检索结果-内蒙古大学图书馆

Lite-HDNet: A Lightweight Domain-Adaptive Segmentation Framework for Improved Finger Vein Pattern Extraction

IEEE ACCESS 2024年 12卷 46165-46180页

作者： Li, Yingxin Chen, Yucong Zeng, Junying Qin, Chuanbo Zhang, Wenguang Jiangmen Cent Hosp Dept Network Informat Jiangmen 529020 Guangdong Peoples R China Wuyi Univ Dept Intelligent Mfg Jiangmen 529020 Guangdong Peoples R China Jiangmen Cent Hosp Dept Neurosurg Jiangmen 529020 Guangdong Peoples R China

Recent times have witnessed significant progress in deep learning-based finger vein pattern extraction methods, but two unavoidable issues still remain to be addressed. One is that the model trained on a single finger vein dataset shows poor generalizability, and the model performance is limited by the image quality of the single dataset;the other is that it is hard for the deep model to extract real-time finger vein patterns because of its large number of parameters and poor real-time performance. To address the aforementioned issues, we propose a novel lightweight domain-adaptive segmentation framework (Lite-HDNet) that learns a generic representation of different domains to improve the extraction of finger vein patterns. We propose a multi-domain feature knowledge transfer strategy and a domain migration loss converter to enable the trunk network to learn the robust representations of different finger vein datasets as well as to compensate for the heterogeneity between them. In the proposed framework, two lightweight segmentation networks are designed as the trunk branch and the auxiliary branch to achieve real-time extraction of finger vein patterns. Our approach has been extensively tested on four finger vein datasets available to the public, and the results show that our Lite-HDNet not only improves segmentation performance on all datasets but also effectively reduces heterogeneity between different domains. In addition, we also validated the real-time performance of Lite-HDNet on NVIDIA embedded terminals, proving the outperformance of our approach compared with previous lightweight segmentation networks.

关键词： Feature extraction image segmentation real-time systems Training Adaptation models Data mining deep learning Knowledge transfer Fingers Veins domain adaptation finger vein extraction knowledge transfer

来源：评论

学校读者我要写书评

暂无评论

Intelligent Infrastructure for Traffic Monitoring Based on deep learning and Edge Computing

引用

JOURNAL OF ADVANCED TRANSPORTATION 2024年第1期2024卷

作者： Villa, Jaime Garcia, Franz Jover, Ruben Martinez, Ventura Armingol, Jose M. Univ Carlos III Madrid Intelligent Syst Lab LSI Res Grp Leganes 28911 Spain Scyr Conces S L Madrid 28027 Spain

In the field of traffic management and control systems, we are witnessing a symbiotic evolution, where intelligent infrastructure is progressively collaborating with smart vehicles to produce benefits for traffic monitoring and security, by rapidly identifying hazardous behaviours. This exponential growth is due to the rapid development of deep learning in recent years, as well as the improvements in computer vision models. These technologies allow for monitoring tasks without the need to install numerous sensors or stop the traffic, using the extensive camera network of surveillance cameras already present in worldwide roads. This study proposes a computer vision-based solution that allows for real-time processing of video streams through edge computing devices, eliminating the need for Internet connectivity or dedicated sensors. The proposed system employs deep learning algorithms and vision techniques that perform vehicle detection, classification, tracking, speed estimation, and vehicle geolocation.

