检索结果-内蒙古大学图书馆

Geometry-aware 3D pose transfer using transformer autoencoder

Computational Visual Media 2024年第6期10卷 1063-1078页

作者： Shanghuan Liu Shaoyan Gai Feipeng Da Fazal Waris Key Laboratory of Measurement and Control of Complex Systems of Engineering Ministry of EducationSoutheast UniversityNanjing210096China

3D pose transfer over unorganized point clouds is a challenging generation task,which transfers a source’s pose to a target shape and keeps the target’s *** deep models have learned deformations and used the target’s identity as a style to modulate the combined features of two shapes or the aligned vertices of the source ***,all operations in these models are point-wise and independent and ignore the geometric information on the surface and structure of the input *** disadvantage severely limits the generation and generalization *** this study,we propose a geometry-aware method based on a novel transformer autoencoder to solve this *** efficient self-attention mechanism,that is,cross-covariance attention,was utilized across our framework to perceive the correlations between points at different ***,the transformer encoder extracts the target shape’s local geometry details for identity attributes and the source shape’s global geometry structure for pose *** transformer decoder efficiently learns deformations and recovers identity properties by fusing and decoding the extracted features in a geometry attentional manner,which does not require corresponding information or modulation *** experiments demonstrated that the geometry-aware method achieved state-of-the-art performance in a 3D pose transfer *** implementation code and data are available at https://***/SEULSH/Geometry-Aware-3D-Pose-Transfer-Using-Transfor mer-Autoencoder.

关键词： 3D pose transfer geometry-aware transfor-mer autoencoder cross-covariance attention

来源：评论

学校读者我要写书评

暂无评论

A Degeneracy-Aware LiDAR-Inertial-Visual SLAM for Degenerate Environments

引用

IEEE Sensors Journal 2025年第11期25卷 19376-19385页

作者： Zhu, Hongwei Zhang, Guobao Huang, Yongming Southeast University School of Automation Key Laboratory of Measurement and Control of Complex Systems of Engineering Ministry of Education Nanjing210096 China

Different sensors may experience degeneracy in specific scenarios, which can lead to failures in multi-sensor fusion optimization. To address the challenges of localization and mapping under such degeneracy conditions, this paper proposes a LiDAR-Inertial-Visual SLAM system with integrated degeneracy awareness. In the frontend, the system utilizes the ESIKF for tracking and incorporates a sensor degeneracy detection method to ensure that degraded sensor data do not affect the state error update. In the backend, we replace the conventional point cloud map with an adaptive voxel map to accelerate the retrieval of planar point cloud, and merge LiDAR planar features within a sliding window via region growing algorithm. Furthermore, the point-to-plane distance formula is optimized via the Rayleigh quotient theorem, which reduces the dimensionality of the optimization variables and simplifies the optimization process. Finally, we evaluate the proposed system on several challenging datasets and compare it with various state-of-the-art sensor fusion systems. The experimental results demonstrate that the proposed system achieves high localization accuracy in the majority of challenging scenarios. © 2001-2012 IEEE.

关键词： Inertial navigation systems

来源：评论

学校读者我要写书评

暂无评论

Regional Multi-Agent Cooperative Reinforcement Learning for City-Level Traffic Grid Signal control

引用

IEEE/CAA Journal of Automatica Sinica 2024年第9期11卷 1987-1998页

作者： Yisha Li Ya Zhang Xinde Li Changyin Sun School of Automation Southeast UniversityNanjing 210096 Key Laboratory of Measurement and Control of Complex Systems of Engineering Ministry of EducationSoutheast UniversityNanjing 210096China

This article studies the effective traffic signal control problem of multiple intersections in a city-level traffic system.A novel regional multi-agent cooperative reinforcement learning algorithm called RegionSTLight is proposed to improve the traffic *** a regional multi-agent Q-learning framework is proposed,which can equivalently decompose the global Q value of the traffic system into the local values of several regions Based on the framework and the idea of human-machine cooperation,a dynamic zoning method is designed to divide the traffic network into several strong-coupled regions according to realtime traffic flow *** order to achieve better cooperation inside each region,a lightweight spatio-temporal fusion feature extraction network is *** experiments in synthetic real-world and city-level scenarios show that the proposed RegionS TLight converges more quickly,is more stable,and obtains better asymptotic performance compared to state-of-theart models.

