检索结果-内蒙古大学图书馆

TiTDet: A tiny text detector with scale-sensitive loss and effective fusion factor

Journal of Intelligent and Fuzzy Systems 2024年第1期1卷 11367页

作者： Wei, Guangcun Fu, Jihua Pan, Zhifei Fang, Qingge Zhang, Zhi College of Intelligent Equipment Shandong University of Science and Technology Shandong Taian China College of Computer Sicence and Engineering Shandong University of Science and Technology Shandong Qingdao China Taian Synergy Software Ltd. Shandong Taian China

The text in natural scenes is often smaller compared to artificially designed text. Due to the small proportion of pixels, low resolution, less semantic information, and susceptibility to complex scenes, tiny text detection often results in many missed detections. To address this issue, this paper draws inspiration from small object detection methods and proposes TiTDet, a detection algorithm more suitable for tiny text. Due to the small proportion of pixels, low resolution, less semantic information, and susceptibility to complex scenes, tiny text detection often results in many missed detections. To address this issue, this paper draws inspiration from small object detection methods and proposes TiTDet, a detection algorithm more suitable for tiny text. Firstly, this paper incorporates a context extraction module and an attention-guided module. These modules guide contextual information learning through a self attention mechanism, while eliminating the possible negative impact caused by redundant information. Regarding multi-scale feature fusion, this paper proposes a fine-grained effective fusion factor, making the fusion process emphasize small object learning more and highlight the feature expression of tiny texts. In terms of post-processing, this paper proposes a differentiable binarization module, incorporating the binarization process into model training. Leveraging the implicit information in the data to drive model improvement can enhance the post-processing effect. Lastly, this paper proposes a scale-sensitive loss, which can handle tiny texts more fairly, fully considering the positional relationship between the predicted and real regions, and better guiding the model training. This paper proves that TiTDet exhibits high sensitivity and accuracy in detecting tiny texts, achieving an 86.0% F1-score on ICDAR2015. The paper also compares the superiority of the method on CTW1500 and Total-Text. © 2024 - IOS Press. All rights reserved.

关键词： Pixels

来源：评论

学校读者我要写书评

暂无评论

Binary Decomposition for Multi-Class Classification Problems: Development and Applications

Binary Decomposition for Multi-Class Classification Problems...

引用

2023 International Conference on Machine Learning and Cybernetics, ICMLC 2023

作者： Li, Peiwei Liu, Han College of Computer Science and Software Engineering Shenzhen University Shenzhen518060 China

ISBN: (纸本)9798350303780

Binary decomposition of a multi-class classification problem is a widely used method in the field of machine learning, which involves using an ensemble of binary classifiers to undertake multi-class classification tasks. The motivation of binary decomposition lies in two main aspects. Firstly, some learning algorithms such as support vector machine can not achieve directly learning a multi-class classifier. Secondly, in some cases, the adoption of binary decomposition for training an ensemble of binary classifiers can even achieve better performance than the direct learning of a multi-class classifier. There are some existing methods that can be applied in achieving binary decomposition and have been proven to be effective. Some of these methods may also be applied in other tasks beyond classification. This paper presents a comprehensive review of the development and applications of binary decomposition methods. Specifically, these methods are put into two categories, namely, ordered and unordered decomposition, which each is introduced in terms of the essence and strategies of decomposition. These methods are also compared through theoretical analysis of their effectiveness and efficiency. On the basis of the comparative analysis, we suggest some future directions towards achieving further advances in binary decomposition. © 2023 IEEE.

关键词： Classification (of information)

来源：评论

学校读者我要写书评

暂无评论

Forecasting Price Prediction for Vegetables and Fruits Using Recurrent Neural Network 5

Forecasting Price Prediction for Vegetables and Fruits Using...

