检索结果-内蒙古大学图书馆

SGformer:Boosting transformers for indoor lighting estimation from a single image

Computational Visual Media 2024年第4期10卷 671-686页

作者： Junhong Zhao Bing Xue Mengjie Zhang Centre for Data Science and Artificial Intelligence&School of Engineering and Computer Science Victoria University of WellingtonWellington 6012New Zealand

Estimating lighting from standard images can effectively circumvent the need for resourceintensive high-dynamic-range(HDR)lighting ***,this task is often ill-posed and challenging,particularly for indoor scenes,due to the intricacy and ambiguity inherent in various indoor illumination *** propose an innovative transformer-based method called SGformer for lighting estimation through modeling spherical Gaussian(SG)distributions—a compact yet expressive lighting *** from previous approaches,we explore underlying local and global dependencies in lighting features,which are crucial for reliable lighting ***,we investigate the structural relationships spanning various resolutions of SG distributions,ranging from sparse to dense,aiming to enhance structural consistency and curtail potential stochastic noise stemming from independent SG component *** harnessing the synergy of local–global lighting representation learning and incorporating consistency constraints from various SG resolutions,the proposed method yields more accurate lighting estimates,allowing for more realistic lighting effects in object relighting and *** code and model implementing our work can be found at https://***/junhong-jennifer-zhao/SGformer.

关键词： lighting estimation transformer spherical Gaussian(SG) augmented reality

来源：评论

学校读者我要写书评

暂无评论

Formation Control Algorithm for Multiple Mobile Robots Based on Fuzzy Mathematics

Informatica (Slovenia)

引用

Informatica (Slovenia) 2024年第10期48卷 51-64页

作者： Fan, Bingqian Artificial Intelligence and Big Data College Hebei University of Engineering Science Langfang050000 China

The formation of multiple mobile robots suffers from unsmooth formation trajectory tracking and large formation control errors. A three closed-loop sliding mode formation control method was proposed to solve these problems, which achieved stable three closed-loop control systems from both linear and angular velocity directions. Meanwhile, fuzzy theory was introduced to design a fuzzy sliding mode control model for multiple mobile robots, which made the gain switching smoother, and the problems of unsmooth and interference were solved in trajectory tracking. The results showed that the linear velocity fluctuation of the research designed model was controlled within 5 seconds, and the angular velocity fluctuation was controlled within 1 second. The overall mean value of position control indicators was 17.08, which was smaller than the comparison model. The average value of the comprehensive control index of position and speed was 2.8, which was smaller than the comparison model. The model designed in this research had higher control accuracy and more stable control effects, ensuring the efficient and high-quality execution of tasks by robots. This research provides a technical basis for the formation motion control of multiple mobile robots. © 2024 Slovene Society Informatika. All rights reserved.

关键词： Angular velocity

来源：评论

学校读者我要写书评

暂无评论

Intelligent Human Behaviour Detection Leveraging ConvLSTM and LRCN Deep Learning Approaches 10

Intelligent Human Behaviour Detection Leveraging ConvLSTM an...

引用

10th International Conference on Signal Processing and Communication, ICSC 2025

作者： Chakradhar Amingad, S. Janmanchi, Harika Sultan, Mohammed Shalini, K. Hariharan, S. Maddala, Bhanu Prakash Artificial Intelligence and Data Science Vardhaman College of Engineering Telangana Hyderabad India Artificial Intelligence and Data Science Vardhaman College of Engineering Hyderabad India

ISBN: (纸本)9798331511890

This research delves into intelligent human behavior detection from video sequences by employing advanced deep learning methodologies. The study harnesses the capabilities of Convolutional Long Short-Term Memory (ConvLSTM) and Long-term Recurrent Convolutional Networks (LRCN) architectures, which are adept at capturing both temporal and spatial dynamics in video data. Using the extensive Kinetics dataset, which includes a wide range of human activities and contexts, the video data was converted into sequential image frames for model training. The ConvLSTM and LRCN models were implemented and trained with the TensorFlow/Keras framework on GPU-accelerated systems. During training, both models exhibited effective learning, with improved accuracy and reduced loss. However, validation metrics revealed fluctuations, indicating possible overfitting and challenges in generalizing to new data. To address these issues, regularization techniques such as dropout and early stopping were applied, alongside comprehensive hyperparameter tuning. Additionally, the study investigated the use of transfer learning to enhance model performance. This research lays the groundwork for sophisticated video-based human behavior detection systems, with applications in surveillance, healthcare, and human-computer interaction. Future work will aim to bolster model robustness and broaden the scope of action recognition. © 2025 IEEE.

