检索结果-内蒙古大学图书馆

DSTANet: learning a dual-stream model for anomaly driving action detection using spatio-temporal and appearance features

引用

MULTIMEDIA SYSTEMS 2024年第6期30卷 1-16页

作者： Peng, Zhonghua Hu, Rong Li, Zuoyong Zeng, Kun Fujian Univ Technol Sch Comp Sci & Math Fuzhou 350118 Peoples R China Fujian Univ Technol Sch Comp Sci & Math Fujian Prov Key Lab Big Data Min & Applicat Fuzhou 350118 Peoples R China Minjiang Univ Coll Comp & Control Engn Fujian Prov Key Lab Informat Proc & Intelligent Co Fuzhou 350121 Peoples R China

Driving action anomaly detection based on in-cab surveillance video has become the mainstream of current driving action research. However, there is a substantial redundancy of spatio-temporal information in the spatio-temporal action features extracted using only the 3D encoder, which will weaken the distinguishability between normal and anomaly videos, and there will be more difficult samples to be distinguished during the anomaly detection. To alleviate this problem, we propose a dual-stream model combining spatio-temporal and appearance features for anomaly driving action detection. Firstly, the model utilises 3D ResNet and 2D ResNet for feature extraction of spatio-temporal action and appearance information in the video. Next, to fully combine the respective advantages of the cross-dimension information and obtain fused features that can be more easily distinguished, a contrastive learning based alignment approach was used before fusion. In addition, cross-dimensional features are fused by introducing Transformer's cross attention block. Moreover, the supervised contrastive learning objective function is employed to guides the model to distinguish between normal and anomaly driving action. Finally, calculating the feature similarity between the target sample and the normal memory center is utilized to obtain the anomaly score of the target sample. In the experiment, the proposed model DSTANet in this paper achieved the 96.81% AUC in the publicly available DAD dataset, which is higher than the popular models in recent years. Furthermore, feature distinguishability was significantly improved in the visualization experiments. The experiment code and data are publicly available at: https://***/CreatedTRYNA/DCC_Driving_Anomaly_Detection-.

关键词： Driving action anomaly detectiona Dual-stream model Contrastive learning Transformer Cross-attention

来源：评论

学校读者我要写书评

暂无评论

A Generic Multicorrespondence Matching Framework for Reference-Based Image Super-Resolution

引用

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT 2024年 73卷 1页

作者： Zheng, Hong Heng, Huabin Yan, Zhiqiang Zeng, Kun Fang, Jinsheng Qiang, Baohua Guilin Univ Elect Technol Sch Comp Sci & Informat Secur Guangxi Key Lab Image & Graph Intelligent Proc Guilin 541004 Peoples R China Minjiang Univ Sch Comp & Big Data Fujian Prov Key Lab Informat Proc & Intelligent Co Fuzhou 350108 Peoples R China Wuyi Univ Key Lab Cognit Comp & Intelligent Informat Proc Fu Wuyishan 354300 Peoples R China Minnan Normal Univ Sch Comp Sci Zhangzhou 363000 Peoples R China

Image super-resolution (SR) plays an important role in vision-based measurements, and reference-based image super-resolution (RefSR) is one such SR technique that enhances the resolution of a low-resolution (LR) image using an external high-resolution (HR) reference image. However, most existing RefSR methods search for the best-matching correspondence in the reference image, typically a single patch or pixel, to facilitate SR reconstructions, leading to reference underuse and misuse issues, since there are several texture details in different positions of the reference image for pairing with each image pixel or patch in the LR image. To this end, a novel generic framework for multicorrespondence matching is proposed in this article, which aims to identify multiple similar correspondences within the reference image for an LR patch or pixel. In our framework, multiple similar features from the reference image are exploited and aligned, and then they are all aggregated with the LR feature. The contribution of each similar feature from the reference image to the SR reconstruction process is determined by its similarity with the corresponding LR feature. Our designed framework facilitates the search for multiple similar correspondences within the same reference image. This approach enables us to fully harness the external content and textures present in the reference image, effectively mitigating the issues of underuse and misuse of the reference image. Furthermore, our innovative framework seamlessly integrates with existing RefSR models. It effortlessly incorporates existing RefSR models and generates multicorrespondence matching versions of them, leading to notable enhancements in their SR performance. We empirically show that existing state-of-the-art RefSR methods can be consistently improved by our framework, further improving the performance of these methods. Extensive quantitative and qualitative experiments validate the effectiveness of the proposed framework a

