检索结果-内蒙古大学图书馆

Two-stage object detection in low-light environments using deep learning image enhancement

PeerJ computer Science 2025年 11卷 e2799-e2799页

作者： Al-refai, Ghaith Elmoaqet, Hisham Al-Refai, Abdullah Alzu’bi, Ahmad Al-Hadhrami, Tawfik Alkhateeb, Abedalrhman Department of Mechatronics Engineering German Jordanian University Amman Jordan Software Engineering Department King Hussein School of Computing Science Princess Sumaya University College for Technology Amman Jordan Department of Computer Science Faculty of Computer and Information Technology Jordan University of Science and Technology Irbid Jordan Department of Computer Science School of Science and Technology Nottingham Trent University Nottingham United Kingdom Department of Computer Science Lakehead University Thunder Bay Canada

This study presents a two-stage object detection system specifically tailored for low-light conditions. In the initial stage, supervised deep learning image enhancement techniques are utilized to improve image quality and enhance features. The second stage employs a computer vision algorithm for object detection. Three image enhancement algorithms—ZeroDCE++, Gladnet, and two-branch exposurefusion network for low-light image enhancement (TBEFN)—were assessed in the first stage to enhance image quality. YOLOv7 was utilized in the object detection phase. The ExDark dataset, recognized for its extensive collection of low-light images, served as the basis for training and evaluation. No-reference image quality evaluators were applied to measure improvements in image quality, while object detection performance was assessed using metrics such as recall and mean average precision (mAP). The results indicated that the two-stage system incorporating TBEFN significantly improved detection performance, achieving a mAP of 0.574, compared to 0.49 for YOLOv7 without the enhancement stage. Furthermore, this study investigated the relationship between object detection performance and image quality evaluation metrics, revealing that the image quality evaluator NIQE exhibited a strong correlation with mAP for object detection. This correlation aids in identifying the features that influence computer vision performance, thereby facilitating its enhancement. © 2025 Al-refai et al.

关键词： Image quality

来源：评论

学校读者我要写书评

暂无评论

Fed RDP: A Robust Federated Learning Framework for Multi Party Decision-Making

引用

IEEE Transactions on Network and Service Management 2025年

作者： An, Jian Tang, Siyu Ping, Ruyuan Li, Ran He, Xin Jin, Xiaolong Xi’an Jiaotong University School of Computer Science and Technology China Shaanxi Province Key Laboratory of Computer Network Xi’an Jiaotong University Xi’an710049 China Xi’an Jiaotong University School of Computer Science and Technology Xi’an710049 China Henan University School of Software Kaifeng475001 China Rocket Force University of Engineering School of Combat Supporting Xi’an710000 China

In the field of intelligent manufacturing, the vast and heterogeneous nature of data across different departments poses significant challenges for collaborative decision-making. Direct data sharing often leads to severe privacy and security concerns. Although federated learning offers a promising solution, issues such as participant selection and privacy protection remain inadequately addressed. During model training, it is crucial to minimize the influence of low-quality participants and prevent the inference of sensitive information through gradient analysis, which could threaten model performance or data privacy. To address these challenges, this paper proposes a robust federated learning model, Fed-RDP, based on participant contribution evaluation and personalized differential privacy. The model leverages blockchain technology for secure data flow and storage, implemented through smart contracts. Historical parameters are submitted to evaluate contributions based on the concept of the least core, with real-time updates to reputation scores. When participants upload their local models, the scores are used as weights to aggregate and update the global model. Additionally, personalized differential noise is added to the uploaded gradients based on participant scores, preserving privacy while maximizing the utility of the data. Experimental results demonstrate that this approach effectively identifies low-quality participants, optimizes evaluation time, and protects privacy through personalized differential noise. © 2004-2012 IEEE.

