检索结果-内蒙古大学图书馆

Hyperbolic cosine transformer for LiDAR 3D object detection

Optoelectronics Letters 2025年

作者： Tong Jigang Yang Fanhang Yang Sen Du Shengzhi Tianjin Key Laboratory for Control Theory & Applications in Complicated Systems and Intelligent Robot Laboratory， Tianjin University of Technology Department of Electrical Engineering， Tshwane University of Technology

Recently， Transformer has achieved great success in computer vision. However， it is constrained because the spatial and temporal complexity grows quadratically with the number of large points in 3D object detection applications from point clouds. Previous point-wise methods are suffering from time consumption and limited receptive fields to capture information among points. To address these limitations， we propose the cosh-attention， which reduces the computation complexity of space and time from the quadratic order to linear order with respect to the number of points. In the cosh-attention， the traditional softmax operator is replaced by non-negative ReLU activation and hyperbolic-cosine-based operator with re-weighting mechanism. Then based on the key component， cosh-attention， we present a two-stage hyperbolic cosine transformer (ChTR3D) for 3D object detection from point clouds. It refines proposals by applying cosh-attention in linear computation complexity to encode rich contextual relationships among points. Extensive experiments on the widely used KITTI dataset and Waymo Open Dataset demonstrate that， compared with vanilla attention， the cosh-attention significantly improves the inference speed with competitive performance. Experiment results show that， among two-stage state-of-the-art methods using point-level features to refine proposals， the proposed ChTR3D is the fastest one.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Liquid refractive index measurement method based on orthogonal polarization laser self-mixing interference

引用

Optoelectronics Letters 2025年

作者： ZHAO Yan WANG Haipeng DU Shengzhi TONG Jigang KANG Jun School of Electrical Engineering and Automation， Tianjin University of Technology Tianjin Key Laboratory for Control Theory and Applications in Complicated Systems Department of Electrical Engineering， Tshwane University of Technology

In this paper， a novel method combining orthogonal polarization laser self-mixing interference is proposed. The method utilizes a rotating cuvette to measure the refractive index of liquids at different concentrations. The cuvette is filled with the liquid to be measured and rotated by a certain angle. The change in the number of interference fringes， caused by comparing an empty cuvette with a liquid-filled cuvette， is used to calculate the refractive index of the liquid. A four-fold logic subdivision algorithm is then used to improve measurement resolution. The experimental results show that for pure water and different NaCl and glucose solutions concentrations， the average relative errors are 0.47%， 0.59%， and 2.17%， respectively， with the maximum relative error within ±2.54%. The standard deviation of all solutions is less than 3.4%.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Weak feedback self-mixing interference fringe slope discrimination method based on deep learning

引用

Optoelectronics Letters 2025年

作者： ZHAO Yan LIN Maohua DU Shengzhi TONG Jigang LIU Bin HAN Fangfang School of Electrical Engineering and Automation， Tianjin University of Technology Tianjin Key Laboratory for Control Theory and Applications in Complicated Systems Department of Electrical Engineering， Tshwane University of Technology

In order to identify the tilt direction of the self-mixing interference (SMI) signals under weak feedback regime interfered by noise， a deep learning method is proposed. The one-dimensional U-Net (1D U-Net) neural network can discriminate the direction of self-mixing fringes accurately and quickly. For experimental SMI signals， the 1D U-Net can be used for discriminant direction after a one-step normalization. Simulation and experimental results show that the proposed method is suitable for SMI signals with noise within the whole weak feedback regime， and can maintain a high discrimination accuracy for signals interfered by 5dB noise. Combined with fringe counting method， accurate and rapid SMI signal displacement reconstruction can be realized.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Differential Enhancement and Commonality Fusion for RGBT Tracking 20

Differential Enhancement and Commonality Fusion for RGBT Tra...

引用

20th IEEE International Conference on Mechatronics and Automation, ICMA 2023

作者： Yang, Jianrong Dong, Enzeng Tong, Jigang Yang, Sen Zhang, Zufeng Li, Wenyu Tianjin University of Technology Tianjin Key Laboratory for Control Theory & Applications in Complicated Systems Intelligent Robot Laboratory 391 Binshui West Road Tianjin China Nankai University College of Artificial Intelligence 94 Weijin Road Tianjin China

ISBN: (纸本)9798350320831

Target tracking is a fundamental task and an important part in the field of computer vision. The image information contained in the visible (RGB) modality and thermal infrared (T) modality has differences. RGBT tracking utilizes this difference to fuse the information of the two modalities, fully combining the uniqueness and complementarity of the information of the two modalities. How to maximize the uniqueness of individual modalities in RGBT tracking and how to effectively combine the complementarity of the two modalities is the key issue in RGBT tracking. For the combination of the uniqueness and complementarity of the two modalities, a novel Differential Enhancement and Commonality Fusion Network (DECFNet) is designed in this paper, in which the main components include a feature enhancement module for the visible modality and a feature enhancement module for the thermal infrared modality, as well as a fusion module that effectively fuses the information of both. The results on three RGB T tracking benchmark datasets demonstrate the effectiveness of the proposed method. © 2023 IEEE.