关键词： learning algorithms

来源：评论

学校读者我要写书评

暂无评论

Automated Reservoir Characterization of Carbonate Rocks using deep learning image Segmentation Approach

引用

SPE JOURNAL 2024年第8期29卷 4356-4375页

作者： Nande, Soumitra B. Patwardhan, Samarth D. Dr Vishwanath Karad MIT World Peace Univ Dept Petr Engn Pune India

The objective of this study is to develop a systematic and novel workflow for the automated and objective characterization of carbonate reservoirs with the help of deep learning architectures. An image database of more than 6,000 carbonate thin- section images was generated using the optical microscope and image augmentation techniques. Five features, namely clay/silt/mineral, calcite, pores, fossils, and opaque minerals, were identified with the help of manual petrography of the thin sections under the microscope. A total of four deep learning models were developed, which included U- Net, U- Net with ResNet34 backbone, U- Net with Mobilenetv2 backbone, and LinkNet with ResNet34 backbone. The Ensemble model of U- Net + ResNet34 and U- Net + MobileNetv2 yielded the highest intersection over union (IoU) score of 75%, followed by the U- Net + ResNet34 model with an IoU score of 61%. The models struggled with class imbalance, which was very prominent in the image database, with classes such as fossils and opaques considered to be rare. The statistical analysis of the relative errors revealed that the major classes play a more important role in increasing the final IoU score as opposed to the common understanding that the rare classes affect the model performance. The novel workflow developed in this paper can be extended to real carbonate reservoirs for time efficient, objective, and accurate characterization.

关键词： geologist neural network artificial intelligence sedimentary rock deep learning rock type machine learning geological subdiscipline geology architecture

来源：评论

学校读者我要写书评

暂无评论

Emotion recognition and interaction of smart education environment screen based on deep learning networks

引用

JOURNAL OF INTELLIGENT SYSTEMS 2025年第1期34卷

作者： Zhao, Wei Qiu, Liguo Hunan Coll Informat Dept Informat Engn Changsha 410200 Peoples R China

Smart education environments combine technologies such as big data, cloud computing, and artificial intelligence to optimize and personalize the teaching and learning process, thereby improving the efficiency and quality of education. This article proposes a dual-stream-coded image sentiment analysis method based on both facial expressions and background actions to monitor and analyze learners' behaviors in real time. By integrating human facial expressions and scene backgrounds, the method can effectively address the occlusion problem in uncontrolled environments. To enhance the accuracy and efficiency of emotion recognition, a multi-task convolutional network is employed for face extraction, while 3D convolutional neural networks optimize the extraction process of facial features. Additionally, the adaptive learning screen adjustment system proposed in this article dynamically adjusts the presentation of learning content to optimize the learning environment and enhance learning efficiency by monitoring learners' expressions and reactions in real time. By analyzing the experimental results on the Emotic dataset, the emotion recognition model in this article shows high accuracy, especially in the recognition of specific emotion categories. This research significantly contributes to the field of smart education environments by providing an effective solution for real-time emotion recognition.

关键词： deep neural network MTCNN 3D-CNN intelligent education emotion recognition

来源：评论

学校读者我要写书评

暂无评论

Digital image Art Style Transfer Based on deep Short and Long Term Memory

Journal of Network Intelligence

引用

Journal of Network Intelligence 2024年第1期9卷 413-426页

作者： Qiao, Hui Bei, Yan-Jing Jang, Dongyeul College of Art Zhejiang Shuren University Zhejiang Hangzhou310000 China Zhejiang Zhongnan Animation Co. Ltd. Zhejiang Hangzhou310000 China International Graduate School of Convergence Design Hanseo University Ruishan 31962 Korea Republic of

With the rise of artificial intelligence, deep learning techniques are increas-ingly being used in real-life applications, especially in image processing. People have started to use image processing techniques based on deep learning technology to accom-plish the task of image art creation, and one of the more popular directions is image style transfer. The traditional image style transfer method is difficult to meet the requirements of practical applications in terms of visual effect, therefore, a digital image art style transfer method based on deep Long Short-Term Memory (LSTM) is proposed. Firstly, the spatial texture feature data is forgotten and filtered by weight through input gates, forgetting gates, output gates and storage elements, so as to selectively choose some data for cyclic iterative training. Secondly, the Whale Optimization Algorithm (WOA) is introduced to perform intelligent solution of the LSTM parameters, thus proposing the WOA-LSTM algorithm. Finally, the total variational regularization based on L2 parametrization is introduced in the image style transfer process to improve the spatial smoothness of the synthesized images. The experimental results show that the problem of significant content distortion in the generated images can be effectively improved by set-ting the WOA parameters reasonably. Compared with other style transfer algorithms, the average absolute value error (MAE) of the proposed WOA-LSTM algorithm is reduced by 6.5 %. The proposed method can effectively solve the problems of unnaturalness and scatter in the synthetic images and obtain better visual effects for the human eye. © 2024.