关键词： Federated learning

来源：评论

学校读者我要写书评

暂无评论

Approximated Coded Computing: Towards Fast, Private and Secure Distributed Machine Learning

引用

IEEE Transactions on Emerging Topics in Computing 2025年

作者： Qiu, Houming Zhu, Kun Luong, Nguyen Cong Niyato, Dusit Jiangxi University of Finance and Economics School of Software and IoT Engineering Jiangxi 330032 China Nanjing University of Aeronautics and Astronautics College of Computer Science and Technology Nanjing210016 China Phenikaa University Faculty of Computer Science Hanoi12116 Viet Nam Nanyang Technological University School of Computer Science and Engineering 639798 Singapore

In a large-scale distributed machine learning system, coded computing has attracted wide-spread attention since it can effectively alleviate the impact of stragglers. However, several emerging problems greatly limit the performance of coded distributed systems. Firstly, an existence of colluding workers who collude results with each other leads to serious privacy leakage issues. Secondly, there are few existing works considering security issues in data transmission of distributed computing systems/or coded distributed machine learning systems. Thirdly, the number of required results for which need to wait increases with the degree of decoding functions. In this paper, we design a secure and private approximated coded distributed computing (SPACDC) scheme that deals with the above-mentioned problems simultaneously. Our SPACDC scheme guarantees data security during the transmission process using a new encryption algorithm based on elliptic curve cryptography. Especially, the SPACDC scheme does not impose strict constraints on the minimum number of results required to be waited for. An extensive performance analysis is conducted to demonstrate the effectiveness of our SPACDC scheme. Furthermore, we present a secure and private distributed learning algorithm based on the SPACDC scheme, which can provide information-theoretic privacy protection for training data. Our experiments show that the SPACDC-based deep learning algorithm achieves a significant speedup over the baseline approaches. © 2013 IEEE.

关键词： Parallel processing systems

来源：评论

学校读者我要写书评

暂无评论

Multi-modality medical image classification with ResoMergeNet for cataract, lung cancer, and breast cancer diagnosis

引用

computers in Biology and Medicine 2025年 187卷 109791-109791页

作者： Ejiyi, Chukwuebuka Joseph Cai, Dongsheng Fiasam, Delali Linda Adjei-Arthur, Bonsu Obiora, Sandra Ayekai, Browne Judith Asare, Sarpong K. Jonathan, Anto Leoba Qin, Zhen College of Nuclear Technology and Automation Engineering Chengdu University of Technology Sichuan China School of Information and Software Engineering University of Electronic Science and Technology of China Sichuan China School of Computer Science and Engineering University of Electronic Science and Technology of China Sichuan China Leeds Business School Leeds Beckett University LeedsLS1 3HB United Kingdom School of Electronic Science and Engineering University of Electronic Science and Technology of China Chengdu China

The variability in image modalities presents significant challenges in medical image classification, as traditional deep learning models often struggle to adapt to different image types, leading to suboptimal performance across diverse datasets. This is critical in the diagnosis of conditions like cataracts and cancers, where imaging data spans various modalities, including visible eye images for cataracts and histopathological images for cancers among others. Cataracts, a leading cause of blindness, and lung and breast cancers, major contributors to cancer-related deaths, require early detection for effective intervention. However, many existing models fall short in handling modality differences, limiting their performance. To address this, we propose ResoMergeNet (RMN), designed to handle multi-modal medical image classification. RMN integrates transfer learning and advanced techniques – the ResBoost framework and ConvMergeNet, enabling the model to effectively extract relevant features from both visible eye images and histopathological images. The model's architecture emphasizes both global and local feature extraction while minimizing the influence of irrelevant data, thus improving classification performance across modalities. Evaluated on the cataract dataset (binary classification), RMN achieved an accuracy of 99.17 %. For lung cancer (3-class classification), it attained 100 % accuracy, while on the BreakHis (8-class classification) dataset, RMN reached 99.24 % accuracy at 100× magnification and 98.28 % at 200× magnification. These results demonstrate RMN's robustness and adaptability to varying image modalities, highlighting its potential as a reliable diagnostic tool in medical settings. Through its versatility, RMN offers a promising solution for improving early diagnosis and healthcare outcomes in the fight against cataracts, lung, and breast cancers. © 2025 Elsevier Ltd

关键词： Lung cancer

来源：评论

学校读者我要写书评

暂无评论

Variational Autoencoders for Spatio-Temporal Dynamic Graph Data 24

Variational Autoencoders for Spatio-Temporal Dynamic Graph D...

引用

Proceedings of the 2024 4th International Conference on Big Data, Artificial Intelligence and Risk Management

作者： Yanhui Peng Sicheng Xi School of Software Engineering Chongqing University of Posts and Telecommunications Chongqing China College of Computer Science and Technology Chongqing University of Posts and Telecommunications Chongqing China

ISBN: (纸本)9798400709753

Most of the existing graph variational autoencoders primarily focus on static graphs, and fail to fully consider the temporal characteristics of dynamic networks, which are difficult to capture the characteristics of real-world network evolution over time, resulting in poor prediction accuracy of downstream tasks. The traditional encoder is improved by introducing the cloud model, the forward cloud generator model is used to extend the normal distribution to the Gaussian cloud distribution, and the idea of grey prediction is combined to serialize each snapshot to expand the downstream link prediction task. By fusing spatial and temporal features, experiments are carried out on three real dynamic network datasets. The results show that the proposed model is superior to the baseline model in link prediction.

