Search

This paper introduces a novel method for accurately estimating the 2D coordinates of hand keypoints from single static images, utilizing a sequential convolutional neural network optimized with Maximum Likelihood Estimation Loss. Unlike traditional heatmap-based techniques, our approach eliminates the need to generate label heatmaps and sidesteps the direct optimization of model parameters based on noisy labels. Instead, it concentrates on modeling the distribution of the discrepancies between predicted results and ground truth, rather than the potential presence of noisy labels, thus enabling the direct prediction of hand keypoint coordinates. Furthermore, we propose a sequential training and inference framework that consists of a deep convolutional backbone network and a multi-stage sequential network. Each stage of this network features similar structures, facilitating the progressive and precise prediction of hand keypoint coordinates. Our extensive experimental results demonstrate that our approach is both highly accurate and robust, outperforming mainstream methods under the experimental conditions detailed in this paper.

The terahertz range, as a type of electromagnetic wave with wavelengths between microwaves and the infrared band, has the characteristics of penetration, low energy and a stable absorption spectrum of specific substances, and is widely used in non-destructive testing, human security inspections, biological tissue diagnoses and military detection. In particular, terahertz wave 3D imaging technology can detect the internal information of the target of detection, and it has become the focus of current research. This study carried out research on 3D reconstruction and object detection algorithms based on terahertz images. In view of the problem that the MVS (Multi-ViewStereo) series of 3D reconstruction algorithms ignore the context information between the cost layers and have unsatisfactory reconstruction effects when used on complex regions, an improved MVSNet 3D reconstruction algorithm FCTMVSNet(Feature and Cost Transformer Depth Inference for Unstructured Multi-view Stereo) based on Transformer is proposed here. A structured object recognition algorithm was designed to provide theoretical support for subsequent terahertz image-based object detection algorithms.

Under the condition of a small time scale (e.g. second), distributed photovoltaic (PV) power generation output has the problems of strongly fluctuating and difficult to accurately simulate. It affects the control strategy and operation mode of hybrid energy systems. To address this problem, a data-driven small-scale distributed PV plant power output model on a 1-second time scale is proposed for the generation of second-by-second PV power output scenarios in hybrid energy systems. Firstly, this work analyzes the characteristics of PV power output at the 1-second time scale based on the probability distribution of power output fluctuations. Secondly, an index system that characterizes the PV power output fluctuation characteristics at the 1-second time scale is constructed. Then, using the data-driven method, a BP neural network model is constructed to simulate the PV power output at the 1-second time scale. Finally, a simulation is performed using the measured data from the PV plant. The findings demonstrate that compared to PV power output models in seconds based on Pearson systematic random numbers: (1) The correlation coefficient (r) of the proposed model is more than 0.8, in a higher degree of fit; (2) The root mean square error (RMSE) of the proposed model achieves 0.005, generally representing a 37.12% reduction. Overall, both the time scale and model accuracy of this model have deep potential value in PV power output modeling and system regulation.

Online education has become a popular teaching method beyond classroom education. However, online teaching resources bring the software and hardware resources cost and security burden of cloud computing, which limits the willingness to share online resources. How to effectively motivate different schools to share teaching resources is a challenging issue. This paper proposes an excitation mechanism to promote high-quality online resource sharing. Firstly, under the resource constraints of the cloud platform, a utility oriented resource sharing model is established. Then, a game of choice between students and course resources is studied, which allows users to participate course strategy selection. Next, a particle swarm optimization algorithm was used to design an optimal cloud computing resource allocation scheme that maximizes revenue. It expects that high-quality course resources will be allocate more cloud computing resources. Finally, experiment simulation shows that appropriately excitation mechanisms can effectively promote the sharing of teaching resources, furthermore, the proposed resource allocation scheme is significantly better than the average allocation scheme.

Blood pressure has a 24-hour repetitive and regular variation which shows circadian rhythm. Using the multivariate time series analysis method of vector autoregressive model, we could realize the simultaneous prediction for both systolic and diastolic blood pressures. We choose blood pressure from 6 AM to 10 AM in 3 weeks as an episode to construct a prediction model. Missing values were imputed by regression models. Subsequently, we defined segments as positive or negative segments according to blood pressure measurements. The predictions were accomplished by vector autoregressive model (VAR). Both positive and negative segments were randomly selected from each patient to summarize the effect of prediction models. In this study, the MAPE (Mean Absolute Percentage Error) of systolic blood pressure and diastolic blood pressure were both less than 10%, indicating that the VAR model was adaptable in predicting the blood pressure of hypertensive patients. Based on VAR, we could provide early warning to breakthrough of blood pressure thresholds. The sensitivity, specificity, and accuracy for patients in the training sets were 77.50%, 81.58 %, and 79.49% respectively, and the sensitivity, specificity, and accuracy for patients in the training sets were 76.92%, 80.00% and 78.43% respectively. This research took information of both systolic and diastolic blood pressures at the same time to establish the VAR models and enabled simultaneous prediction for systolic and diastolic blood pressure.

