检索结果-内蒙古大学图书馆

FSS: algorithm and neural network accelerator for style transfer

Science China(Information Sciences) 2024年第2期67卷 253-266页

作者： Yi LING Yujie HUANG Yujie CAI Zhaojie LI Mingyu WANG Wenhong LI Xiaoyang ZENG State Key Laboratory of ASIC & System Fudan University Shanghai Explore X Technology Co. Ltd.

Neural networks(NNs), owing to their impressive performance, have gradually begun to dominate multimedia processing. For resource-constrained and energy-sensitive mobile devices, an efficient NN accelerator is necessary. Style transfer is an important multimedia application. However, existing arbitrary style transfer networks are complex and not well supported by current NN accelerators, limiting their application on mobile devices. Moreover, the quality of style transfer needs improvement. Thus, we design the Fast Style system(FSS), where a novel algorithm and an NN accelerator are proposed for style transfer. In FSS, we first propose a novel arbitrary style transfer algorithm, Fast Style. We propose a light network that contributes to high quality and low computational complexity and a prior mechanism to avoid retraining when the style changes. Then, we redesign an NN accelerator for Fast Style by applying two improvements to the basic NVIDIA deep learning accelerator(NVDLA) architecture. First, a flexible dat FSM and wt FSM are redesigned to enable the original data path to perform other operations(including the GRAM operation)by software programming. Moreover, statistics and judgment logic are designed to utilize the continuity of a video stream and remove the data dependency in the instance normalization, which improves the accelerator performance by 18.6%. The experimental results demonstrate that the proposed Fast Style can achieve higher quality with a lower computational cost, making it more suitable for mobile devices. The proposed NN accelerator is implemented on the Xilinx VCU118 FPGA under a 180-MHz clock. Experimental results show that the accelerator can stylize 512×512-pixel video with 20 FPS, and the measured performance reaches up to 306.07 GOPS. The ASIC implementation in TSMC 28 nm achieves about 22 FPS in the case of a 720-p video.

关键词： neural network accelerator style transfer neural network deep learning

来源：评论

学校读者我要写书评

暂无评论

Modelling and experimental investigation of micro-dimpled structures milling with spiral trajectory tool reciprocating motion

引用

Chinese Journal of Aeronautics 2025年第2期38卷 577-596页

作者： Guangzhou WANG Linjie ZHAO Qi LIU Xiguang LI Yazhou SUN Mingjun CHEN School of Mechatronics Engineering Harbin Institute of TechnologyHarbin 150001China Centre for Precision Manufacturing DMEMUniversity of StrathclydeGlasgow G11XJUK State Key Laboratory of Robotics and System Harbin Institute of TechnologyHarbin 150001China

To mill fine and well-defined micro-dimpled structures,a machining manner of spiral trajectory tool reciprocating motion,where the tool repeats the process of‘feed milling–retract–cutting feed–feed milling again’along the spiral trajectory,was *** the kinematics analysis,it is found that the machining quality of micro-dimpled structures is highly dependent on the machining trajectory using spiral trajectory tool reciprocating *** reveal this causation,simulation modelling and experimental studies were carried out.A simulation model was developed to quantitatively and qualitatively investigate the influence of the trajectory discretization strategies(constant-angle and constant-arc length)and parameters(discrete angle,discrete arc length,and pitch)on surface texture and residual height of micro-dimpled ***,micro-dimpled structures were milled under different trajectory discretization strategies and parameters with spiral trajectory tool reciprocating motion.A comprehensive comparison between the milled results and simulation analysis was made based on geometry accuracy,surface morphology and surface roughness of milled ***,the errors and factors affecting the above three aspects were *** results demonstrate both the feasibility of the established simulation model and the machining capability of this machining way in milling high-quality micro-dimpled *** trajectory tool reciprocating motion provides a new machining way for milling micro-dimpled structures and micro-dimpled functional *** an appropriate machining trajectory can be generated based on the optimized trajectory parameters,thus contributing to the improvement of machining quality and efficiency.

关键词： Tool reciprocating motion Spiral trajectory Micro-dimpled structure Micromachining Simulation modelling with diamond tool Surface texture

来源：评论

学校读者我要写书评

暂无评论

Data-Driven Robust Model Predictive Control for Pose Tracking of Redundant Manipulators

Data-Driven Robust Model Predictive Control for Pose Trackin...

