检索结果-内蒙古大学图书馆

High-Performance Image Restoration with SpectraNet: A Lightweight CNN Architecture

Procedia computer science 2025年 258卷 2268-2277页

作者： Nitesh Patnaik Rishi Raj Piyush Kumar Singh Shreekara Mahapatra Debashis Pani Ravi Shankar Singh Dept. of Computer Science and Engineering Kalinga Institute Of Industrial Technology Bhubaneswar 751024 India Dept. of Computer Science Central University of South Bihar Bihar 824236 India Dept. of Computer Science and Engineering Indian Institute of Technology BHU Varanasi 221005 India

Deep learning-based image restoration faces significant challenges when deployed on resource-constrained platforms due to the computational demands and large number of parameters in existing models. The high computational load and extensive memory requirements make it difficult to implement these models on devices with limited processing power and storage capacity, such as mobile phones and embedded systems. Additionally, maintaining real-time performance while ensuring high-quality image restoration is a critical challenge, as traditional deep learning models often fail to meet the stringent latency and efficiency requirements of such platforms. This paper introduces a new, lightweight convolutional neural network (CNN) architecture tailored for efficient pixelation detection and restoration. Our approach combines a pre-trained MobileNetV2 with finetuning for pixelation detection, and SpectraNet, a architecture incorporating dept.-wise separable convolutions for image restoration. The proposed architecture was trained on the HQ-50k dataset, with 70% of the data used for training and 30% for testing. It achieved a Peak Signal-to-Noise Ratio (PSNR) of 17-23, outperforming 10 state-of-the-art architectures in both efficiency and image quality. This design addresses the dual problems of high computational load and extensive memory usage, prioritizing real-time performance and resource efficiency. The proposed architecture is ideal for deployment on mobile and embedded devices since it maintains high accuracy while significantly reducing the number of trainable parameters. Our findings advance the feasibility of deploying advanced image restoration techniques in real-world, resource-constrained environments.

关键词： dept.-wise Separable Convolutions Embedded Devices Image Restoration Lightweight Real-Time Performance Resource-Constrained Platforms Deep Learning

来源：评论

学校读者我要写书评

暂无评论

Enhancing 3D Modeling Quality Through Orthorectified Image Accumulation

Enhancing 3D Modeling Quality Through Orthorectified Image A...

引用

2025 International Conference on Electronics, Information, and Communication, ICEIC 2025

作者： Kwag, Cheolsu Kim, Euntaek Lee, Yongsuk Jung, Seungmin Hwang, Sung Soo Ryu, Hwangryol Moon, Youngmin Handong Global University Dept. of Computer Science and Electrical Engineering Pohang Korea Republic of School of Computer Science and Electrical Engineering Handong Global University Pohang Korea Republic of Research Institute of Industrial Science and Technology Smart Engineering Research Laboratory Pohang Korea Republic of

ISBN: (纸本)9798331510756

This study introduces a novel methodology designed to facilitate the capture of comprehensive image datasets, crucial for accurate 3D modeling of expansive indoor spaces. Leveraging orthophotos generated from panoramic camera data and visualizing image coverage density through heatmaps, our approach effectively identifies under-captured and unvisited areas. This enables targeted image acquisition, minimizes redundant reshoots, and enhances the overall quality and fidelity of 3D models. © 2025 IEEE.

关键词： Digital elevation model

来源：评论

学校读者我要写书评

暂无评论

InterACT: Enhancing Human-AI Interaction 2

InterACT: Enhancing Human-AI Interaction

引用

2nd International Conference on Machine Learning and Autonomous Systems, ICMLAS 2025

作者： Katta, Jithendra Kodumuru, Nikhil Reddy Katta, Swathi School of Engineering and Applied Sciences The George Washington University WashingtonDC United States Vignana Bharathi Institute of Technology Computer Science & Engineering dept. Hyderabad India