关键词： Human-machine cooperation mixed domain attention mechanism multi-agent reinforcement learning spatio-temporal feature traffic signal control

来源：评论

学校读者我要写书评

暂无评论

Sensorless control of PMSM Drives Using Octant Newton-Raphson Method and Reduced-Order EKF Considering Speed Reversal and Noise

引用

IEEE Transactions on Power Electronics 2025年第8期40卷 10793-10803页

作者： Xiang, Fuyuan Yu, Ke Chen, Congyan Li, Shihua Southeast University School of Automation and the Key Laboratory of Measurement and Control of Complex Systems of Engineering Ministry of Education Nanjing210096 China

For back electromotive force-based sensorless control in permanent magnet synchronous motor (PMSM) drives, the conventional phase-locked loop (PLL) exhibits a ±π position estimation error during speed reversal. This paper presents an octant Newton-Raphson method (ONRM) based PLL, which employs a newly defined objective function and a position octant method. It not only ensures the position estimation converges correctly during speed reversal but also improves dynamic performance compared to the conventional PLL. The applicability condition of the ONRM in the presence of noise is also analyzed in detail. Subsequently, due to the noise sensitivity of the ONRM, a reduced-order extended Kalman filter is designed to filter out the non-Gaussian noise in the position estimation while simultaneously estimating the speed. The effectiveness of the proposed sensorless control scheme is verified on a 1.5 kW PMSM drive platform. © 1986-2012 IEEE.

关键词： Extended Kalman filters

来源：评论

学校读者我要写书评

暂无评论

Efficient Circuit Diagram Retrieval via Graph Representation

Efficient Circuit Diagram Retrieval via Graph Representation

引用

2024 China Automation Congress, CAC 2024

作者： Gao, Ming Chen, Hongcai Zhang, Kanjian Wei, Haikun Key Laboratory of Measurement and Control of Complex System Ministry of Education School of Automation Southeast University Nanjing China

ISBN: (纸本)9798350368604

In the field of electronic engineering, the retrieval of circuit diagrams has become an area of considerable research interest. This is attributed to the critical role it plays in consulting legacy electronic designs and in the identification of design plagiarism. Existing image retrieval techniques, which are proficient in managing natural images, typically involve the conversion of images into feature vectors and retrieve similar images according to the closeness of these vectors. However, these methods encounter limitations when adapted to the more specialized and complex domain of circuit diagrams. This paper introduces an innovative approach to the retrieval of circuit diagrams through the use of a graph representation of these diagrams. This novel approach effectively transforms the retrieval task into a graph retrieval problem. The methodology proposed in this paper is comprised of two main components: A recognition algorithm for circuit diagrams designed to accurately extract the components and the topological structure of the circuit and a retrieval strategy based on graph similarity. This methodology is pioneering in its use of graph representation for circuit diagram retrieval, incorporating topological features that are often not considered by conventional image retrieval methods. The experimental results presented in this paper validate the effectiveness of the proposed approach. © 2024 IEEE.

关键词： Circuit Diagram Recognition Circuit Diagram Retrieval Circuit Graph Representation Graph Retrieval

来源：评论

学校读者我要写书评

暂无评论

Disturbance Observer-based Pointing control of Leighton Chajnantor Telescope

引用

Research in Astronomy and Astrophysics 2024年第1期24卷 135-150页

作者： Weirui Chen Zheng Wang School of Automation Southeast UniversityNanjing 210096China Key Laboratory of Measurement and Control of Complex Systems of Engineering Ministry of EducationSoutheast UniversityNanjing 210096China

Leighton Chajnantor Telescope(LCT), i.e., the former Caltech Submillimeter Observatory telescope, will be refurbished at the new site in Chajnantor Plateau, Chile in 2023. The environment of LCT will change significantly after its relocation, and the telescope will be exposed to large wind disturbances directly because its enclosure will be completely open during observation. The wind disturbance is expected to be a challenge for LCT's pointing control since the existing control method cannot reject this disturbance very well. Therefore, it is very necessary to develop a new pointing control method with good capability of disturbance rejection. In this research, a disturbance observer—based composite position controller(DOB-CPC) is designed, in which an H∞feedback controller is employed to compress the disturbance, and a feedforward linear quadratic regulator is employed to compensate the disturbance precisely based on the estimated disturbance signal. Moreover, a controller switching policy is adopted, which applies the proportional controller to the transient process to achieve a quick response and applies the DOB-CPC to the steady state to achieve a small position error. Numerical experiments are conducted to verify the good performance of the proposed pointing controller(i.e., DOB-CPC) for rejecting the disturbance acting on LCT.

关键词： instrumentation:miscellaneous methods:analytical methods:numerical telescopes

来源：评论

学校读者我要写书评

暂无评论

Balanced prioritized experience replay in off-policy reinforcement learning

引用

Neural Computing and Applications 2024年第25期36卷 15721-15737页

作者： Lou, Zhouwei Wang, Yiye Shan, Shuo Zhang, Kanjian Wei, Haikun Key Laboratory of Measurement and Control of Complex Systems of Engineering Ministry of Education Nanjing210096 China School of Automation Southeast University Nanjing210096 China

In Off-Policy reinforcement learning (RL), the experience imbalance problem can affect learning performance. The experience imbalance problem refers to the phenomenon that the experiences obtained by the agent during the learning process are unevenly distributed in the state space, resulting in the agent’s inability to accurately estimate the value of each potential state. This problem is typically caused by environments with high-dimensional state and action spaces, as well as the exploration–exploitation mechanism inherent in RL. This article proposes a balanced prioritized experience replay (BPER) algorithm based on experience rarity. First, an evaluation metric to quantify experience rarity is defined. Then, the sampling priority of each experience is calculated according to this metric. Finally, prioritized experience replay is performed according to the sampling priority. BPER increases the sampling frequency of high-rarity experiences and decreases the sampling frequency of low-rarity experiences, enabling the agent to learn more comprehensive knowledge. We evaluate BPER on a series of MuJoCo continuous control tasks. Experimental results show that BPER can effectively improve the learning performance while mitigating the impact of the experience imbalance problem. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2024.