引用

2024 IEEE International Conference on robotics and Biomimetics, ROBIO 2024

作者： Chen, Zhaoyang Ni, Fenglei Min, Kang Liu, Hong Harbin Institute of Technology State Key Laboratory of Robotics and System Heilongjiang Province Harbin150001 China

ISBN: (纸本)9781665481090

In this paper, we present a data-driven robust model predictive control (DD-RMPC) method for pose tracking of redundant manipulators. Firstly, a MPC-based trajectory tracking framework is established, in which the joint limits at three different levels (angle, velocity, and acceleration) are satisfied. Secondly, to deal with the system model's uncertainty, accelerate the error convergence speed, and reduce tracking errors, a novel data-driven RMPC is proposed, in which the system's input-output data are used to compensate for the conservatism of the system. Finally, the simulation results on a seven degrees of freedoms (DOFs) redundant manipulator show that DD-RMPC provides a faster error convergence rate and achieves smaller pose tracking errors than that of the comparison methods. © 2024 IEEE.

关键词： Predictive control systems

来源：评论

学校读者我要写书评

暂无评论

A Kinematic Decoupling Whole-body Control Method for a Mobile Humanoid Upper Body Robot

A Kinematic Decoupling Whole-body Control Method for a Mobil...

引用

2023 International Conference on Frontiers of robotics and Software Engineering, FRSE 2023

作者： Ju, Fengjia Jin, Hongzhe Zhao, Jie Harbin Institute of Technology State Key Laboratory of Robotics and System Harbin China

ISBN: (纸本)9798350301113

The mobile humanoid upper body robot combines the humanoid upper body robot with the mobile platform. It has high redundancy and can complete complex tasks. However, coupling joints, including the waist joints and the mobile platform, exist in the configuration of the robot. During the task execution, the motion of the end-effectors on the dual arms are not independent, but interfere with each other through the motion of the waist and the mobile platform. Therefore, we need to decouple the waist and mobile platform from the dual arms. Based on the pseudo-inverse method of the redundant manipulators, we consider the motion of the coupling joints as a disturbance term and propose a kinematic decoupling whole-body control method. The method realizes the decoupling and cooperation of dual arms, avoids the interference of the inverse kinematics model of the dual arms, and fully releases the application potential of the high redundancy. The decoupling method allows the waist joints and the mobile platform to move randomly without affecting the task operation at the end-effectors, and to further plan their movements separately in order to smooth the whole body motion. The simulations on the mobile humanoid upper body robot verifies our method. © 2023 IEEE.

关键词： End effectors

来源：评论

学校读者我要写书评

暂无评论

A Rigid-Soft Pneumatic Wrist with Fixed Rotation Axes and Active Jamming Variable Stiffness Mechanisms 16th

A Rigid-Soft Pneumatic Wrist with Fixed Rotation Axes and Ac...

引用

16th International Conference on Intelligent robotics and Applications, ICIRA 2023

作者： Ding, Kehan Jiang, Li Zhen, Ruichen State Key Laboratory of Robotics and System Harbin Institute of Technology Harbin China

ISBN: (纸本)9789819964888

The wrist can greatly enhance the manipulative performance of dexterous hands. In addition, due to the flexibility of soft materials, the soft wrist has obvious advantages in environmental adaptability and safety. However, existing soft wrists often have small motion curvatures and cannot modulate stiffness. Inspired by the human wrist, we proposed a rigid-soft pneumatic wrist design with fixed axes, realizing variable stiffness through jamming mechanisms. The wrist is composed of two rotatory joints in series to achieve pronation/supination and flexion/extension motions. Each joint is driven by two opposite-mounted actuators, resulting in active bidirectional movement. The pre-stretched mounting of pneumatic actuators eliminates the stiffness torque effects. Both joints have independent jamming stiffening mechanisms controlled by air pressure and require no additional power source, which makes the stiffness control more flexible. Experimental results demonstrated that the maximum output torques of the two joints reached 1.18 Nm and 1.40 Nm, and the jamming mechanism increased the stiffness of the joints by 55.3% and 67.2% in the practical test, respectively. The proposed rigid-soft wrist provides a feasible solution for enhancing the manipulative ability of soft dexterous hands. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023.