ISBN: (纸本)9798331505745

The rapid advancement of large language models and computer vision systems has opened new frontiers in artificial intelligence. This paper introduces InterACT, a novel cross-modal system that integrates leading language and vision models to enable more intuitive and context-aware human AI interactions. By leveraging the strengths of both modalities, InterACT processes natural language queries to guide object detection in visual environments, bridging the gap between linguistic and visual understanding. We evaluate the performance of three language models of varying sizes in conjunction with an advanced object detection model, exploring the balance between computational efficiency and interaction quality. Our experiments demonstrate InterACT's potential across diverse real-world scenarios, revealing promising results in maintaining contextual relevance and adaptability. This research contributes to the growing field of cross-modal AI systems and lays the groundwork for more sophisticated human AI interactions, with implications for applications ranging from assistive technologies to educational tools. © 2025 IEEE.

关键词： Object detection

来源：评论

学校读者我要写书评

暂无评论

Live Audio Recognition and Text Conversion System 2

Live Audio Recognition and Text Conversion System

引用

2nd International Conference on Computational Intelligence, Communication technology and Networking, CICTN 2025

作者： Sharma, Pratyaksha Gupta, Sumita Bhavya, Boddu Asmitha Sharma, Khushi Dhir, Saru Gambhir, Sapna Department of Computer Science & Engineering Amity School of Engineering & Technology Amity University UP India Department of Information Technology Amity School of Engineering & Technology Amity University UP India Dept. of Computer Science College of Engineering and Computing George Mason University United States

ISBN: (纸本)9798331530389

With the advancement of technologies, different methods are currently being used for converting spoken language into text. These systems offer a hands-free alternative to traditional input methods, especially for individuals with physical disabilities or those requiring efficient documentation of spoken content. The goal of this study is to review recent advancements in this field and highlight gaps in current research, with a focus on accuracy, language adaptability, ease of use, and the integration of various technologies. We also suggest future directions for enhancing and developing this technology further. © 2025 IEEE.

关键词： Hidden Markov models

来源：评论

学校读者我要写书评

暂无评论

Sustainable Agriculture in Food Security Integrating Satellite Data Risk Assessment by Cyberattack Detection: AI Applications

引用

Remote Sensing in Earth Systems sciences 2025年第2期8卷 435-443页

作者： Kumar, M. Prema Krishnammal, N. Gupta, Manu Begum, Mairaj Unisa Sultana, Sameera Degala, Divya Priya Dept. of ECE Shri Vishnu Engineering College for Women (A) Andhra Pradesh Bhimavaram 534202 India Department of Computer Science and Engineering Dr. N.G.P. Institute of Technology Tamilnadu Coimbatore India Department of Electronics and Communication Engineering Sreenidhi Institute of Science and Technology Telangana Hyderabad India Department of CSE-AIML MALLAREDDY UNIVERSITY Telangana Hyderabad India Department of Computer Science and Engineering MLR Institute of Technology Hyderabad India

Globally, chronic food insecurity persists and is made worse by shocks brought on by climate change, like floods and droughts. In order to guarantee prompt assistance delivery, humanitarian programming prioritizes predicting levels of food insecurity and identifying homes that are at risk. Using complicated and diverse data, machine learning (ML) models have been widely used to promote food security. This research proposes a novel technique in sustainable agriculture-based food security analysis and satellite data cyberattack detection using machine learning models. Here, the input is collected based on satellite data, and its cyber detection has been analyzed using a convolutional spatio adversarial encoder neural network. Then, from this satellite data, the food security analysis data has been collected and processed for noise removal and normalization. Processed data has been classified using the fuzzy Q-Boltzmann ResNet transfer neural network. Experimental analysis is carried out for the food security analysis dataset in terms of data integrity, detection accuracy, average precision, F-1 score, and recall. The proposed approach yielded 97% detection accuracy, data integrity of 98%, 96% average precision, 93% recall, and 94% F-1 score. © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2025.