关键词： Feature extraction Superresolution Image reconstruction Image resolution Transformers Task analysis Optical flow Generic framework multicorrespondence matching reference-based image super-resolution (RefSR) similarity super-resolution (SR)

来源：评论

学校读者我要写书评

暂无评论

LTDNet-EEG: A Lightweight Network of Portable/Wearable Devices for Real-Time EEG Signal Denoising

引用

IEEE TRANSACTIONS ON CONSUMER ELECTRONICS 2024年第3期70卷 5561-5575页

作者： Huang, Jingwei Wang, Chuansheng Zhao, Wanqi Grau, Antoni Xue, Xingsi Zhang, Fuquan Fuzhou Univ Coll Comp & Big Data Fuzhou 350002 Peoples R China Univ Politecn Cataluna Dept Automat Control Barcelona 08034 Spain Fujian Univ Technol Fujian Prov Key Lab Big Data Min & Applicat Fuzhou 350118 Peoples R China Minjiang Univ Coll Comp & Control Engn Fujian Prov Key Lab Informat Proc & Intelligent Co Fuzhou 350108 Peoples R China Minjiang Univ Fuzhou Technol Innovat Ctr Intelligent Mfg informa Fuzhou 350108 Peoples R China

Portable/Wearable (P/W) electroencephalography (EEG) devices capture and analyze EEG signals, which are widely used in different research fields, such as consumer psychology prediction, attention and fatigue monitoring. Nonetheless, EEG signals obtained through P/W devices are sensitive to environmental conditions and physiological activities, rendering real-time denoising a challenge on computation and memory limited consumer electronics (CE). In this work, we propose a lightweight network of P/W devices for real-time EEG signal denoising (LTDNet-EEG). Specifically, LTDNet-EEG performs automatic linearized modeling of nonlinear EEG signals via Taylor series expansion, then utilizes a Kalman smoothing filter to remove noise from EEG signals and designs a lightweight network based on depthwise separable convolution (DSC) to update Kalman gain and other parameters. Besides, it applies data layout and common subexpression elimination to optimize model structure and code computation respectively. Experiments on the benchmark EEGdenoiseNet database show that LTDNet-EEG outperforms the existing state-of-the-art algorithm. Additionally, the LTDNet-EEG can be effectively implemented on the hardware platform equipped with a 4th generation Raspberry Pi (4GB RAM, 16GB Flash). compared to training and reasoning on CPU, the LTDNet-EEG with optimized approaches achieves approximately a 2.5-fold reduction in execution time which has great potential widely to be used in CE.

关键词： Electroencephalography Brain modeling Noise computational modeling Real-time systems Noise reduction Mathematical models Portable/wearable (P/W) EEG signals denoising consumer electronics (CE) real-time Kalman smoothing filter

来源：评论

学校读者我要写书评

暂无评论

Effective Continual Learning for Text Classification with Lightweight Snapshots 37

Effective Continual Learning for Text Classification with Li...

引用

37th AAAI Conference on Artificial Intelligence (AAAI) / 35th Conference on Innovative Applications of Artificial Intelligence / 13th Symposium on Educational Advances in Artificial Intelligence

作者： Wang, Jue Dong, Dajie Shou, Lidan Chen, Ke Chen, Gang Zhejiang Univ Key Lab Intelligent Comp Based Big Data Zhejiang Hangzhou Peoples R China Zhejiang Univ Coll Comp Sci & Technol Hangzhou Peoples R China

ISBN: (纸本)9781577358800

Continual learning is known for suffering from catastrophic forgetting, a phenomenon where previously learned concepts are forgotten upon learning new tasks. A natural remedy is to use trained models for old tasks as 'teachers' to regularize the update of the current model to prevent such forgetting. However, this requires storing all past models, which is very space-consuming for large models, e.g. BERT, thus impractical in real-world applications. To tackle this issue, we propose to construct snapshots of seen tasks whose key knowledge is captured in lightweight adapters. During continual learning, we transfer knowledge from past snapshots to the current model through knowledge distillation, allowing the current model to review previously learned knowledge while learning new tasks. We also design representation recalibration to better handle the class-incremental setting. Experiments over various task sequences show that our approach effectively mitigates catastrophic forgetting and outperforms all baselines.

