To enhance the estimation accuracy and dynamic performance of sensorless surface-mounted permanent magnet synchronous motor(SPMSM) drives,a sensorless control scheme based on generalized super-twisting observer(GSTO) ...
To enhance the estimation accuracy and dynamic performance of sensorless surface-mounted permanent magnet synchronous motor(SPMSM) drives,a sensorless control scheme based on generalized super-twisting observer(GSTO) and nonsmooth controller is ***,a GSTO for back electromotive force(back-EMF) estimation is *** with the conventional super-twisting observer,the GSTO has a faster convergence rate and stronger robustness due to the additional *** a linear extended state observer(LESO) is adopted to estimate the position,speed,and lumped disturbance at the same ***,a non-smooth composite speed controller is designed by combining the disturbance feed-forward *** with the conventional PI speed controller,the non-smooth controller has a shorter settling time and a better disturbance rejection ***,the effectiveness of the proposed method is verified by simulation results.
We construct the first markerless deformable interaction dataset recording interactive motions of the hands and deformable objects, called HMDO (Hand Manipulation with Deformable Objects). Our motivation is the curren...
Fall events have unique dynamic features,which are not fully utilized by existing fall detection *** on video understanding,we propose Fall-LSTM to learn such features pertinently without additional ***-LSTM is compos...
Fall events have unique dynamic features,which are not fully utilized by existing fall detection *** on video understanding,we propose Fall-LSTM to learn such features pertinently without additional ***-LSTM is composed of CNN-LSTM framework and two excitation modules i.e.,Spatial Attention Module(SAM) and Temporal Location Module(TLM).SAM provides spatial constraints on motion for feature layers through foreground extraction and spatial *** emphasizes frames with high probability of fall events to LSTM by inferring the rate and trend of motion in *** results show that our proposed modules significantly improve the performance of LSTM model,outperforming the state-of-theart methods on two public Fall Detection Datasets i.e.,Le2 i and UR.
Cross-domain synergy and intelligence will become important features of future warfare. Cross-domain guidance is one of the typical forms of cross-domain synergy. It can significantly improve the ability to respond qu...
详细信息
This paper investigates the speed regulation control of switched reluctance motor(SRM) *** improve the antidisturbance performance of SRM,a composite non-smooth control strategy is ***,the structure of SRM is analyzed...
This paper investigates the speed regulation control of switched reluctance motor(SRM) *** improve the antidisturbance performance of SRM,a composite non-smooth control strategy is ***,the structure of SRM is analyzed,and a simplified nonlinear model is obtained based on a segmented representation of the varying phase inductance.A virtual control function is introduced to represent the nonlinear part of the torque equation,whose inverse function is cascaded to linearized the nonlinear ***,a generalized proportional integral observer(GPIO) is constructed to estimate the lumped disturbance of the system,which is used for feedforward compensation ***,a composite speed controller is designed based on a combination of finite time proportional feedback and feed-forward compensation based on GPIO(FTP+GPIO).The speed error closed-loop system can be regarded as a first-order finite time controlsystem with bounded *** analysis shows that the proposed scheme can improve the anti-disturbance performance of the closed-loop *** effectiveness of the proposed method is verified by simulation ***,it is compared with proportional feedback combining feed-forward compensation(P+GPIO) method and proportional integral(PI) control method.
In this paper, we propose a pylon reconstruction method based on Neural Radiance Fields (NeRF) technology. The advantage of this method lies in its ability to reconstruct pylons from images, with a multi-view model ma...
详细信息
ISBN:
(数字)9798350361896
ISBN:
(纸本)9798350361902
In this paper, we propose a pylon reconstruction method based on Neural Radiance Fields (NeRF) technology. The advantage of this method lies in its ability to reconstruct pylons from images, with a multi-view model maintaining good performance even under complex lighting and occlusion conditions. We demonstrate the process of using NeRF to reconstruct electric power towers through experiments. During the experimental process, we collect images of electric power towers from different environments and angles, then apply NeRF to process these images, using deep learning to learn and generate a network model capable of expressing the scene. The experimental results show that our method can complete the reconstruction of scenes with only image inputs, and the reconstructed models have a certain accuracy in terms of geometric details and textures.
This paper investigates the problem of Traffic Signal control (TSC) in large-scale road networks. In extensive road networks, it is customary to define each intersection as an agent, however, the issue of partial obse...
详细信息
ISBN:
(数字)9798350373691
ISBN:
(纸本)9798350373707
This paper investigates the problem of Traffic Signal control (TSC) in large-scale road networks. In extensive road networks, it is customary to define each intersection as an agent, however, the issue of partial observability is particularly prominent. In this paper, Predictive State Representation (PSR) is employed to address the challenge of partial observability in large-scale multi-agent systems. A Multi-agent Deep Reinforcement Learning (DRL) model based on PSR called PSR-XLight is proposed in Large-Scale TSC systems. Multi-agent PSR is conducted with centralized training and independent filtering which overcome the challenge of prohibitive computations when the number of agents is large. Parameters sharing is adopted between each agent’s PSR model to enhance learning efficiency and facilitate utilization in large-scale multi-agent environments. Each agent undergoes independent DRL training and execution while parameters sharing is adopted. Experiments are conducted on real-world road networks and a large-scale road network comprising 1000 intersections.
Video-based person re-identification (Re-ID) aims at matching the video snippets of the same person across multiple cameras. The ubiquitous appearance misalignment is a critical challenge in video person re-identifica...
Video-based person re-identification (Re-ID) aims at matching the video snippets of the same person across multiple cameras. The ubiquitous appearance misalignment is a critical challenge in video person re-identification. Existing alignment-based methods rely on off-the-shelf human parsing models and cannot handle anomalous appearance information (e.g., obstacles and pedestrian interference) in video sequences. In this paper, we propose Anomaly-Aware Semantic Self-Alignment (ASSA), a novel video-based person Re-ID framework that seeks out body parts without prior human topology information and learns part-based feature representations against anomalous information. The proposed ASSA performs part classifier training and part-aligned representation learning alternately. For the classifier training, we design a Salient Region Extraction module to segment the entire foreground from the background in each input frame. Furthermore, a novel Anomaly-Aware Refinement module is proposed to suppress the influence of anomalous interference. Extensive experiments on three prevalent benchmarks demonstrate the effectiveness and superiority of the proposed framework.
This paper presents an new approach to enhance the detection of smoking behavior using object detection neural networks, focusing on the challenge of small object detection, namely cigarettes in video frames. We intro...
详细信息
ISBN:
(数字)9798331510138
ISBN:
(纸本)9798331510145
This paper presents an new approach to enhance the detection of smoking behavior using object detection neural networks, focusing on the challenge of small object detection, namely cigarettes in video frames. We introduce a Multi-Spatial Pooling (MSP) bottleneck mechanism integrated into a YOLOv8 backbone, which significantly improves the network's ability to discern smaller objects without compromising real-time inference capabilities. This enhancement facilitates the extraction of relevant features across various scales and efficiently handles the computational overhead, thereby boosting the detection accuracy and operational efficiency. Our results demonstrate a substantial improvement in performance metrics over traditional and existing neural network approaches, making this method highly suitable for real-time applications in diverse scenarios.
A distributed fault-tolerant formation control law is designed in this paper for multi-agent systems under external disturbances and actuator faults including time-varying loss of effectiveness faults. The initial pos...
详细信息
暂无评论