Chinese characters have been created into many font styles, such as official script, running script, and regular script. Other than that, some famous calligraphers, such as Ouyang Xun and Yan Zhenqing, have produced fonts with their style. Being able to detect and recognize these font styles quickly and accurately has essential applications in graphic design, page layout, handwriting identification, and other use cases. Distinguishing between font styles requires professional knowledge, which almost inevitably leads to errors for unprofessional people. Therefore, this paper presents a sword-like model based on a convolutional neural network with a sword structure to recognize font styles for Chinese characters. This model includes 15 convolutional layers. For each layer, we gradually increase the number of convolutional kernels to better extract the classification features of the input image. This paper uses four downsampling layers in the model. For each downsampling operation, the length and width of the image become half of their original values while the number of channels gradually increases, leading to a sword-like shape. As a result, we name our model as SwordNet. We also created a Chinese font dataset called the Nankai Chinese Font Style dataset and made it available on Github. Using the above dataset, we compared the accuracy of our model with six other state-of-the-art network models. The experiments showed that SwordNet could achieve an average recognition accuracy of 99.03% in multiple experiments, while the other six models can only achieve accuracy up to 94.91%. So we can conclude that SwordNet could perform better in font style recognition than other models.

Cloud computing services have to gain fast awareness by many organizations because of their cost-effectiveness, but they are faced with many security issues in protecting client business data in the current technological generation. Order-preserving encryption (OPE) is a very important technique for databases, and cloud storage encryption executes range queries efficiently, but regrettably, limited systems have stood to accomplish confirmable security. However, these schemes leak the distribution of repeated plaintext values. This paper presents a random perturbation distribution scheme (RPDS), which is secure and does not leak the distribution of repeated plaintext values. Our technology is based on Popa’s mutable order preserving encoding (mOPE) and storage-aware order-preserving encoding (stOPE). We supported RPDS with an insertion algorithm in which we added a random bit value to handle a repeated plaintext value that stops the server from recognizing repeated values, and we developed a proof to show the correctness of its syntax. This paper suggests that the arbitrary collision probability and proof show that collisions can only occur with a lower probability in the RPDS. This paper presents a proof of the safety of the RPDS and its applicability, usability, and functionalities. Finally, we compared the RPDS with existing OPE schemes and provided an experimental result for the practicality of the RPDS.

Aiming at the speed regulation problem of permanent magnet synchronous motor (PMSM) drives, a novel backstepping nonsingular terminal sliding mode control (BNTSMC) method with finite-time disturbance observer (FTDO) is proposed. In order to ensure excellent tracking and anti-disturbance performance, the backstepping nonsingular terminal sliding mode speed controller based on equivalent motor model and the recursive principle is designed for PMSM. The benefits of this approach are that the controller has asymptotic stability based on backstepping design and strong robustness for the matched disturbance based on sliding mode control. Then, to further improve the anti-disturbance performance, a finite-time disturbance observer is designed, and it is used to estimate the mismatched external disturbance in motor drive system. The stability of the system is also proved with Lyapunov criterion. Finally, the experimental results show that the proposed method has the advantages of fast convergence and strong robustness against parameter variations and external disturbance.

Detecting loop closures in dynamic environments is a severe challenge for the simultaneous localization and mapping (SLAM) system. Convolutional neural networks (ConvNet) could provide high-level and abstract representations extracted directly from images as image descriptors. Some novel ConvNet-based methods have been presented. In dynamic environments, they perform better than the state-of-the-art methods which use hand-crafted features. In this paper, (1) We proposed a flexible loop closure detection workflow based on the holistic representations; (2) In this workflow, a post-processing method is applied to the raw holistic ConvNet representations for redundant information compression; (3) In addition, a compression ratio is introduced in (2) to determine how much information will be retained depending on background change and moving objects. We evaluated our workflow in four open datasets. The experimental results demonstrate that the proposed workflow performs better than many state-of-the-art methods and ConvNet-based approaches.