引用

5th International Conference on Electronics and Sustainable Communication Systems, ICESC 2024

作者： Gothai, E. Rajalaxmi, R.R. Thamilselvan, R. Harshath, S.M. Kongu Engineering College Department of Computer Technology Tamilnadu Perundurai India Kongu Engineering College Department of Computer Science and Engineering Tamilnadu Perundurai India Kongu Engineering College MSc Software Systems Erode India

ISBN: (纸本)9798350379945

Unpredictable fruit and vegetable prices create significant challenges for farmer livelihoods. This research proposes an innovative approach using recurrent neural networks (RNNs) to predict both minimum and maximum prices. By analyzing historical market data, these models can forecast potential price ranges, empowering farmers to make informed decisions about planting, harvesting, and sales strategies. This enhanced foresight helps farmers navigate market volatility and improve income stability. © 2024 IEEE.

关键词： Vegetables

来源：评论

学校读者我要写书评

暂无评论

Unsupervised Video Object Segmentation via Weak User Interaction and Temporal Modulation

引用

Chinese Journal of Electronics 2023年第3期32卷 507-518页

作者： FAN Jiaqing ZHANG Kaihua ZHAO Yaqian LIU Qingshan College of Computer Science and Technology Nanjing University of Aeronautics and Astronautics College of Computer and Software Nanjing University of Information Science and Technology Engineering Research Center of Digital Forensics Ministry of Education Inspur Suzhou Intelligent Technology Corporation

In unsupervised video object segmentation(UVOS), the whole video might segment the wrong target due to the lack of initial prior information. Also, in semi-supervised video object segmentation(SVOS), the initial video frame with a fine-grained pixel-level mask is essential to good segmentation accuracy. It is expensive and laborious to provide the accurate pixel-level masks for each training sequence. To address this issue, We present a weak user interactive UVOS approach guided by a simple human-made rectangle annotation in the initial frame. We first interactively draw the region of interest by a rectangle, and then we leverage the mask RCNN(region-based convolutional neural networks) method to generate a set of coarse reference labels for subsequent mask propagations. To establish the temporal correspondence between the coherent frames, we further design two novel temporal modulation modules to enhance the target representations. We compute the earth mover's distance(EMD)-based similarity between coherent frames to mine the co-occurrent objects in the two images, which is used to modulate the target representation to highlight the foreground target. We design a cross-squeeze temporal modulation module to emphasize the co-occurrent features across frames, which further helps to enhance the foreground target representation. We augment the temporally modulated representations with the original representation and obtain the compositive spatio-temporal information, producing a more accurate video object segmentation(VOS) model. The experimental results on both UVOS and SVOS datasets including Davis2016,FBMS, Youtube-VOS, and Davis2017, show that our method yields favorable accuracy and complexity. The related code is available.

关键词： Earth Training Codes Annotations Modulation Object segmentation Complexity theory

来源：评论

学校读者我要写书评

暂无评论

Enhanced Loan Approval Prediction System using Ensemble Machine Learning Techniques 3

Enhanced Loan Approval Prediction System using Ensemble Mach...

引用

3rd International Conference on Automation, Computing and Renewable Systems, ICACRS 2024

作者： Gothai, E. Rajalaxmi, R.R. Thamilselvan, R. Sridhar, K. Department of Computer Technology Kongu Engineering College Tamilnadu Perundurai India Department of Computer Science and Engineering Kongu Engineering College Tamilnadu Perundurai India MSc Software Systems Kongu Engineering College Erode India

ISBN: (纸本)9798331532420

Banks play a pivotal role in generating significant profits through loan operations. However, the challenge lies in accurately identifying genuine loan applicants who are likely to repay their loans. Manual processes are prone to misconceptions, prompting the development of a loan prediction system using deep learning. This system aims to automate the selection of eligible candidates, benefiting both bank staff and applicants by drastically reducing the loan approval timeframe. Two key machine learning algorithms such as Artificial Neural Network (ANN) and Support Vector Machine (SVM) are employed in this work for applicant selection. SVM, a popular supervised learning algorithm, creates a decision boundary or hyperplane to segregate classes in n-dimensional space. Conversely, ANN is inspired by biological neural networks, which mimic the organization of neurons in the input layer, hidden and output layers of the human brain. The integration of ANN and SVM in an ensemble model enhances accuracy compared to traditional methods. This collaborative approach ensures precise predictions of loan repayment likelihood based on applicant data and past loan eligibility records. The model's performance is measured using accuracy. The outcomes of the research were promising and the loan prediction system defines the minimum and maximum loan amounts that can be sanctioned to an applicant. By leveraging the strengths of both ANN and SVM, the ensemble model serves as a robust tool for accurately identifying worthy loan applicants, streamlining the loan approval process, and minimizing the risk of default. The accuracy of the algorithm is improved by this method to 91%. © 2024 IEEE.