关键词： Feature Selection

来源：评论

学校读者我要写书评

暂无评论

Advanced Parkinson’s Disease Classification using Hybrid Model: SVM and Random Forest 3

Advanced Parkinson’s Disease Classification using Hybrid Mo...

引用

3rd International Conference on Intelligent data Communication Technologies and Internet of Things, IDCIoT 2025

作者： Archana, S.H. Rajeshkannan, M. Anish, M.J. Sandeep, Pasupuleti Artificial Intelligence and Data Science Saveetha Engineering College Chennai India

ISBN: (纸本)9798331527549

Parkinson’s disease (PD) is a progressive and chronic neurodegenerative condition that significantly impairs motor and cognitive functions, posing a considerable burden on global healthcare systems. Early and precise diagnosis is vital to improving patient outcomes, yet traditional diagnostic techniques often encounter limitations due to overlapping symptoms with other neurological disorders and the subtle nature of early PD indicators. This research introduces a machine learning-driven approach that employs a hybrid ensemble model integrating Support Vector Machine (SVM) and Random Forest classifiers to enhance diagnostic accuracy by analyzing vocal and acoustic features. To address dataset imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) is applied, enabling the model to identify meaningful patterns within underrepresented data. Principal Component Analysis (PCA) is utilized to reduce dimensionality, preserving critical data variance while minimizing computational complexity and mitigating overfitting. Hyperparameter tuning further optimizes each classifier, bolstering the model’s accuracy and robustness. Comparative analyses demonstrate that this hybrid ensemble model effectively distinguishes PD patients from healthy individuals, showcasing its potential as a reliable predictive tool in clinical applications. This study highlights the promise of advanced machine learning methodologies in supporting early PD detection, offering clinicians a valuable resource for timely intervention and improved patient care. © 2025 IEEE.

关键词： hybrid classification Machine learning models Parkinson’s Disease principal component analysis (PCA) random forest algorithm support vector machine (SVM)

来源：评论

学校读者我要写书评

暂无评论

Deep learning framework for automatic detection and classification of sleep apnea severity from polysomnography signals

引用

Neural Computing and Applications 2024年第28期36卷 17483-17493页

作者： Raja Brundha, A. Lakshmi Sangeetha, A. Balajiganesh, A. Department of Artificial Intelligence and Data Science Sri Sai Ram Engineering College Chennai India Department of Computer Science Engineering Velammal Engineering College Chennai India

Sleep apnea (SA) is a sleep-related breathing disorder characterized by breathing pauses during sleep. A person’s sleep schedule is significantly influenced by that person’s hectic lifestyle, which may include unhealthy eating habits and their line of work. Polysomnographic (PSG) sleep studies examine sleep-related disorders by recording various biosignals from the human body. However, SA classification methods could be more robust in terms of performance because they rely on feature-engineering strategies or employ a particular signal from PSG recording for diagnosis. This study aims to classify the severity of SA according to the apnea–hypopnea index (AHI) into normal, mild, and moderate-to-severe groups using oxygen saturation (SpO2), electroencephalogram (EEG), and electrocardiogram (ECG) signals. The proposed deep neural network (DNN)-bidirectional long short-term memory (DNN-BiLSTM) framework addresses the issue of low detection accuracy in analysis. The DNN-BiLSTM approach employs features extracted from a multiscale dilation attention 1D convolutional neural network (MSDA-1DCNN) as input for detection and classification purposes. The MSDA-1DCNN network extracts deep features from processed SpO2, EEG, and ECG signals. The developed firefly combined electric fish optimization (FCEFO) algorithm improves performance by optimizing the hidden neuron count of the DNN and the learning rate of the BiLSTM framework. The performance measures proved the effectiveness of the model over conventional machine and deep learning approaches. With the integration of deep features, the proposed DNN-BiLSTM model provides enhanced performance in terms of accuracy and precision. Thus, the proposed approach is progressive and can be used for medical diagnosis. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2024.