关键词： cloud model

来源：评论

学校读者我要写书评

暂无评论

Trauma-Informed Insights from Co-Design of Self-Disclosure Robots with Domestic Abuse Survivors 25

Trauma-Informed Insights from Co-Design of Self-Disclosure R...

引用

ACM/IEEE International Conference on Human-Robot Interaction (HRI)

作者： Nyomi Carrine Morris Tom Williams Ben Jelen MIRRORLab Colorado School of Mines Golden CO USA Department of Computer Science and Software Engineering Rose-Hulman Institute of Technology Terre Haute IN USA

ISBN: (数字)9798350378931

ISBN: (纸本)9798350378948

Survivors of domestic abuse face significant challenges securing recovery resources, such as housing, mental health care, and social connections. Accessing these resources requires survivors to disclose their status as survivors of domestic abuse; a process that can be traumatic and emotionally burdensome. In this work, we consider how social robots might help domestic abuse survivors to address challenges surrounding self-disclosure. To do so, we conducted a three-session co-design workshop involving 8 participants from a domestic abuse shelter. Our results provide three key contributions: (1) new insights into the benefits and barriers to self-disclosure; (2) new insights into the ways that social robots can help survivors to better achieve those benefits and overcome those barriers before, during, and after disclosure, including key design recommendations associated with each of these phases; and (3) observations into how a trauma-informed computing perspective may enable more effective design work in Human-Robot Interaction.

关键词： Conferences Social robots Mental health Faces

来源：评论

学校读者我要写书评

暂无评论

SolSearch: An LLM-Driven Framework for Efficient SAT-Solving Code Generation

arXiv

引用

arXiv 2025年

作者： Sheng, Junjie Lin, Yanqiu Wu, Jiehao Huang, Yanhong Shi, Jianqi Zhang, Min Wang, Xiangfeng School of Computer Science and Technology East China Normal University Shanghai China Software Engineering Institute East China Normal University Shanghai China

The Satisfiability (SAT) problem is a core challenge with significant applications in software engineering, including automated testing, configuration management, and program verification. This paper presents SolSearch, a novel framework that harnesses large language models (LLMs) to discover and optimize SAT-solving strategies automatically. Leveraging a curriculum-based, trial-and-error process, SolSearch enables the LLM to iteratively modify and generate SAT solver code, thereby improving solving efficiency and performance. This automated SAT-solving paradigm has the advantage of being plug-and-play, allowing integration with any SAT solver and accelerating the development or design process of new SAT solvers (new methods). Our preliminary experimental results are encouraging by demonstrating that the LLM-powered paradigm improves state-of-the-art SAT solvers on general SAT benchmarks and significantly enhances the performance of the widely used Z3 solver (11% on PAR-2 score). These results highlight the potential for using LLM-driven methods to advance solver adaptability and effectiveness in real-world software engineering challenges. Future research directions are discussed to further refine and validate this approach, offering a promising avenue for integrating AI with traditional software engineering tasks. © 2025, CC BY.

关键词： Application programs

来源：评论

学校读者我要写书评

暂无评论

Task-Oriented Feature Compression for Multimodal Understanding via Device-Edge Co-Inference

arXiv

引用

arXiv 2025年

作者： Yuan, Cheng Liu, Zhening Lv, Jiashu Shao, Jiawei Jiang, Yufei Zhang, Jun Li, Xuelong of China Telecom China Department of Electronic and Computer Engineering Hong Kong University of Science and Technology Hong Kong School of Software and Microelectronics Peking University China School of Electronic and Information Engineering Harbin Institute of Technology Shenzhen China

