical priority collision avoidance strategy between the MUSVs is proposed, aiming at the problem that when the USVs judge each other as obstacles, both of them adopt the obstacle avoidance behavior resulting in two vehicles' courses deviating from the direction of the target point. Finally, a simulation platform for the formation control of MUSVs was established by taking the Dolphin-I prototype USV as the experimental object, and the feasibility of the proposed method was verified by a simulation test.

We present a saliency-guided algorithm for detecting the locations of repetitive structures on building facades. First, the global and local saliencies of each point are determined by measuring the global rarity and the local distinctness. The saliency map is utilized to adaptively extract the salient points. Second, the salient points are vertically sliced. A curve can be derived by counting the total number of points in each slice. Then, the curve is converted into a square wave to locate the vertical splitting position. Next, each segment is horizontally sliced, similar to the vertical splitting. The salient points are partitioned into repetitive candidates after the vertical and horizontal splitting. Finally, the repetitive candidates are refined according to the similarity of the neighborhood and the regularity of the arrangement. The experimental results demonstrate that our method can quickly and effectively extract repetitions from facade point clouds.

The main task of future networks is to build, as much as possible, intelligent networking architectures for intellectualization, activation, and customization. Software-defined networking (SDN) technology breaks the tight coupling between the control plane and the data plane in the traditional network architecture, making the controllability, security, and economy of network resources into a reality. As one of the important actualization methods of artificial intelligence (AI), machine learning (ML), combined with SDN architecture will have great potential in areas, such as network resource management, route planning, traffic scheduling, fault diagnosis, and network security. This paper presents the network applications combined with SDN concepts based on ML from two perspectives, namely the perspective of ML algorithms and SDN network applications. From the perspective of ML algorithms, this paper focuses on the applications of classical ML algorithms in SDN-based networks, after a characteristic analysis of algorithms. From the other perspective, after classifying the existing network applications based on the SDN architecture, the related ML solutions are introduced. Finally, the future development of the ML algorithms and SDN concepts is discussed and analyzed. This paper occupies the intersection of the AI, big data, computer networking, and other disciplines; the AI itself is a new and complex interdisciplinary field, which causes the researchers in this field to often have different professional backgrounds and, sometimes, divergent research purposes. This paper is necessary and helpful for researchers from different fields to accurately master the key issues.

Based on model-free adaptive control (MFAC) theory, this paper presents a variable output constraint MFAC (VOC-MFAC) algorithm to enhance the robustness of an unmanned surface vehicle's (USV's) heading subsystem. The contributions of this paper are as follows. First, a controller output constraint function is proposed to solve the system's control performance sensitivity to the redefined output gain when the redefined compact format model free adaptive control (RO-CFDL-MFAC) method is used to control an unmanned surface vehicle's heading. Second, the compact format dynamic linearization data models for a USV's angular velocity subsystem and heading subsystem are established, and the convergence of the closed-loop system under environmental disturbances is proven through rigorous theoretical analysis. Finally, the control algorithm proposed in this paper is simulated and tested in the field using the “Dolphin IB” unmanned surface vehicle platform developed by our research group, and the effectiveness of the VOC-MFAC algorithm is verified by the experimental results.

This paper presents a novel swarm control framework for path following of multiple underactuated unmanned marine vehicles (UMVs) with uncertain dynamics and unmeasured velocities. Main contributions are as follows: (1) unlike previous master-slave formation control, a swarm system function with distributed and self-organized capability is designed; (2) a center-of-swarm (COS) guidance scheme without vehicle number constraints is proposed for swarm path following, where an improved artificial potential field (APF) using ring-shaped repulsion is further employed for collision avoidance and obstacle avoidance; (3) a nonlinear velocity observer is incorporated into the proposed swarm control framework to estimate the unmeasured velocities, thereby contributing to robust controllers based on fuzzy sliding mode against uncertain dynamics and time-varying disturbances. Simulations are carried out to illustrate the universal applicability and effectiveness of the proposed swarm control framework.

Recently, broadband maritime communication has attracted much attention due to the rapid development of blue economy. In addition to the conventional MF/HF/VHF bands, there has been increasing interests in the utilization of higher frequency bands to provide broadband data service to the sea area. To design efficient maritime communication systems, the first and a fundamental requirement is to develop a framework to understand the wireless channels. In an integrated air-ground-sea communications network, there are two major type of channels to be investigated, namely the air-to-sea channel (e.g., for communication links from the aircraft-based base stations or relays) and the near-sea-surface channel (for land-to-ship/ship-to-land or ship-to-ship communications). Due to the unique features of the maritime propagation environment such as sparse scattering, sea wave movement, and the ducting effect over the sea surface, the modeling of these maritime channel links differs from conventional terrestrial wireless channels in many aspects and, consequently, will result in significant impact on the transceiver design. In this survey, we highlight the most notable differences from the modeling perspective as well as the channel characteristics for the air-to-sea and near-sea-surface channel links, with more focus on the latter. After a thorough review of existing modeling approaches and measurement campaigns, we conclude that the sparse and the location-dependent properties constitute the most important and distinctive characteristics of the maritime wireless channels. As such, we further remark on the challenges and research topics for future development of maritime communications.