关键词： Support vector machines

来源：评论

学校读者我要写书评

暂无评论

Based on feature reconstruction and feature intersection, the Deep CTR Prediction Model 2

Based on feature reconstruction and feature intersection, th...

引用

2nd International Conference on Big Data, Information and computer Network, BDICN 2023

作者： Bian, Ning Wang, Li University of Science and Technology Liaoning College of Computer and Software Engineering Anshan China

ISBN: (纸本)9798350321364

Collecting a lot of feature data can enhance the prediction impact for the recommender system's CTR prediction task. The second-order feature cross factorization machine model, which is the most popular, has the ability to automatically learn implicit features based on existing features. The factorization machine model, however, has the following issues: (1) The encoding is fixed, and the same feature represents the same in various contexts. (2) The encoding type is single, and the feature is represented by a single weight vector, which cannot completely express the information of the feature. To harvest possible feature information, recommender systems employ neural networks. A novel method of feature intersection is created by the Compressed Interaction Network (CIN) component suggested in xDeepFM. In order to address the issue of encoding fixation in the factorization machine, the original feature encoding is reconstructed in this paper using xDeepFM in conjunction with the feature refinement network (FRNet). Additionally, different feature representations are expressed in various contexts for the same feature. To address the issue of a single encoding type in factorization machines, the field-aware interactive machine (FIM) component provides feature information at the domain and feature levels, respectively. The model is known as the Field-aware Feature Refinement Depth Factorization Machine (FRDFFM). On the Criteo dataset and the Avazu dataset, respectively, the suggested model was evaluated, and it produced promising results. © 2023 IEEE.

关键词： Forecasting

来源：评论

学校读者我要写书评

暂无评论

SRGAN Optimization through Enhanced Network Structure and Multi-Loss Functions 4

SRGAN Optimization through Enhanced Network Structure and Mu...

引用

4th IEEE International Conference on Data science and computer Application, ICDSCA 2024

作者： Chen, Zexin Chen, Xiaoxin College of Computer Science and Software Engineering Shenzhen University Shenzhen China College of Medical Information Engineering Guangdong Pharmaceutical University Guangzhou China

ISBN: (纸本)9798350368239

An enhanced SRGAN image super-resolution algorithm is proposed to address SRGAN's training instability and limited global information capture. This enhancement optimizes both the network structure and loss function. In network structure optimization, group normalization is introduced as an alternative to batch normalization, addressing its dependency on batch size and improving stability during small-batch training. Additionally, Global Context Block (GCBlock) is integrated into the residual block, enhancing the network’s global information capture and detail representation capabilities. In regard to the loss function, this paper introduces L1 loss and MS-SSIM loss as alternatives to the conventional MSE loss. This approach aims to enhance the quality of the generated image in both pixel space and structural information while reducing the smoothing effect. In the experimental section, the ablation experiments evaluate the performance of models with and without GCBlock on five test sets, including Urban100 and Manga109. The results demonstrate that GCBlock improves both detail expressiveness and global information capture, with an average PSNR gain of 0.1388 and SSIM improvement of 0.008. In comparison experiments, the improved algorithm achieved an average PSNR increase of 0.270 and an average SSIM increase of 0.018 on the aforementioned test set, indicating a significant improvement in detail expression and visual quality. Finally, the limitations of the enhanced algorithm are discussed, and potential future research directions are proposed. ©2024 IEEE.

关键词： Image reconstruction

来源：评论

学校读者我要写书评

暂无评论

Federated Reinforcement Learning for Sharing Experiences Between Multiple Workers

Federated Reinforcement Learning for Sharing Experiences Bet...