SLAM in dynamic environments is still a severe challenge for most feature-based SLAM systems. Moving objects will lead to terrible errors in the calculation of frame tracking and local mapping. We propose a novel method for keypoints selection to lower the negative effect brought by moving objects during map construction. To address this challenge, we concentrate on the combination of coarse semantic information and a feature-based SLAM system. In this article, a modified CenterNet object detector is proposed as the moving object detection thread for providing coarse semantic information and 2D location. The modified CenterNet can provide a faster and more accurate prediction. For each frame in a sequence, objects in a scenario will be classified into two motion states, non-static and static, according to the category prediction from the moving object detection thread. Then important processing called semantic data association is presented for motion removal in frame tracking thread and local mapping thread. In this way, the keypoints on the non-static objects will not be chosen for calculation. Finally, on this basis, a modified real-time SLAM system is presented. Experimental results in TUM indoor dataset show that the proposed SLAM system outperforms the state-of-the-art ORB-SLAM2 in dynamic environments. Besides that, our method has a better speed-accuracy trade-off than other deep learning-based SLAM systems in dynamic environments.

The pipeline system on the outside of aero-engine needs to work in resonance environment in some cases, therefore, it is necessary to reduce the vibration amplitude of pipeline systems to improve the reliability in the dynamic design stage. In this paper, a single pipeline system with multi-hoop supports was taken as the object and a method based on genetic algorithm to realize the layout of hoops and effectively reduce the resonance amplitude of pipeline system was proposed. Considering that the system belongs to the statically indeterminate structure, a new semi-analytical model was developed, that is, the pipeline is modeled under free boundary conditions firstly, and then the hoop is introduced into the pipeline system in the form of spring-damping structure. Meanwhile, in the process of modeling, to improve the analysis accuracy of the model, the non-uniform distribution springs were used to simulate the support stiffness of hoops, and the uniformly distributed dampers were used to simulate the support damping of hoops. Taking the position of the hoop as the design variable, the optimization model of the hoop layout with minimizing the maximum vibration amplitude of pipeline systems as the optimization objective was established, furthermore the optimization solution process of the hoop layout based on genetic algorithm was given. Finally, a case study was carried out to verify the rationality of the proposed semi-analytical model and the optimal hoop support positions were obtained by the proposed optimization model and method.

Simultaneous Localization and Mapping (SLAM) plays an important role in the computer vision and robotic field. The traditional SLAM framework adopts a strong static world assumption for convenience of analysis. It is very essential to know how to deal with the dynamic environment in the entire industry with widespread attention. Faced with these challenges, researchers consider introducing semantic information to collaboratively solve dynamic objects in the scene. So, in this paper, we proposed a PSPNet-SLAM: Pyramid Scene Parsing Network SLAM, which integrated the Semantic thread of pyramid structure and geometric threads of reverse ant colony search strategy into ORB-SLAM2. In the proposed system, a pyramid-structured PSPNet was used for semantic thread to segment dynamic objects in combination with context information. In the geometric thread, we proposed a OCMulti-View Geometry thread. On the one hand, the optimal error compensation homography matrix was designed to improve the accuracy of dynamic point detection. On the other hand, we came up with a reverse ant colony collection strategy to enhance the real-time performance of the system and reduce its time consumption during the detection of dynamic objects. We have evaluated our SLAM in public data sheets and real-time world and compared it with ORB-SLAM2, DynaSLAM. Many improvements have been achieved in this system including location accuracy in high-dynamic scenarios, which also outperformed the other four state-of-the-art SLAM systems coping with the dynamic environments. The real-time performance has been delivered, compared with the geometric thread of the excellent DynaSALM system.

User interests modeling has been exploited as a critical component to improve the predictive performance of recommender systems. However, with the absence of explicit information to model user interests, most approaches to recommender systems exploit users activities (user generated contents or user ratings) to inference the interest of users. In reality, the relationship among users also serves as a rich source of information of shared interest. To this end, we propose a framework which avoids the sole dependence of user activities to infer user interests and allows the exploitation of the direct relationship between users to propagate user interests to improve system's performance. In this paper, we advocate a novel modeling framework. We construct a probabilistic user interests model and propose a user interests propagation algorithm (UIP), which applies a factor graph based approach to estimate the distribution of the interests of users. Moreover, we incorporate our UIP algorithm with conventional matrix factorization (MF) for recommender systems. Experimental results demonstrate that our proposed approach outperforms previous methods used for recommender systems.