关键词： Electrocardiograms

来源：评论

学校读者我要写书评

暂无评论

Leveraging Advanced Deep Learning Architectures for Precise Knee Osteoporosis Prediction: A Paradigm Shift in Diagnostic Healthcare 2

Leveraging Advanced Deep Learning Architectures for Precise ...

引用

2nd International Conference on Advances in Computation, Communication and Information Technology, ICAICCIT 2024

作者： Keerthika, S. Janani, A. Krishnasamy, Lalitha DevaBrindha, S. Priyanka, S. Dharun, S.R. Department of Artificial Intelligence Kongu Engineering College Erode India Department of Artificial Intelligence and Data Science Kongu Engineering College Erode India Department of Artificial Intelligence and Data Science Nandha Engineering College TamilNadu Perundurai India

ISBN: (纸本)9798331541217

Osteoporosis is a condition characterized by reduced bone mass and deterioration of bone tissue. It affects the people by increasing bone fragility and a higher risk of fractures. Globally, it affects approximately 200 million people, predominantly postmenopausal women and the elderly. Early detection of osteoporosis is crucial for reducing fracture risks, improving patient outcomes, and lowering healthcare costs. However, traditional diagnostic methods are often invasive and costly, particularly in developing regions. Current methods lack efficiency and affordability, underscoring the need for noninvasive and cost-effective diagnostic solutions. This paper explores the use of various deep learning techniques such as CNN, VGG16, ResNet, DenseNet, AlexNet, and GoogLeNet. Using deep learning models on knee X-ray images, this effort provides a simple, non-invasive approach to early osteoporosis detection. The models are assessed using accuracy metrics, and cross-validation on training, validation, and test datasets is used to validate their performance. The models' classification performance is also evaluated using F1-scores, accuracy, recall, and confusion matrices. The models are employed to analyze the images, with the aim of improving diagnostic accuracy and enabling early intervention. The overall efficiency is improved by using data augmentation approaches, optimizing the models and implementing efficient preprocessing steps. This study's primary objective is to develop and evaluate deep learning models that can use knee X-ray images to identify osteoporosis early, offering a quicker, non-invasive, and easier method of diagnosis. © 2024 IEEE.

关键词： Bone Fragility Deep Learning Early Detection Healthcare Efficiency Osteoporosis

来源：评论

学校读者我要写书评

暂无评论

A Long Short-Term Memory Network-Based Approach for Predicting the Trends in the S&P 500 Index

引用

Journal of The Institution of Engineers (India): Series B 2024年第1期105卷 19-26页

作者： Siddesh, G.M. Sekhar, S. R. Mani Srinivasa, K.G. Department of Artificial Intelligence & Data Science M S Ramaiah Institute of Technology Bangalore India Department of Information Science and Engineering M S Ramaiah Institute of Technology Bangalore India Data Science and Artificial Intelligence Programme International Institute of Information Technology Naya Raipur India

The price level of a considered security in the stock market is largely determined by perceived consumer demand. Variations in price level in the stock market are essentially a manifestation of public psychology aiming to either cut losses or maximize profits. The assumption that market movements are controlled by a guiding trend is proved by establishing that the time-series variations of daily S&P 500 indices are a non-stationary process using the Dickey–Fuller test. This paper presents a long short-term memory (LSTM) network adaptation of an autoregressive predictive model for S&P 500 indices. Later, an LSTM network-based model is devised to learn the underlying patterns that are responsible for driving the trends in market dynamics. The proposed model accepts as input at each time-step a vector of prior S&P indices over the past two weeks and estimates the current market index. The network weights assume the correct settings by training the model using daily S&P indices over the past ten years. The proposed model decisively captures subtleties in market trends, which are in turn verified using the Granger causality approach. The result demonstrates that the proposed model achieved an accuracy of 82.6%, whereas random forest results in 79.12%. The results obtained are truly the latest in this line of predictive models and reaffirm that LSTM networks are powerful learning models capable of deciphering intricate underlying patterns. © 2023, The Institution of Engineers (India).

关键词： Long short-term memory

来源：评论

学校读者我要写书评

暂无评论

Quasi-Parallel Dense Video Captioning: A Novel Approach to Achieving Ground Truth Event Captions using Hierarchical Attention 2

Quasi-Parallel Dense Video Captioning: A Novel Approach to A...