引用

2023 International Conference on Machine Learning and Cybernetics, ICMLC 2023

作者： Feng, Wenliang Liu, Han Peng, Xiaogang College of Computer Science and Software Engineering Shenzhen University Shenzhen518060 China

ISBN: (纸本)9798350303780

Reinforcement learning has been successfully applied in various fields, such as games and robots. However, there are still some issues in the traditional reinforcement learning paradigm that involves one agent per environment, such low efficiency in data generation and insufficient amount of training data. In this paper, we propose a federated reinforcement learning framework that utilizes multiple agents to generate data for simultaneous training of models. Specifically, the proposed framework allows the best performing agent to share its learning experience with other agents for improving the learning performance and protecting the privacy of agents. The proposed framework is applied to the Cart-pole, Acrobot and LunarLander environments in OpenAI Gym, and the results show that a significant improvement of the learning performance can be achieved by adopting our framework. © 2023 IEEE.

关键词： Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

Support vector machine with discriminative low-rank embedding

引用

CAAI Transactions on Intelligence Technology 2024年第5期9卷 1249-1262页

作者： Guangfei Liang Zhihui Lai Heng Kong Computer Vision Institute College of Computer Science and Software EngineeringShenzhen UniversityShenzhenChina Shenzhen Institute of Artificial Intelligence and Robotics for Society ShenzhenChina Department of Breast and Thyroid Surgery BaoAn Central Hospital of ShenzhenShenzhenChina

Support vector machine(SVM)is a binary classifier widely used in machine ***,neglecting the latent data structure in previous SVM can limit the performance of SVM and its *** address this issue,the authors propose a novel SVM with discriminative low-rank embedding(LRSVM)that finds a discriminative latent low-rank subspace more suitable for SVM *** extension models of LRSVM are introduced by imposing different orthogonality constraints to prevent computational inaccuracies.A detailed derivation of the authors’iterative algorithms are given that is essentially for solving the SVM on the low-rank ***,some theorems and properties of the proposed models are presented by the *** is worth mentioning that the subproblems of the proposed algorithms are equivalent to the standard or the weighted linear discriminant analysis(LDA)*** indicates that the projection subspaces obtained by the authors’algorithms are more suitable for SVM classification compared to those from the LDA *** convergence analysis for the authors proposed algorithms are also ***,the authors conduct experiments on various machine learning data sets to evaluate the *** experiment results show that the authors’algorithms perform significantly better than other algorithms,which indicates their superior abilities on classification tasks.

关键词： iterative methods machine leaning support vector machunes

来源：评论

学校读者我要写书评

暂无评论

Recent Advancements and Challenges in Multimodal Sentiment Analysis: A Survey

Recent Advancements and Challenges in Multimodal Sentiment A...

引用

2023 International Conference on Machine Learning and Cybernetics, ICMLC 2023

作者： Dong, Zhen-Xun Liu, Han College of Computer Science and Software Engineering Shenzhen University Shenzhen518060 China

ISBN: (纸本)9798350303780

Sentiment analysis aims to extract people's underlying attitudes, opinions and thoughts toward various topics, products and services. Sentiment analysis is a popularly adopted technique that involves applications across domains such as business, healthcare, government and marketing. Traditional sentiment analysis methods mainly focus on text contents. However, recent technological advancements have enabled the expression of opinions through different forms of media, such as images or audios, leading to sentiment analysis using multiple modalities rather than text alone. Multimodal sentiment analysis results in a more comprehensive understanding of the emotions and opinions expressed in a given content by combining information from different modalities. A model of multimodal sentiment analysis involves three main parts, namely, unimodal feature extraction, multimodal fusion and sentiment polarity prediction. This paper provides a comprehensive overview of multimodal fusion methods along with their corresponding model architectures. Moreover, several more recently released models are introduced and the strengths and limitations of these models are discussed comparatively. Finally, some challenges and future research directions are identified. © 2023 IEEE.

关键词： Sentiment analysis

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：