关键词： Distillation

来源：评论

学校读者我要写书评

暂无评论

A multi-UAV assisted non-orthogonal multiple access based relay system for minimal average receiving rate maximization

引用

SOFT compUTING 2024年第11-12期28卷 7125页

作者： Tang, Qiang Qu, Xinyu Wang, Jin He, Shiming Changsha Univ Sci & Technol Sch Comp Sci & Commun Engn Changsha Hunan Peoples R China Changsha Univ Sci & Technol Sch Comp & Commun Engn Hunan Prov Key Lab Intelligent Proc Big Data Trans Changsha 410004 Peoples R China

In this paper, we proposed a data relay system consisting of multiple unmanned aerial vehicles (UAVs) and acess point (AP) with the aim of maximizing the minimum receiving rate of Internet of things devices (IoTDs) as the optimization objective. To enhance the utilization of communication resources, we employed non-orthogonal multiple access (NOMA) and successive interference cancellation (SIC) techniques at the AP side. In addition, we used the Wireless power transmission (WPT) technique on the UAV side to charge the IoTDs and prolong their operational lifetime. We formulated the problem as a mixed integer nonlinear programming (MINLP). To solve the MINLP problem, it was decomposed into four subproblems: connection scheduling between UAVs and IoTDs, time slot allocation, UAV trajectory, and AP's power distribution. Finally, we solved all subproblems by using convex optimization techniques, such as successive convex approximations (SCA), and proposed an iterative algorithm. In the simulation, we compared our proposed algorithm with different optimization schemes. The results indicated that our scheme outperforms the others.

关键词： data relay system Non-orthogonal multiple assess UAV trajectory optimization Mimimal average receiving rate maximization

来源：评论

学校读者我要写书评

暂无评论

Dynamic finegrained structured pruning sensitive to filter texture distribution

引用

AI COMMUNICATIONS 2023年第4期36卷 311-323页

作者： Li, Ping Wang, Yuzhe Wu, Cong Kang, Xiatao Changsha Univ Sci & Technol Sch Comp & Commun Engn Changsha Hunan Peoples R China Hunan Prov Key Lab Intelligent Proc Big Data Tran Changsha Hunan Peoples R China

Pruning of neural networks is undoubtedly a popular approach to cope with the current compression of large-scale, high-cost network models. However, most of the existing methods require a high level of human-regulated pruning criteria, which requires a lot of human effort to figure out a reasonable pruning strength. One of the main reasons is that there are different levels of sensitivity distribution in the network. Our main goal is to discover compression methods that adapt to this distribution to avoid deep architectural damage to the network due to unnecessary pruning. In this paper, we propose a filter texture distribution that affects the training of the network. We also analyze the sensitivity of each of the diverse states of this distribution. To do so, we first use a multidimensional penalty method that can analyze the potential sensitivity based on this texture distribution to obtain a pruning-friendly sparse environment. Then, we set up a lightweight dynamic threshold container in order to prune the sparse network. By providing each filter with a suitable threshold for that filter at a low cost, a massive reduction in the number of parameters is achieved without affecting the contribution of certain pruning-sensitive layers to the network as a whole. In the final experiments, our two methods adapted to texture distribution were applied to ResNet Deep Neural Network (DNN) and VGG-16, which were deployed on the classical CIFAR-10/100 and ImageNet datasets with excellent results in order to facilitate comparison with good cutting-edge pruning methods. Code is available at https://***/wangyuzhe27/CDP-and-DTC.

关键词： Threshold neural network penalty pruning

来源：评论

学校读者我要写书评

暂无评论

Biological age construction for prediction of mortality in the Chinese population

引用

GEROSCIENCE 2025年 1-12页

作者： Wang, Kaiyue Gao, Jingli Liu, Ying Liu, Zuyun Li, Yaqi Chen, Shuohua Sun, Liang Wu, Shouling Gao, Xiang Fudan Univ Inst Nutr Sch Publ Hlth Dept Nutr & Food Hyg Shanghai Peoples R China Kailuan Gen Hosp Dept Intens Care Unit Tangshan 063000 Peoples R China Zhejiang Univ Affiliated Hosp 2 Ctr Clin Big Data & Analyt Key Lab Intelligent Prevent Med Zhejiang ProvSch Hangzhou 310058 Zhejiang Peoples R China Zhejiang Univ Sch Med Dept Big Data Hlth Sci Sch Publ Hlth Key Lab Intelligent Prevent Med Zhejiang Prov Hangzhou 310058 Zhejiang Peoples R China Kailuan Gen Hosp Dept Cardiol Tangshan 063000 Peoples R China