引用

2nd International Conference on Intelligent data Communication Technologies and Internet of Things, IDCIoT 2024

作者： Kalai Selvi, T. Naveenkumar, S. Thompson, Michael Kesavaraj, R. Easwari Engineering College Artificial Intelligence and Data Science Chennai India

ISBN: (纸本)9798350327533

Video captioning is a complex process, involving the automatic generation of descriptive, natural language narratives for video content. The current landscape of video captioning models primarily emphasizes crafting coherent, grammatically sound captions, yet often falls short in comprehensively capturing the essential events and actions portrayed in the video. Capturing ground truth events and localization of the generated predictions stays a challenge. This study introduces a novel video captioning model, Quasi-Parallel Dense Video Captioning (QPDVC), which directly targets authentic events within the video. QPDVC operates by concurrently analysing random frames, generating captions for each frame. These captions are then amalgamated, localized, and condensed to construct a cohesive story or summary. The distinct design of QPDVC is meticulously crafted to achieve elevated precision, rendering it particularly apt for applications such as sophisticated AI-based CCTV surveillance and advancing the realm of video captioning research. This approach not only enhances the interpretative capability of video captioning systems but also ensures a more accurate portrayal of the actual content. The model's framework provides an efficient mechanism to process and synthesize information, thereby offering a more detailed and nuanced understanding of the video's content. As the necessity for accurate video understanding grows across various domains, from security surveillance to content analysis, QPDVC's precision-driven methodology represents a significant step forward in the evolution of video captioning technology. © 2024 IEEE.

关键词： Attention Computer Vision Decoder Parallelization Deep Learning Encoder Feature Extraction Machine Learning Video Captioning

来源：评论

学校读者我要写书评

暂无评论

Advancements in Breast Imaging: CNN-based Thermogram Quality Enhancement 2

Advancements in Breast Imaging: CNN-based Thermogram Quality...

引用

2nd International Conference on Intelligent data Communication Technologies and Internet of Things, IDCIoT 2024

作者： Venkatachalam, Nirmala Risha Dharshini, S. Charanya, B. Ashwati, K. Easwari Engineering College Artificial Intelligence and Data Science Chennai India

ISBN: (纸本)9798350327533

Breast thermography is a medical imaging technique that is free from radiation and is non-invasive which apprehends variations in temperature on the breast surface. The significant information, like early detection of potential abnormalities for breast health assessments, are provided. This study proposes an innovative approach to enhance the visual quality of breast thermograms utilizing a specialized Convolutional Neural Networks. The proposed methodology involves a tailored image enhancement model that is CNN-based to ascend the resolution, contrast, and overall quality of breast thermogram images. This research not only advances the state-of-the-art in breast thermogram image processing but also holds promise for enhancing the diagnostic capabilities of thermography in early breast cancer detection. The proposed CNN-based framework demonstrates its potential as a valuable tool in the realm of medical image analysis, contributing to improved clinical decision-making and patient outcomes in breast health assessments. © 2024 IEEE.

关键词： Abnormality Detection Breast Health Assessment Breast Thermography Convolutional Neural Network Diagnostic accuracy Image Enhancement Non-invasive Radiation free

来源：评论

学校读者我要写书评

暂无评论

AI Driven Restoration of Damaged Paintings and Historical Artifacts Using Generative Adversarial Networks 3

AI Driven Restoration of Damaged Paintings and Historical Ar...

引用

3rd International Conference on Disruptive Technologies, ICDT 2025

作者： Buvaneshwaran, B. Ashwin, V. Abishekrupan, B.S. Sasidharan, Sandhya Revathi, K.P. Easwari Engineering College Department of Artificial Intelligence and Data Science Chennai India

ISBN: (纸本)9798331519582

Restoration of damaged art pieces, paintings, and artifacts is vital to the preservation of cultural heritage. Conventional techniques are done by hand, hence time-consuming and inconsistent. This paper proposes an AI-based restoration system with Generative Adversarial Networks (GANs) and Context Encoders to automate and improve the restoration process of digital artifacts. The process includes data collection, training, and model optimization to achieve good-quality outputs. The Context Encoder generates predictions for lost parts from adjacent image features, and the GAN model increases texture and color for full integration. The platform is easy to use with convenient uploads and rapid restoration. The model is assessed with quantitative metrics and expert judgment, showing better restoration accuracy. This AI-based solution is a productive and scalable alternative to human restoration and serves museums, historians, and cultural heritage. © 2025 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：