In recent years, e-learning systems such as massive open online courses (MOOC) have been widely employed in academic institutions and have shown its power in enhancing the students' learning ability. Automatic detection and classification of knowledge points in lecture video would significantly enhance the performance of the online learning platform. Most of the previously presented approach for knowledge discovery focused on the text and audio documents, whereas the identification of knowledge points in videos still remains a challenge. To bridge this gap, we proposed a novel convolutional neural network which was designed for the characteristics of lecture video. It could both extract the temporal-spatial and semantic information from the multimedia record. To evaluate the performance of the proposed technique, we conducted comparison experiments between the state-of-the-art methods and ours. The experimental results demonstrated that the presented approach outperformed the state-of-the-art techniques and could be potentially invaluable for the accurate discovery of knowledge points within videos.

High frequency Eddy Current Testing (ECT) is one of the key non-destructive testing techniques for ensuring integrity of a structure of Carbon Fiber Reinforced Polymer (CFRP) material. An efficient numerical simulator is indispensable to enhance the performance of quantitative ECT for CFRP structures from both point of view of probe optimization and defect sizing. In this paper, a fast forward simulation scheme based on the A-Φ formulation and databases approach is proposed, implemented and experimentally validated for the rapid and high precision simulation of ECT signals due to defects in a CFRP plate by updating an FEM-BEM hybrid code for ECT problem. Comparison of numerical results of the present method with those of the conventional full FEM-BEM code and the experimental results for artificial cracks in CFRP laminate plates indicates that the proposed novel fast forward scheme can predict ECT signals over 300 times faster but without worsening numerical accuracy, which enables it to be applied to efficient reconstruction of cracks in CFRP plates and for probe optimization.

Cloud computing allows applications to share resources on physical machines in an independent manner so as to effectively increase the utilization of physical machines. To enable cloud service users to obtain high-quality services, it is significant for cloud service providers to reduce the time required to allocate and deploy virtual machines, as well as to provide time windows for the deployment of physical servers. To ensure demand is met, one effective implementation method is to predict the workload of virtual machines in the future, as this contributes to the efficient deployment of a data center's requests from virtual machines to dynamically changing physical servers. However, due to the complexity of virtual machine requests, virtual machine workload prediction remains a significant challenge at present; on the one hand, it is difficult for standard recurrent neural networks to capture long-term dependencies because of the disappearance of gradients, while on the other hand, the long short-term memory method cannot handle irregular intervals. Accordingly, a new virtual machine workload prediction method that is capable of managing irregular time intervals is proposed in this paper. The proposed method combines the amount and time intervals of historical virtual machine work requests in order to accurately predict the virtual machine's future workload using a fixed length of time as a unit. Experiments show that the proposed model can generate more accurate prediction results than the long short-term memory method.

Flywheel motor topology was analyzed, and a main performance optimization model and stator/rotor core loss optimization model were provided. Then, an adaptive kriging surrogate model based on the MaxPro design (AKMMP) was used, improving the optimization precision and efficiency. Based on these preliminary outcomes, a stepwise optimization design method of flywheel motor based on AKMMP was proposed. The brief design process includes five steps: 1) initial structure of the flywheel motor: outer rotor, 16 poles/18 slots, double layer concentrated winding, inclined shoulder flat bottom slot, water cooling, and F grade insulation; 2) first step optimization: performance optimization using AKMMP to obtain the performance Pareto solution set; 3) according to the performance of a flywheel motor, selection of the suitable case from the performance Pareto solution set; 4) second step optimization: further optimization of the stator/rotor core loss for the case obtained before using AKMMP, and; 5) creation of the final optimization design structure. Then, complete the flywheel motor design by the stepwise optimization design method. Finally, the prototype of the flywheel motor was produced, and performance testing was performed.

Joint fraud is one of the most common fraud types existing in medical fraud. However, joint fraud detection is a difficult problem because fraudsters take only a very small part of the population and fraudsters do everything to bypass fraud detection constraints. Most existing fraud detection studies focus on finding normal behavior patterns and treat those who possess behaviors that violate behavior patterns as fraudsters. However, these methods generally have high false positives because normal people may also sometime behave contrary to normal behavior patterns. To address this issue, we propose an abnormal group-based joint fraud detection method named abnormal group-based joint fraud detection method. This method can distinguish suspicious fraudsters from normal persons who have unusual behaviors by abnormal group mining in person similarity adjacency graph so that the occurrence of false positives caused by non-fraudulent abnormal behavior can be reduced. The extensive experiments using medical insurance data show that our approach has improved the precision of joint fraud detection by more than 10% compared with the conventional methods.