With the rapid development of large multimodal models (LMMs), multimodal understanding applications are emerging. As most LMM inference requests originate from edge devices with limited computational capabilities, the predominant inference pipeline involves directly forwarding the input data to an edge server which handles all computations. However, this approach introduces high transmission latency due to limited uplink bandwidth of edge devices and significant computation latency caused by the prohibitive number of visual tokens, thus hindering delay-sensitive tasks and degrading user experience. To address this challenge, we propose a task-oriented feature compression (TOFC) method for multimodal understanding in a device-edge co-inference framework, where visual features are merged by clustering and encoded by a learnable and selective entropy model before feature projection. Specifically, we employ density peaks clustering based on K nearest neighbors to reduce the number of visual features, thereby minimizing both data transmission and computational complexity. Subsequently, a learnable entropy model with hyperprior is utilized to encode and decode merged features, further reducing transmission overhead. To enhance compression efficiency, multiple entropy models are adaptively selected based on the characteristics of the visual features, enabling a more accurate estimation of the probability distribution. Comprehensive experiments on seven visual question answering benchmarks validate the effectiveness of the proposed TOFC method. Results show that TOFC achieves up to 60% reduction in data transmission overhead and 50% reduction in system latency while maintaining identical task performance, compared with traditional image compression methods. Copyright © 2025, The Authors. All rights reserved.

关键词： Image compression

来源：评论

学校读者我要写书评

暂无评论

Safeguarding Privacy in Real-Time Trajectory Releases for AI-Driven Consumer Applications

引用

IEEE Transactions on Consumer Electronics 2025年

作者： Cai, Sujin Tian, Wei Lv, Taizhi Wang, Xiangchen Li, Xin Jiangsu Maritime Institute School of Information and Engineering Nanjing211106 China Guizhou University of Commerce College of Computer and Information engineering Guiyang550014 China Southern University of Science and Technology College of Computer Science Shenzhen518055 China Hohai University College of Computer Science and Software Engineering Nanjing211106 China

Intelligent consumer applications optimize product recommendations by analyzing users' real-time trajectory information and mobility patterns to gain insights into their consumption preferences. However, privacy concerns often cause individuals to hesitate when sharing their locations. This paper introduces a real-time trajectory release scheme called Dynamic Trajectory Releasing based on Geo-indistinguishability (DTRG) that ensures client-side data protection. A significant contribution of this work is the introduction of a novel concept, w-event Geo-indistinguishability, which extends the measurement of privacy for real-time trajectories. Building upon the theoretical foundation of w-event Geo-indistinguishability, we design a dynamic privacy budget allocation strategy that doesn't require pre-setting trajectory length. This strategy recycles unused privacy budgets and progressively reallocating them in a backward cumulative manner, enhancing the utilization efficiency of the privacy budget. By incorporating users' travel information for prediction, DTRG adaptively perturbs real locations based on prediction results. Through the collaborative operation of the Prediction Mechanism (PM), Detection Mechanism (DM) and Noisy Mechanism (NM), this approach further optimizes the utility of the budget. Experimental results demonstrate that DTRG significantly improves data availability, achieving up to a 36% increase compared to some competing algorithms while maintaining the desired level of privacy. © 1975-2011 IEEE.

关键词： Differential privacy

来源：评论

学校读者我要写书评

暂无评论

AVENet: Disentangling Features by Approximating Average Features for Voice Conversion

arXiv

引用

arXiv 2025年

作者： Wang, Wenyu Zhou, Yiquan Zhu, Jihua Ding, Hongwu Xu, Jiacheng Li, Shihao School of Software Engineering Xi’an Jiaotong University Xi’an China School of Computer Science and Technology Anhui University Hefei China School of Software Engineering East China Normal University Shanghai China Division of Music and Audio Union Wheatland Culture and Media Ltd Chengdu China

Voice conversion (VC) has made progress in feature disentanglement, but it is still difficult to balance timbre and content information. This paper evaluates the pre-trained model features commonly used in voice conversion, and proposes an innovative method for disentangling speech feature representations. Specifically, we first propose an ideal content feature, referred to as the average feature, which is calculated by averaging the features within frame-level aligned parallel speech (FAPS) data. For generating FAPS data, we utilize a technique that involves freezing the duration predictor in a Text-to-Speech system and manipulating speaker embedding. To fit the average feature on traditional VC datasets, we then design the AVENet to take features as input and generate closely matching average features. Experiments are conducted on the performance of AVENet-extracted features within a VC system. The experimental results demonstrate its superiority over multiple current speech feature disentangling methods. These findings affirm the effectiveness of our disentanglement approach. Copyright © 2025, The Authors. All rights reserved.

关键词： Feature extraction

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：