关键词： Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

Self-supervised Depth Completion with Adaptive Online Adaptation 23

Self-supervised Depth Completion with Adaptive Online Adapta...

引用

7th International Conference on Machine Learning and Soft Computing, ICMLSC 2023

作者： Chen, Yang Tan, Yang Key Laboratory of Measurement and Control of Complex System of Engineering Ministry of Education School of Automation Southeast University Nanjing210096 China

ISBN: (纸本)9781450398633

Although depth completion has achieved remarkable performance relying on deep learning in recent years, these models tend to suffer a performance degradation when exposed to new environments. Online adaptation, where the model is trained in a self-supervised manner during testing, seems a promising technique to alleviate the drop. However, continuous online adaptation may cause the model to over-adapt and miss the optimal parameters, resulting in oscillation or even degradation of the model performance, in addition to wasting computational resources. Therefore, this paper proposes an adaptive online adaptation framework to make model adaptively trigger online adaptation when encountering novel environments and stop adaptation when model has adapted to the current environment. In detail, we design a trigger to detect the familiarity of model to the current scenario based on image similarity and then launch online adaptation when the scenario is novel. Besides, we elaborate a stopper to monitor the error between prediction and depth input and convert online adaptation to inference when online adaptation does not bring improvement for model. Experimental results demonstrate that our method improves the accuracy of model prediction and increases average running speed of the model on each frame in online adaptation. © 2023 ACM.

关键词： Deep learning

来源：评论

学校读者我要写书评

暂无评论

RSD-Diff: Boundary-Frequency Feature-Enhanced Rail Surface Defect Inspection with Diffusion-Based Transformer Decoder 10

RSD-Diff: Boundary-Frequency Feature-Enhanced Rail Surface D...

引用

10th International Conference on systems and Informatics, ICSAI 2024

作者： Wang, Dezhou Su, Shixiang Lu, Xiaobo School of Automation Southeast University Key Laboratory of Measurement and Control of Complex Systems of Engineering Ministry of Education Nanjing China

ISBN: (纸本)9798331510138

Rail surface defect inspection is an essential task for the railway system. However, due to the similarity of the background and defect foreground pixels, uneven textures, irregular shapes and multiple scales of the existing defects, the inspection accuracy is still required to be further improved. In this paper, we first expand our dataset by GAN-based method for the scarcity of the defect samples. Then, we propose a diffusion-based model to perform the classification and segmentation task. The diffusion model predicts the mask from the Gaussian noise broken ground truth by successive designed transformer decoder. Meanwhile, to better distinguish the pixels in the boundary or other illegible regions, a boundary-frequency feature enhancement module (BFM) is proposed to make the model pay more attention to the defect-related features. Experiments on the rail surface defect dataset revealed that the proposed diffusion-based transformer decoder and BFM can work well together and efficiently improve the rail surface defect segmentation accuracy. © 2024 IEEE.

关键词： Decoding

来源：评论

学校读者我要写书评

暂无评论

Guided Transformative GAN: Joint Face Image Super-Resolution and Frontalization 10

Guided Transformative GAN: Joint Face Image Super-Resolution...

引用

10th International Conference on systems and Informatics, ICSAI 2024

作者： Mu, Minghao Wang, Wei Lu, Xiaobo School of Automation Southeast University Key Laboratory of Measurement and Control of Complex Systems of Engineering Ministry of Education Nanjing China

ISBN: (纸本)9798331510138

Joint face super-resolution and frontalization of non-frontal low-resolution faces is of significant importance for many face analysis applications. However, at the same time, it is a challenging task. In this paper, a novel Guided Transformative Generative Adversarial Network (GT-GAN) is proposed to address the challenge of generating frontal high-resolution face images from non-frontal low-resolution face images. Existing methods usually treat super-resolution and frontalization as two separate tasks, leading to unsatisfactory processing results. GTG AN innovatively achieves both goals within a unified frame-work. Specifically, we propose Profile Attention Guidance (PAG) to generate landmark heatmaps that provide a priori pose information to the generative network and direct its attention to critical facial regions. In addition, an Attention Fusion Transformation Module (AFTM) is proposed to significantly improve the hallucination performance of the network by utilizing the attention mechanism and STNs, combined with sequential feature enhancement strategies. The experimental results show that GT-GAN exhibits superior performance in the joint face super-resolution and frontalization task on the public face dataset Multi-PIE, which provides new ideas and references for this challenging topic. © 2024 IEEE.

关键词： Generative adversarial networks

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：