关键词： Pneumatic actuators

来源：评论

学校读者我要写书评

暂无评论

Optimizing Dataset Configuration for Improving Convolutional Neural Networks Performance in Chest X-ray Image Classification 8

Optimizing Dataset Configuration for Improving Convolutional...

引用

8th International Conference on Intelligent Computing and Signal Processing, ICSP 2023

作者： He, Zengmin Ren, Yi Wang, Honggang Harbin Institute of Technology State Key Laboratory of Robotics and System Harbin China

ISBN: (纸本)9798350302455

Chest X-ray (CXR) is essential for physicians to diagnose lung diseases in clinical medicine. With the development of computer and deep learning techniques, creating CXR datasets to train convolutional neural networks (CNNs) has become a popular research topic. However, there are still challenges in creating datasets due to class imbalance and the ratio setting of datasets. This study investigated the impact of class imbalance and the ratio of training, validation, and test sets on CNNs classification performance by optimizing the dataset configuration. This was achieved by directly modifying the ratio, applying oversampling based on the adaptive contrast enhancement (ACE) algorithm, and random undersampling to balance the dataset classes, followed by modifying the ratio again. Therefore, seven datasets were obtained, which were utilized to individually train four CNNs based on transfer learning and fine-tuning techniques. Evaluation metrics based on the confusion matrix were utilized to demonstrate the enhanced classification performance of CNNs. The results indicated that at least 17% of the accuracy of CNNs trained by the dataset with modified ratios was improved as compared with the dataset with an unreasonable initial configuration. Additionally, the overall evaluation metrics of CNNs were further improved by balancing the dataset classes and modifying the ratios. ChexNet demonstrated the best classification performance among the four CNNs, as evidenced by the area under the receiver operating characteristic (ROC) curves (AUC). Furthermore, ChexNet was trained by a class balanced dataset with a ratio of 7:2:1, resulting in the best evaluation metrics, including F1 Score, balanced accuracy, accuracy, recall, specificity, and precision, which were 97.81%, 97.78%, 97.78%, 98.73%, 96.84%, and 96.89%, respectively. Therefore, optimizing dataset configuration can effectively improve the performance of CNNs, providing empirical guidance for researchers in creating dat

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Fast measurement and prediction method for electromagnetic susceptibility of receiver

引用

Journal of systems Engineering and Electronics 2024年第2期35卷 275-285页

作者： CHEN Yan LU Zhonghao LIU Yunxia State Key Laboratory of Complex Electromagnetic Environment Effects on Electronics and Information System College of Electronic Science and TechnologyNational University of Defense TechnologyChangsha 410073China State Key Laboratory of Astronautic Dynamics Xi’an 710043China

Aiming at evaluating and predicting rapidly and accurately a high sensitivity receiver’s adaptability in complex electromagnetic environments,a novel testing and prediction method based on dual-channel multi-frequency is proposed to improve the traditional two-tone ***,two signal generators are used to generate signals at the radio frequency(RF)by frequency scanning,and then a rapid measurement at the intermediate frequency(IF)output port is carried out to obtain a huge amount of sample data for the subsequent ***,the IF output response data are modeled and analyzed to construct the linear and nonlinear response constraint equations in the frequency domain and prediction models in the power domain,which provide the theoretical criteria for interpreting and predicting electromagnetic susceptibility(EMS)of the *** experiment performed on a radar receiver confirms the reliability of the method proposed in this *** shows that the interference of each harmonic frequency and each order to the receiver can be identified and predicted with the sensitivity *** on this,fast and comprehensive evaluation and prediction of the receiver’s EMS in complex environment can be efficiently realized.

关键词： electromagnetic susceptibility(EMS) receiver dualchannel multi-frequency nonlinear response frequency domain power domain

来源：评论

学校读者我要写书评

暂无评论

Data Augmentation for Improving CNNs in Medical Image Classification 8

Data Augmentation for Improving CNNs in Medical Image Classi...