关键词： Target tracking

来源：评论

学校读者我要写书评

暂无评论

Deep Neural Network Based Object Detection Algorithm With optimized Detection Head for Small Targets 20

Deep Neural Network Based Object Detection Algorithm With op...

引用

20th IEEE International Conference on Mechatronics and Automation, ICMA 2023

作者： Zhu, Yongchang Dong, Enzeng Tong, Jigang Yang, Sen Zhang, Zufeng Li, Wenyu Tianjin University of Technology Tianjin Key Laboratory for Control Theory & Applications in Complicated Systems Intelligent Robot Laboratory 391 Binshui West Road Tianjin China Nankai University College of Artificial Intelligence 94 Weijin Road Tianjin China

ISBN: (纸本)9798350320831

Target detection is a significant research direction in the realm of computer vision, aimed at analyzing and localizing targets in input images to obtain their categories and locations. However, this task is faced with several challenges in complex scenarios, such as large variations in target scales, similarity to the background, dense number, and overlapping with each other, making it difficult for the target detection model to extract different scales and discriminative target features adaptively, leading to low detection accuracy and weak generalization ability. To tackle the aforementioned problems, this study proposes a novel approach by optimizing the detection head of the model, invoking decoupled detection heads, dividing the classification and localization tasks, and optimizing for small targets. This approach effectively reduces the false detection rate and the missed detection rate, enabling the network to better discriminate and recognize distant objects as well as objects of relatively small size. Moreover, the proposed approach outperforms the yolov7 model by 0.8% in the COCO dataset, not only improving the accuracy of detecting distant and small targets but also exhibiting better performance in conventional target detection. Additionally, this approach is capable of handling small targets, overlapping targets, or relatively dense targets, significantly expanding its practical use range. © 2023 IEEE.

关键词： Object detection

来源：评论

学校读者我要写书评

暂无评论

A Global Perception Attention-based Network for Point Cloud Completion 20

A Global Perception Attention-based Network for Point Cloud ...

引用

20th IEEE International Conference on Mechatronics and Automation, ICMA 2023

作者： Wu, Fengyong Dong, Enzeng Tong, Jigang Yang, Sen Li, Wenyu Tianjin University of Technology Tianjin Key Laboratory for Control Theory & Applications in Complicated Systems Intelligent Robot Laboratory 391 Binshui West Road Tianjin China Nankai University College of Artificial Intelligence 94 Weijin Road Tianjin China

ISBN: (纸本)9798350320831

The purpose of point cloud completion is designed to high-precision complete the complete shape from local observations. However, the previous approaches have tended to focus on a good perception of local details and lack a better overall representation of the point cloud. To solve this problem, we propose a global attention point cloud reconstruction method based on SnowflakeNet. We designed a global focus module to integrate the global structure information of the original point cloud on the basis of the local shape context features, so as to enhance the perception ability of the network to the local fine structure and the overall shape of the point cloud. At the same time, we improved the point cloud generation part of the network and added SPD that can re-optimize the position of the completion point cloud, so as to achieve more accurate completion results. We conducted a series of experiments on the Completion3D dataset and confirmed that our network outperformed the original algorithm. © 2023 IEEE.

关键词： Computer vision

来源：评论

学校读者我要写书评

暂无评论

A multi-directional controllab.e multi-scroll conservative chaos generator:Modelling,analysis,and FPGA implementation

引用

Chinese Physics B 2021年第2期30卷 232-239页

作者： En-Zeng Dong Rong-Hao Li Sheng-Zhi Du Tianjin Key Laboratory for Control Theory&Applications in Complicated Systems Tianjin University of TechnologyTianjin 300384China Department of Electrical Engineering Tshwane University of TechnologyPretoria 0001South Africa

Combing with the generalized Hamiltonian system theory,by introducing a special form of sinusoidal function,a class of n-dimensional(n=1,2,3)controllab.e multi-scroll conservative chaos with complicated dynamics is *** dynamics characteristics including bifurcation behavior and coexistence of the system are analyzed in detail,the latter reveals abundant coexisting ***,the proposed system passes the NIST tests and has been implemented physically by *** to the multi-scroll dissipative chaos,the experimental portraits of the proposed system show better ergodicity,which have potential application value in secure communication and image encryption.