关键词： Cyberattack detection Food security analysis Machine learning models Satellite data Sustainable agriculture

来源：评论

学校读者我要写书评

暂无评论

Edge Node Load Balancing Using Game Theoretic Approaches 5

Edge Node Load Balancing Using Game Theoretic Approaches

引用

5th International Conference on Emerging Systems and Intelligent Computing, ESIC 2025

作者： Haldar, Rajdeep Rooj, Saptadip Sharma, Neeraj Das, Ritambhar Mandal, Arnab Bandyopadhyay, Anjan Dept. of Information Technology Bhubaneswar India Lovely Professional University School of Computer Science and Engineering Jalandhar India School of Computer Engineering Bhubaneswar India

ISBN: (纸本)9798331522100

This paper explores the application of game theoretic approaches to load balancing across edge nodes within distributed computing systems. Focusing on a non-cooperative game model, we aim to enhance system efficiency by strategically optimizing resource allocation and load distribution among autonomous edge nodes. Initial simulations demonstrate that our approach not only addresses the inherent complexities of edge environments but also outperforms traditional load balancing mechanisms in terms of both operational efficiency and system responsiveness. © 2025 IEEE.

关键词： Edge computing

来源：评论

学校读者我要写书评

暂无评论

Improvising Early Detection of Diabetes through Deep Learning on Retinal Fundus images 3

Improvising Early Detection of Diabetes through Deep Learnin...

引用

3rd International Conference on Electronics and Renewable Systems, ICEARS 2025

作者： Balaji, Krishna Ganesh, Maniprabha Shankar Murali, Daneesh Pachiappan, Kalaivani Iyyanar, P. Jayakumar, Nithya Kongu Engineering College Dept. of Computer Science Erode India Sona College of Technology Dept. of Information Technology Salem India K.S Rangasamy College of Technology Dept. of Information Technology Tiruchengode India

ISBN: (纸本)9798331509675

Diabetes is a chronic disease whose timely and accurate diagnosis will prevent serious complications from health. This paper explores using iridology principles in a deep learning method to detect diabetes from retinal images in order to streamline the process and automate the procedure. The objective here is to create a non-invasive procedure for early and reliable detection of the disease. Deep learning in medical imaging is the new way to interfere with conventional modes of diagnostics. The retinal imaging is non-invasive and has, over the years, been put to use in medical examinations, and its potential is now absolutely being utilized in the early diagnosis of diabetes. A retinal analysis that is based on principles, integrating them into iridology, concentrating on the patterns and signs portrayed in the iris for possible reflections of a general health condition, is aimed at achieving earlier indications of diabetes in a more effortless and rapid way. Models such as ResNet50, ResNet101, and EfficientNetB7 were precious but actually, it was InceptionV3 that delivered a great accuracy of 97.14%. This result proves the feasibility application of InceptionV3 for diagnosing diabetes in the early stages, thus eventually providing an easy and non-invasive way of moving forward to better health. © 2025 IEEE.

关键词： Ophthalmology

来源：评论

学校读者我要写书评

暂无评论

Diversity Seeking Techniques for Red-Teaming Large Language Models

Diversity Seeking Techniques for Red-Teaming Large Language ...

引用

2025 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2025

作者： Lee, Seokhan Ku, Bonhwa Ko, Hanseok Dept. Computer Science and Engineering Korea University Seoul Korea Republic of Machine Learning Big Data Institute Korea University Seoul Korea Republic of Dept. Electrical and Computer Engineering Korea University Seoul Korea Republic of

ISBN: (纸本)9798350368741

In this paper, we present new techniques for increasing the diversity of red-teaming prompts generated by automated machine learning-based methods, thereby enabling the discovery of more vulnerabilities in large language models. Using reinforcement learning to train models to output effective prompts for this task results in the models converging deterministically to a single output. Our first technique, which we term Defender, acts by blocking the reward signal for prompts that have already been discovered, thus making what was a stationary problem into a non-stationary problem that compels the reward maximizing algorithm to continually seek new prompts. Our second technique, Teamplay, trains two prompt generation models in tandem and adds the KL divergence between them to the reward in order to make them search in disparate regions of the space of prompts. Our techniques are shown experimentally to increase the effectiveness and diversity of prompts generated by existing reinforcement learning baselines. © 2025 IEEE.