Efforts to increase health span bring to light the necessity of constructing biological age (BA) for measuring aging. However, universally adaptive BA needs further investigation, especially among the Chinese population. Therefore, this study aimed to construct BA using routine clinical markers for the Chinese population. Included were two Chinese prospective cohorts, the Kailuan Study I (n = 83,571) for developing BA and the Kailuan Study II (n = 21,229) for validation. Leveraging baseline age-related clinical markers, we developed phenotypic BA (Pheno-Age) using Levine's methods and Klemera-Doubal BA (KDM-Age) using KDM methods and calculated the residuals of regressions of the two BA measured at baseline and during follow-up on chronological age, namely BA acceleration. The predictive performance of baseline, cumulative average, and updated BAs on mortality was evaluated using the area under the curve (AUC) and calibration plots. COX regressions were used to estimate hazard rations (HRs) and 95% confidence intervals (CIs) for the BA acceleration and risk of mortality. During 1,443,857 person-years of follow-up, 12,679 deaths were recorded in the two cohorts. Baseline Pheno-Age and KDM-Age produced desirable predictions for mortality in both the Kailuan Study I (AUC, 0.810 and 0.806, respectively) and the Kailuan Study II (AUC, 0.867 and 0.819, respectively). Calibration plots showed reasonable agreement between predicted and observed probabilities. The pooled multivariable-adjusted HRs (95% CIs) for per standard deviation increment of baseline Pheno-Age acceleration and mortality was 1.24 (1.18, 1.30), and for KDM-Age acceleration was 1.16 (1.10, 1.21). Similar predictive performance and association were observed when using cumulative average or updated BA. The associations were stronger in the adults aged <= 60 years, smokers, and drinkers, relative to their counterparts (P for interaction <0.05 for all). Pheno-Age and KDM-Age, developed and validated in the two

关键词： Biological age acceleration Physical measures Blood biomarkers Mortality

来源：评论

学校读者我要写书评

暂无评论

Experimental study on cutting force and surface quality of Aramid fiber reinforced polymer based on minimum quantity lubrication

引用

PROCEEDINGS OF THE INSTITUTION OF MECHANICAL ENGINEERS PART C-JOURNAL OF MECHANICAL ENGINEERING SCIENCE 2025年第9期239卷 3448-3464页

作者： Wang, Lin Li, Na Shi, Wentian Li, Ru Bu, Kunting Zhang, Hanyu Qilu Inst Technol Sch Intelligent Mfg & Control Engn 3028 Jingshidong Rd Jinan 250200 Peoples R China Shandong Prov Key Lab Ind Big Data & Intelligent M Jinan Peoples R China Beijing Technol & Business Univ Sch Comp & Artificial Intelligence Beijing Peoples R China

In order to explore the milling mechanism of aramid fiber reinforced polymer under minimal quantity lubrication and improve the milling quality of aramid fiber reinforced polymer, the milling experiments of aramid fiber reinforced polymer were carried out under dry cutting and minimal quantity lubrication, and the cutting force, fiber section, and surface quality were compared and analyzed. The experiment results show that in the dry cutting state, the surface burr is more, the surface quality is poor, the fiber section is rough, and the resin matrix has a melting trend, accompanied by the phenomenon of chip wrapping around the tool, tooltip collapse, and coating peeling. In minimal quantity lubrication, the surface burr is reduced, the surface quality is good, the fiber section is neat, and the resin interface is clean and tidy. The resin matrix has no melting trend, the tooltip is only slightly worn, and the phenomenon of chip entanglement is effectively avoided. Further research shows that the minimal quantity lubrication can reduce the temperature and friction, enhance the stiffness of the fiber, and change the fracture model of the fiber to a certain extent, that is, from bending to tensile fracture (or accompanied by tensile phenomenon) to shear fracture, which also leads to the overall high cutting force. The total milling force synthesized by the average value increases from 2.45 N to 20.93 N, with a maximum increase of 88.29%. However, due to the cooling and friction reduction of the lubricating fluid and the chip removal and diversion, the overall fluctuation of the cutting force under the minimal quantity lubrication condition is reduced. The overall cutting force is more stable than that under dry cutting conditions. The results show that the milling quality is improved under minimal quantity lubrication, the cutting force is more significant but more stable, and the milling process is more stable. This study has important reference significance for mill