关键词： multi-directional controllab.e multi-scroll conservative chaos coexisting flows field programmable gate array(FPGA)

来源：评论

学校读者我要写书评

暂无评论

Multilevel Decomposition Time Aggregation Graph Convolution Networks for Skeleton-Based Action Recognition 20

Multilevel Decomposition Time Aggregation Graph Convolution ...

引用

20th IEEE International Conference on Mechatronics and Automation, ICMA 2023

作者： Li, Wenhua Dong, Enzeng Tong, Jigang Yang, Sen Zhang, Zufeng Li, Wenyu Tianjin University of Technology Tianjin Key Laboratory for Control Theory & Applications in Complicated Systems Intelligent Robot Laboratory 391 Binshui West Road Tianjin China Nankai University College of Artificial Intelligence 94 Weijin Road Tianjin China

ISBN: (纸本)9798350320831

Skeleton-based human action recognition has become a popular topic among researchers. This is because using skeletal data provides a robust solution to problems encountered in complex environments, such as changes in perspective and background interference. The robustness of skeletal data enables recognition methods to focus on more specific and relevant features. We propose a model called multilevel decomposition time aggregation graph convolution network (MDT-GCN), which utilizes a multilevel graph convolution kernel to capture higher-order spatial dependence relationships between joints. This is achieved by decomposing a human topology graph into smaller graphs, each of which has its own graph convolution kernel. To further enhance the performance of our model, we employ a two-flow framework and channel topology refinement strategy. Our experiments on the NTU-RGB+D60 and NTU-RGB+D120 datasets demonstrate that our MDT-GCN network outperforms the previous algorithm and significantly improves the accuracy of action recognition. © 2023 IEEE.

关键词： Convolution

来源：评论

学校读者我要写书评

暂无评论

A Single-Frame Face Anti-Spoofing Algorithm With Circular CDC and Multi-Scale Spatial Attention 20

A Single-Frame Face Anti-Spoofing Algorithm With Circular CD...

引用

20th IEEE International Conference on Mechatronics and Automation, ICMA 2023

作者： He, Shan Dong, Enzeng Tong, Jigang Yang, Sen Liang, Xuehui Li, Wenyu Tianjin University of Technology Tianjin Key Laboratory for Control Theory & Applications in Complicated Systems Intelligent Robot Laboratory 391 Binshui West Road Tianjin China Nankai University College of Artificial Intelligence 94 Weijin Road Tianjin China

ISBN: (纸本)9798350320831

Face anti-spoofing is to prevent face images with attack properties from entering face recognition and causing confusion or spoofing of the face recognition function. Since most face presentation attacks do not possess true deep information, some recent works use this information as supervisory signal. In this paper, we continue such a pattern by proposing a circular CDC (Central Difference Convolution) with larger sensory field and contextual semantic awareness, and use it in a backbone network for improving the network model's ability to characterize texture features and multi-scale information. Combined with the spatial attention method concat multiple scale feature maps are used to fuse the feature map information from multiple scales. We conduct a series of experiments on OULU-NPU and demonstrate that our network accuracy is better than the original algorithm, especially in terms of generalizability across devices and different environments. © 2023 IEEE.

关键词： Face recognition

来源：评论

学校读者我要写书评

暂无评论

Modeling and control Design for Quadrotor Transporting System With a Cable-Suspended Rigid Payload 20

Modeling and Control Design for Quadrotor Transporting Syste...

引用

20th IEEE International Conference on Mechatronics and Automation, ICMA 2023

作者： Xia, Tianqi Dong, Enzeng Yang, Sen Tong, Jigang Du, Shengzhi Tianjin University of Technology Tianjin Key Laboratory for Control Theory & Applications in Complicated Systems Intelligent Robot Laboratory 391 Binshui West Road Tianjin China Tshwane University of Technology Department of Electrical Engineering Pretoria0001 South Africa

ISBN: (纸本)9798350320831

In this paper, facing the aerial transportation task with a large volume cargo, the system of quadrotor with a cable-suspended rigid payload is investigated. The dynamic model of proposed system is built with Lagrange d'Alembert principle. The dynamic model is linearized to design the feedback control method. Finally, two groups of numerical simulation with proposed feedback controller are designed to demonstrate the effectiveness of system model and the results of simulation illustrate satisfactory performance. © 2023 IEEE.

关键词： Dynamic models

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：