关键词： AI safety large language models red-teaming reinforcement learning toxicity

来源：评论

学校读者我要写书评

暂无评论

Machine Learning Approaches for Predicting Lung Cancer Risk and Early Diagnosis 2

Machine Learning Approaches for Predicting Lung Cancer Risk ...

引用

2nd International Conference on Machine Learning and Autonomous Systems, ICMLAS 2025

作者： Jegan, K. Nikesh, Infant Pravin, A. Sathyabama Institute of Science and Technology Dept of Computer Science and Engineering Tamil Nadu Chennai600119 India

ISBN: (纸本)9798331505745

Using a healthcare dataset, this study applies a variety of ML techniques to analyze and forecast the likelihood of lung cancer. This dataset will be initialized by cleaning, encoding categorical variables, and removing duplicates to ensure quality and efficiency. Correlation analysis is carried out to find important characteristics that raise the risk of lung cancer. engineering including interaction terms enhances the predictive efficiency of the dataset. The various machine learning models are introduced they are Logistic Regression, SVC, Random Forests, GBM, and XGB. Over-sampling with ADASYN addresses class imbalance to improve model training. The logistic regression machine learning model has attained 97% accuracy. It will ensure consistency and reliability. Comprehensive visualizations such as heatmaps and bar plots, explore feature correlations and class distributions. The study represents the efficiency of ensemble methods like Random Forests and XGBoost, achieving superior predictive performance compared to baseline models. This system predicts whether they might have lung cancer, helping doctors to diagnose or early intervention. © 2025 IEEE.

关键词： Visualization

来源：评论

学校读者我要写书评

暂无评论

Geometrically Shaped Constellation in Short-Packet Visible Light Communications for IIoT Applications

引用

IEEE Internet of Things Journal 2025年第11期12卷 18220-18234页

作者： Guo, Jia-Ning Zhang, Jian Gong, Chen Li, Longguang Yang, Xu Zhou, Jing Chen, Ru-Han information engineering university Henan Province 450000 China The CAS Key Laboratory of Wireless-Optical Communications School of Information Science and Technology University of Science and Technology of China Hefei230027 China The SJTU Paris Elite Institute of Technology Shanghai Jiao Tong University Shanghai200240 China Sixty-Third Research Institute National University of Defense Technology Nanjing China Dept. Computer Science and Engineering Shaoxing University Shaoxing China

Due to the high data rates, license-free operations and inherent security, visible light communication (VLC) is highly valued in industrial Internet of Things (IIoT) applications. Short packet communication, which offers ultra-low latency and efficient resource utilization, is suitable for the low-latency and massive connectivity demand of the IIoT system. It adopts finite blocklength codewords for data transmissions. In this paper, we present a general framework of designing geometrically shaped constellations in short-packet VLC with peak and average intensity constraints for IIoT applications. By leveraging tools from large deviation theory, we first characterize the second-order asymptotics of the optimal constellation shaping region under aforementioned intensity constraints, which serves as a good performance metric for the best geometric shaping in finite blocklength. To further incorporate a sufficiently large coding gain and a nearly maximum shaping gain, we construct multidimensional constellations by the nested structure of Construction B lattices, where the constellation shaping is implemented by controlling the boundary of the embedded sublattice, i.e., a strategy called coarsely shaping and finely coding. Fast algorithms for constellation mapping and demodulation are presented as well. As an illustrative example, we present an energy-efficient 24-dimensional constellation design based on the Leech lattice, whose superiority over existing constellation designs is verified by numerical results. © 2014 IEEE.

关键词： Network security

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：