Identification of different risk factors and early prediction of mortality for patients with heart failure are crucial for guiding clinical decision-making in Intensive care unit cohorts. In this paper, we developed a comprehensive risk model for predicting heart failure mortality with a high level of accuracy using an improved random survival forest (iRSF). Utilizing a novel split rule and stopping criterion, the proposed iRSF was able to identify more accurate predictors to separate survivors and nonsurvivors and thus improve discrimination ability. Based on the public MIMIC II clinical database with 8059 patients, 32 risk factors, including demographics, clinical, laboratory information, and medications, were analyzed and used to develop the risk model for patients with heart failure. Compared with previous studies, more critical laboratory predictors were identified that could reveal difficult-to-manage comorbidities, including aspartate aminotransferase, alanine aminotransferase, total bilirubin, serum creatine, blood urea nitrogen, and their inherent effects on events; these were determined to be critical indicators for predicting heart failure mortality with the proposed iRSF. The experimental results showed that the developed risk model was superior to those used in previous studies and the conventional random survival forest-based model with an out-of-bag C-statistic value of 0.821. Therefore, the developed iRSF-based risk model could serve as a valuable tool for clinicians in heart failure mortality prediction.

This paper focuses on the modeling, diving controller design, and experiment of a special cable-driven underwater parallel platform with eight-cable coupling drive structure. Kinematic and dynamic models of the platform are established utilizing a simplified approach, and the hydraulic driven control model is derived based on joint-space method. To improve the diving control precision and system robustness despite the complex dynamic behaviors and manifold unknown disturbances, a three-layer adaptive diving control strategy is proposed. Among the three control layers, layer 1 is responsible for dive planning and online monitoring, layer 2 places emphasis on synchronous control by employing an improved relative coupling strategy, layer 3 utilizes an adaptive radial basis function neural network-based backstepping sliding mode control algorithm (ARBFNN-BSMC) to achieve high precision speed control of the single driving branch. Hardware-in-the-loop simulations and experimental results illustrate that the proposed three-layer adaptive diving control strategy can asymptotically drive the cable-driven underwater parallel platform onto a predefined diving trajectory with favorable precision, robustness, and stability.

In this paper, a distributed adaptive critic design-based asymptotical cooperative tracking control scheme is established for a class of unknown high-order multi-agent systems with disturbances on undirected graphs. The cooperative tracking problem of high-order multi-agent systems is first transformed to the stabilizing problem of filtered tracking errors dynamic systems, which only depend on the information from neighboring agents. Then, a novel distributed adaptive critic design-based controller is developed for the newly obtained filtered tracking errors dynamic systems. The special local filtered tracking errors and local critic signals are defined for facilitating the implementation of distributed adaptive critic design. Novel distributed weights updating laws for critic neural networks, actor neural networks, and adaptive parameters are proposed. Furthermore, the RISE technique is introduced to suppress the external disturbances and neural network approximation errors. The asymptotical stability of a closed-loop system is demonstrated by using Lyapunov stability theory. Finally, a simulation example justifying the effectiveness of the proposed control scheme is given.

With increasing security and privacy requirements, electrocardiogram (ECG)-based biometric human identification and authentication is gaining extensive attention. This paper aims to solve three major problems: stable identity feature is hard extracted from the inferior quality ECG, the performance of authentication system falls down when the size of registered sample set increases, and the authentication system needs to retrain when a new registered identity is added. To improve the robustness of identity feature, this paper proposed a multiscale feature extraction method using a multiscale autoregressive model (MSARM). First, the performance of multiscale feature was tested by simple matching method based on Chi-square distance in identification system. The test was performed on self-built SIAT-ECG and public PTB databases, which contain 146 and 100 (50 healthy volunteers and 50 patients with myocardial infarction) individuals, respectively. The recognition rate exceeded 93.15% for both databases in identification scenario. The results revealed that the MSARM has more excellent performance than other feature extraction methods. Then, this paper proposed a combination classifier method with one-to-one structure in authentication mode. It yielded a true rejection rate (TRR) of 98.99% and true acceptance rate (TAR) of 95.04% when registered sample set contains 140 individuals from SIAT-ECG database. Therefore, the proposed MSARM and combination classifier not only significantly improve the accuracy but also enhance the practicability of ECG-based biometric systems.

While random linear network coding is known to improve network reliability and throughput, its high costs for delivering coding coefficients and decoding represent an obstacle where nodes have limited power to transmit and decode packets. In this paper, we propose sparse network codes for scenarios where low coding vector weights and low decoding cost are crucial. We consider generation-based network codes where source packets are grouped into overlapping subsets called generations, and coding is performed only on packets within the same generation in order to achieve sparseness and low complexity. A sparse code is proposed that is comprised of a precode and random overlapping generations. The code is shown to be much sparser than existing codes that enjoy similar code overhead. To efficiently decode the proposed code, a novel low-complexity overhead-optimized decoder is proposed where code sparsity is exploited through local processing and multiple rounds of pivoting. Through extensive simulation comparison with existing schemes, we show that short transmissions of the order of 102 -103 source packets, a denomination convenient for many applications of interest, can be efficiently decoded by the proposed decoder.

With the rapid development of mobile health technologies and applications in recent years, large amounts of electrocardiogram (ECG) signals that need to be processed timely have been produced. Although the CPU-based sequential automated ECG analysis algorithm (CPU-AECG) designed for identifying seven types of heartbeats has been in use for years, it is single-threaded and handling lots of concurrent ECG signals still poses a severe challenge. In this paper, we propose a novel GPU-based automated ECG analysis algorithm (GPU-AECG) to effectively shorten the program executing time. A new concurrencybased GPU-AECG, named cGPU-AECG, is also developed to handle multiple concurrent signals. Compared with the CPU-AECG, our cGPU-AECG achieves a 35 times speedup when handling 24-h-long ECG data, without reducing the classification accuracy. With cGPU-AECG, we can handle 24-h-ECG signals from thousands of users in a few seconds and provide prompt feedback, which not only greatly improves the user experience of mobile health services, but also reduces the economic cost of building healthcare platforms.