This paper is mainly focusing on the problem of speed tracking control for permanent magnet synchronous motor (PMSM) drive system subject to matched and mismatched disturbance. This is achieved by a novel Port-Controlled Hamiltonian control method with nonlinear disturbance observer. Different from the traditional speed-current cascade control for PMSM, the speed-current single loop controller is designed. Firstly, the Port-Controlled Hamiltonian model of PMSM is established, then the Hamiltonian speed controller is designed by using interconnection assignment and damping assignment method. However, the standard Hamiltonian controller cannot achieve the satisfying behavior in the presence of the parameter variations and external disturbance, thus a nonlinear disturbance observer is designed to estimate the lump disturbance of the system, which is used to the feed-forward compensation control. Finally, simulation and experiment on the speed control system of PMSM are implemented. The results show that the proposed method has fast speed control performance and strong robustness for the different disturbance.

This paper takes on the central heating secondary network and establishes a two-level diagnosis model of leakage fault of heating pipe network based on deep belief network (DBN) under the condition of constant and small supply flow quality regulation. Firstly, a leakage condition hydraulic calculation model of the heating pipe network is established with graph theory, which provides the pressure changes of the pressure monitoring points in the pipe network. Then, the first-level diagnostic model for the leakage of the heating pipe network is designed to diagnose the leaky pipe segment by using a deep belief network. Based on the results of the first-level diagnostic model, each leaky pipe segment is treated as a unit and a second-level diagnosis model is then developed to predict the specific leak location. Finally, the model is verified with a branch-pipe network and a loop-pipe network. Experimental results showed that the first-level diagnostic model had a high accuracy rate in the prediction of leaky pipe segments, which was superior to traditional fault diagnosis methods such as BP (Back Propagation Neural Network) and SVM (Support Vector Machines). The second-level diagnostic model can detect the leak location of the leaky pipe with satisfactory results.

A terminal sliding mode control (SMC) method based on nonlinear disturbance observer is investigated to realize the speed and the current tracking control for the permanent magnet synchronous motor (PMSM) drive system in this paper. The proposed method adopts the speed-current single-loop control structure instead of the traditional cascade control in the vector control of the PMSM. First, considering the nonlinear and the coupling characteristic, a single-loop terminal sliding mode controller is designed for PMSM drive system through feedback linearization technology. This method can make the motor speed and current reach the reference value in finite time, which can realize the fast transient response. Although the SMC is less sensitive to parameter uncertainties and external disturbance, it may produce a large switching gain, which may cause the undesired chattering. Meanwhile, the SMC cannot keep the property of invariance in the presence of unmatched uncertainties. Then, a nonlinear disturbance observer is proposed to the estimate the lump disturbance, which is used in the feed-forward compensation control. Thus, a composite control scheme is developed for the PMSM drive system. The results show that the motor control system based on the proposed method has good speed and current tracking performance and strong robustness.

A port-controlled Hamiltonian (PCH) control approach is presented to solve the position tracking problem of gantry robot based on induction motor (IM) drives. First, a robot model is established. Second, a PCH controller is designed to realize accurate position tracking of a gantry robot. For IM drives, it is convenient to choose a direct torque control strategy based on the sliding mode control, which overcomes the higher ripples of torque and flux. Third, a voltage reconstruction technique is introduced to calculate the stator voltage of the IM, which replaces the stator voltage measurement of the IM. Finally, the load torque observer is developed to estimate an unknown load torque. The asymptotic stability of the robot system is proved by the Lyapunov stability theory. Simulation results indicate that the system has excellent position tracking performances and load disturbance attenuation ability.

Person re-identification refers to matching people across disjoint camera views. Most existing person re-identification methods use the same feature descriptors and similarity metrics for all pedestrian pairs. However, these methods ignore that image pairs with different visual consistency conditions are sensitive to different features and metrics. In this paper, we propose to optimally organize multiple similarity measures of global pedestrian and body part pairs with respect to different visual consistency measures (VCM). First, we compute multiple similarity measures for global image and body parts. Then, we group the global image and body part set into three classes based on their VCM value, respectively. Finally, the VCM-specific similarity measures of pedestrian as well as body part pairs are selected and optimally organized to form an ensemble by the reliability estimation and adaptively weighting combination. This method is termed as multiple similarity ensemble based on the visual consistency measure (MSE-VCM). Our contributions are 1) the visual consistency measure method which can select the most appropriate similarity measures for image pairs and 2) optimal organization of these VCM-specific features and metrics on global image and body parts. Extensive experiments on three challenging data sets are conducted. Results demonstrate that our method achieves the comparable performance versus the state-of-the-art methods.