检索结果-内蒙古大学图书馆

RepColor: deep coloring algorithm combining semantic categories

Optoelectronics Letters 2025年

作者： Lei Sun Dongjie Ju School of Electrical Engineering and Automation， Tianjin University of Technology Tianjin Key Laboratory for Control Theory and Complicated Industry Systems， Tianjin University of Technology

Image coloring is an inherently uncertain and multimodal problem. By inputting a grayscale image into a coloring network， visually plausible colored photos can be generated. Conventional methods primarily rely on semantic information for image colorization. Although effective in coloring images with clear semantic information， these methods still suffer from color contamination and semantic confusion. This is largely due to the limited capacity of convolutional neural networks to effectively learn deep semantic information inherent in *** this paper， we propose a network structure that addresses these limitations by leveraging multi-level semantic information classification and fusion. Additionally， we introduce a global semantic fusion network to combat the issues of color contamination. The proposed coloring encoder accurately extracts object-level semantic information from *** further enhance visual plausibility， we employ a self-supervised adversarial training method. We train the network structure on various datasets with varying amounts of data and evaluate its performance using the ImageNet validation set and COCO validation set. Experimental results demonstrate that our proposed RepColor can generate more realistic images compared to previous approaches， showcasing its high generalization ability.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Composite-mask GAN based on refined optical flow and disparity map for SLAM visual odometry

引用

Optoelectronics Letters 2025年

作者： JI Yuehui JIANG Jingwei LIU Junjie SONG Yu GAO Qiang School of Electrical Engineering and Automation， Tianjin University of Technology Tianjin Key Laboratory of Control Theory and Application for Complex Systems

Although deep learning methods have been widely applied in slam visual odometry over the past decade with impressive improvements， the accuracy remains limited in complex dynamic environments. In this paper， a composite mask-based generative adversarial network is introduced to predict camera motion and binocular depth maps. Specifically， a perceptual generator is constructed to obtain the corresponding parallax map and optical flow from between two neighboring frames. Then， an iterative pose improvement strategy is proposed to improve the accuracy of pose estimation. Finally， a composite mask is embedded in the discriminator to sense structural deformation in the synthesized virtual image， thereby increasing the overall structural constraints of the network model， improving the accuracy of camera pose estimation， and reducing drift issues in the Visual Odometer. Detailed quantitative and qualitative evaluations on the KITTI dataset show that the proposed framework outperforms existing conventional， supervised learning and unsupervised depth VO methods， providing better results in both pose estimation and depth estimation.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Study on the Falling Off Position and Heat Shift of Bonding Wire Based on Temperature Distribution Coefficient 17th

Study on the Falling Off Position and Heat Shift of Bonding ...

引用

17th Annual Conference of China Electrotechnical Society, CES 2022

作者： Dong, Chao Hu, Jingwei Du, Mingxing Yin, Jinliang Wei, Yifan Liu, Xueting Tianjin Key Laboratory of Control Theory and Application of Complex Systems Tianjin University of Technology Tianjin300384 China Tianjin Sino-German University of Applied Sciences Tianjin300384 China

ISBN: (纸本)9789819904501

When the bonding wire of IGBT module is aged, most traditional researches only consider the change rule of the electrothermal parameters caused by the number of dropped bonding wires, ignoring the influence caused by the difference of the distribution of the dropped bonding wires. Based on the temperature distribution coefficient k, the heat source offset caused by the bond wire falling off position is analyzed in this paper. Firstly, the temperature distribution coefficient k in the healthy state is determined by capturing the temperature at the bottom, middle and edge of the chip in a fixed position with a fiber grating sensor. Then, the FEM is used to analyze the heat source offset caused by the different number and location of bond wire falling off, and explore the law of temperature distribution coefficient k and heat source offset. Finally, an experimental platform is set up to verify the rule of heat source offset and temperature distribution coefficient k, and the specific bond wire falling off position is determined according to the values and changes of temperature distribution coefficient k on both sides of IGBT module. © 2023, Beijing Paike Culture Commu. Co., Ltd.

关键词： Finite element method

来源：评论

学校读者我要写书评

暂无评论

Multi-Feature Fusion Transformer for Natural Language Inference 5

Multi-Feature Fusion Transformer for Natural Language Infere...

引用

5th International Conference on Robotics, Intelligent control and Artificial Intelligence, RICAI 2023

作者： Sun, Lei Li, Xiangyang Yan, Hengxin School of Electrical Engineering and Automation Tianjin University of Technology Tianj in Key Laboratory for Control Theory and Complicated Industry Systems Tianjin China School of Electrical Engineering and Automation Tianjin University of Technology Tianjin China

ISBN: (纸本)9798350357950

Natural Language Inference (NLI) is a branch of Natural Language Processing (NLP) whose main task is to determine the relationship between two sentences. Such tasks essentially use pre-trained models to ensure accuracy, and many applications are found in human-computer dialog and question-answering systems. However, in some cases where small devices are deployed offline, lightweight model implementations are often required to save computational resources. To address this problem and ensure more efficient inference capabilities, this paper describes how to improve the preexisting Transformer and proposes a Multi-Feature Fusion Transformer Network for NLI processing. The model integrates sequence features and introduces a priori information to enhance local features. It also achieves a comprehensive fusion of sequence information, local features, and non-local features, thus compensating for many potential shortcomings of Transformer. At the same time, it maintains the lightweight feature of the model and facilitates the implementation of downstream tasks. We verified the above point by experiments, our model has better inference performance compared to previous models used for the same task. The accuracy in the SNLI dataset reached 89.0%, while the matched and mismatched versions of MultiNLI reached 79.3% and 78.7%. © 2023 IEEE.

关键词： Feature Fusion Natural language inference Natural language processing Transformer

来源：评论

学校读者我要写书评

暂无评论

Hyperbolic cosine transformer for LiDAR 3D object detection

引用

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

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

Research on hearing-impaired EEG Emotion Recognition Based on Deep Residual Shrinkage Networks

Research on hearing-impaired EEG Emotion Recognition Based o...

引用

第42届中国控制会议

作者： Meiyi Cheng Qiang Gao Zhongli Bai Yirui Chen Yang Liu Yu Song Tianjin Key Laboratory for Control Theory and Applications in Complicated Systems

With the development of affective computing and computer science,electroencephalogram(EEG) based emotion recognition has attracted much more *** this paper,we collected EEG signals from fifteen healthy people and fifteen hearing-impaired people when they were watching five kinds of emotional pictures(fear,anger,sadness,happiness and neutral).The collected EEG signals are preprocessed to remove artifacts by independent component analysis(ICA).Then the differential power spectral density,entropy,and wavelet entropy features were extracted(PSD,DE,WE).The Deep Residual Shrinkage Networks(DRSNs) were composed of residual building units(RBUs) and an attention mechanism was used to capture the representative features,and then made a *** classification results prove that using DE as a feature for classification is better than other features,with an average accuracy being 76.35% and 80.47% for the hearing-impaired group and the normal group,***,from the brain topography,we found that there are certain differences in brain function between hearing-impaired and healthy people.

关键词：

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

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 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

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：