引用

8th International Conference on Intelligent Computing and Signal Processing, ICSP 2023

作者： Ren, Yi He, Zengmin Deng, Yang Huang, Bo Harbin Institute of Technology State Key Laboratory of Robotics and System Harbin China

ISBN: (纸本)9798350302455

Medical image is essential for physicians to diagnose diseases. And convolutional neural networks (CNNs) have gained momentum for computer-aided diagnosis (CAD) medical image. However, there are still challenges in CNNs related to a lack of data and class-imbalanced datasets. Data augmentation is utilized to address the issues mentioned above. In this study, the Pseudo-color enhancement algorithm (OCEAN and TWILIGHT), common linear transform, CLAHE enhancement, and K-means clustering algorithm were applied to pneumonia X-ray images for data augmentation. Thus, five processed datasets were finally obtained. The results indicated that five data-augmentation methods effectively improved the performance of CNNs for the pneumonia X-ray images binary classification detection, in which the CLAHE enhancement and OCEAN enhancement methods had the accuracy and F1 Score with 97.42%, 97.19%, and 97.42%, 98.36%, respectively. DenseNet121 showed the best classification performance of four CNNs utilized in this study, as evidenced by the area under the receiver operating characteristic (ROC) curves (AUC). To verify the generalizability of the two enhancement methods, CLAHE and OCEAN enhancement methods were applied to the Magnetic Resonance Imaging (MRI) dataset of Alzheimer's disease, which had the accuracy and F1 Score with 98.15% , 98.15%, and 97.22%, 97.68%, respectively. Therefore, these simple and efficient data-augmentation methods effectively improve the performance of CNNs for diagnosing medical image, which can provide theoretical reference for physicians to diagnose the above two diseases, and advance the application of clinical medicine. © 2023 IEEE.

关键词： Magnetic resonance imaging

来源：评论

学校读者我要写书评

暂无评论

A Novel Approach for Solving Time-Varying Linear Equations and Its Application in Motion Tracking of Redundant Robots

A Novel Approach for Solving Time-Varying Linear Equations a...

引用

2023 International Conference on Frontiers of robotics and Software Engineering, FRSE 2023

作者： Yin, Hong Jin, Hongzhe Zhao, Jie Harbin Institute of Technology State Key Laboratory of Robotics and System Harbin China

ISBN: (纸本)9798350301113

Solving time-varying linear equations is a common problem in many fields. In this article, a novel approach is proposed based on the cost function constructed by the mixed error, which is made up of the residual error and its integration. The merits of the novel approach include high accuracy, noise immunity, and asymptotic convergence. Since the derivative information of time-varying signals is used in the proposed method, a proportional-integral differentiator is designed based on the mixed estimation error for the first derivative estimation. The proposed differentiator has a high degree of precision, asymptotic convergence, and no chattering. In addition, theoretical analyses prove the convergence of the proposed optimal strategy and differentiator. Simulation results validate the effectiveness and superiority of the proposed strategy and the designed differentiator. Finally, the proposed approach is employed in the motion tracking of redundant robots. © 2023 IEEE.

关键词： Errors

来源：评论

学校读者我要写书评

暂无评论

GCA-Net: A Global Context Aggregation Network for Effective Optical Flow 49

GCA-Net: A Global Context Aggregation Network for Effective ...

引用

49th Annual Conference of the IEEE Industrial Electronics Society, IECON 2023

作者： Xie, Tao Gao, Jinghan Wang, Ke Li, Ruifeng Harbin Institute of Technology State Key Laboratory of Robotics and System Harbin China

ISBN: (纸本)9798350331820

Optical flow seeks to estimate the per-pixel 2D motion between two frames by identifying corresponding pixels. Current flow estimators typically involve per-pixel feature extraction, multi-scale 4D correlation volume construction, and iterative flow field updates through a Conv-GRU module. However, the locality of convolutional features in these methods renders the calculated correlations vulnerable to different noises. In addition, the Conv-GRU module of these methods is only executed with a convolution layer, which is incapable of exploiting context clues from larger window sizes even inside the image being queried itself, thus raising it more difficult for the model to process images with challenging regions, e.g., textureless areas. To this end, we introduce GCA-Net, a global context aggregation network for credible yet effective optical flow estimation. More specifically, we propose a highly efficient multi-scale transformer (MSFormer) layer which enables the per-pixel feature to aggregate long-range information from other features, hence building more accurate 4D correlation volumes. Besides, we develop an attention-enhanced Conv-GRU block (AttGRU) that can incorporate information alongside a larger context window even within itself empowering our network to estimate optical flow in even the most challenging regions. Experimentally, we demonstrate that GCA-Net outperforms previous state-of-the-art methods by large margins on Sintel (Final) and KITTI 2015 (background and foreground) benchmarks. © 2023 IEEE.

关键词： Optical flows

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：