关键词： Aramid fiber reinforced polymer minimal quantity lubrication fiber fracture model cutting force surface quality

来源：评论

学校读者我要写书评

暂无评论

EATN: An Efficient Adaptive Transfer Network for Aspect-Level Sentiment Analysis

引用

IEEE TRANSACTIONS ON KNOWLEDGE AND data ENGINEERING 2023年第1期35卷 377-389页

作者： Zhang, Kai Liu, Qi Qian, Hao Xiang, Biao Cui, Qing Zhou, Jun Chen, Enhong Univ Sci & Technol China Sch Data Sci Anhui Prov Key Lab Big Data Anal & Applicat Hefei 230027 Anhui Peoples R China Univ Sci & Technol China Sch Comp Sci & Technol Anhui Prov Key Lab Big Data Anal & Applicat Hefei 230027 Anhui Peoples R China Ant Financial Serv Grp Hangzhou 310000 Zhejiang Peoples R China

Aspect-level sentiment analysis is a granular emotional classification task that refers to identifying sentiment polarities towards aspects in a sentence. Although previous research has reached a great achievement, this task remains very challenging. First, previous approaches only focus on one specific domain, which lacks the capability of transferring to other domains. Moreover, the majority of prior studies ignore the direct relationship between aspects and the corresponding sentiment words. To this end, in this paper, we propose a novel model named Efficient Adaptive Transfer Network (EATN) for aspect-level sentiment analysis which emphasizes the need to incorporate the correlation among multiple domains. The proposed EATN provides a Domain Adaptation Module (DAM) to learn common features from the sufficiently labelled source domain and guide the target domain's classification performance. Specifically, DAM comprises two special tasks, with one sentiment classification task aiming to learn sentiment knowledge and the other domain classification task focusing on learning domain-invariant features. Here, we adopt a multiple-kernel selection method to further reduce the feature discrepancy among domains. Besides that, EATN contains a novel aspect-oriented multi-head attention to capture the direct associations between the aspects and the contextual sentiment words, which is beneficial to learn the aspect-aware semantic knowledge. Extensive experiments on six public datasets with two granularities, compared with current state-of-the-art methods, demonstrate the effectiveness and universality of our method.

关键词： Aspect-level sentiment analysis domain adaptation transfer learning multi-head attention mechanism

来源：评论

学校读者我要写书评

暂无评论

Diagram Perception Networks for Textbook Question Answering via Joint Optimization

引用

INTERNATIONAL JOURNAL OF compUTER VISION 2024年第5期132卷 1578-1591页

作者： Ma, Jie Liu, Jun Chai, Qi Wang, Pinghui Tao, Jing Xi An Jiao Tong Univ Sch Cyber Sci & Engn Minist Educ Key Lab Intelligent Networks & Network Secur Xian 710049 Peoples R China Xi An Jiao Tong Univ Sch Comp Sci & Technol Shaanxi Prov Key Lab Big Data Knowledge Engn Xian Peoples R China

Textbook question answering requires a system to answer questions with or without diagrams accurately, given multimodal contexts that include rich paragraphs and diagrams. Existing methods usually utilize a pipelined way to extract the most relevant paragraph from multimodal contexts and only employ convolutional neural networks to comprehend diagram semantics under the supervision of answer labels. The former will result in error accumulation, while the latter will lead to poor diagram understanding. To provide a remedy for the above issues, we propose an end-to-end DIagraM Perception network for textbook question answering (DIMP), which is jointly optimized by the supervision of relation predicting, diagram classification, and question answering. Specifically, knowledge extracting is regarded as a sequence classification task and optimized through the supervision of answer labels to alleviate error accumulation. To capture diagram semantics effectively, DIMP uses an explicit relation-aware method that first parses a diagram into several graphs under specific relations and then grasps the information propagation within them. Evaluation on two benchmark datasets shows that our method achieves competitive or better results without large data pre-training and constructing auxiliary tasks compared with current state-of-the-art methods. We provide comprehensive ablation studies and thorough analyses to determine what factors contribute to this success. We also make in-depth analyses for relational graph learning and joint optimization.

关键词： Textbook question answering Multi-modality Diagram understanding Visual question answering Multimodal learning

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：