关键词： Long short-term memory

来源：评论

学校读者我要写书评

暂无评论

Application of Product Form Recognition Combined with deep learning Algorithm

引用

Computer-Aided Design and Applications 2024年第S15期21卷 54-68页

作者： Wu, Xia Zhu, Leiming Creative Arts School Jinhua Polytechnic Jinhua321017 China College of Design Wenzhou Technology Wenzhou325035 China

CAD plays an important role in current product form recognition. How to accurately identify the product form and improve the design efficiency has become an urgent demand for garment CAD design. This article aims to explore the application of deep learning (DL) technology in clothing product form recognition and design. This article constructs a clothing product shape recognition and classification model based on ACNN and machine vision technology, which adopts the ACNN approach. After experimental verification, the model exhibits superior performance in processing time and efficiency. Compared to traditional CNN and LSTM models, ACNN has higher classification accuracy and shorter processing time when processing clothing product images. These results provide intelligent methods and tools for clothing CAD design and demonstrate the potential of DL technology. In the future, further optimization of models, exploration of multimodal data fusion, cross-domain applications, and real-time interactive design can bring more innovation and breakthroughs to the field of product design. © 2024 U-turn Press LLC.

关键词： Computer vision

来源：评论

学校读者我要写书评

暂无评论

Tiny Machine learning for Concept Drift

引用

IEEE TRANSACTIONS ON NEURAL NETWORKS AND learning SYSTEMS 2024年第6期35卷 8470-8481页

作者： Disabato, Simone Roveri, Manuel Politecn Milan Dipartimento Elettron Informaz Bioingegneria DEIB I-20133 Milan Italy

Tiny machine learning (TML) is a new research area whose goal is to design machine and deep learning (DL) techniques able to operate in embedded systems and the Internet-of-Things (IoT) units, hence satisfying the severe technological constraints on memory, computation, and energy characterizing these pervasive devices. Interestingly, the related literature mainly focused on reducing the computational and memory demand of the inference phase of machine and deep learning models. At the same time, the training is typically assumed to be carried out in cloud or edge computing systems (due to the larger memory and computational requirements). This assumption results in TML solutions that might become obsolete when the process generating the data is affected by concept drift (e.g., due to periodicity or seasonality effect, faults or malfunctioning affecting sensors or actuators, or changes in the users' behavior), a common situation in real-world application scenarios. For the first time in the literature, this article introduces a TML for concept drift (TML-CD) solution based on deep learning feature extractors and a k-nearest neighbors (k-NNs) classifier integrating a hybrid adaptation module able to deal with concept drift affecting the data-generating process. This adaptation module continuously updates (in a passive way) the knowledge base of TML-CD and, at the same time, employs a change detection test (CDT) to inspect for changes (in an active way) to quickly adapt to concept drift by removing obsolete knowledge. Experimental results on both image and audio benchmarks show the effectiveness of the proposed solution, whilst the porting of TML-CD on three off-the-shelf micro-controller units (MCUs) shows the feasibility of what is proposed in real-world pervasive systems.

关键词： Adaptation concept drift deep learning (DL) k-nearest neighbor (k-NN) tiny machine learning (TML)

来源：评论

学校读者我要写书评

暂无评论

real time Car Model and Plate Detection System by Using deep learning Architectures

引用

IEEE ACCESS 2024年 12卷 107616-107630页

作者： Mustafa, Twana Karabatak, Murat Firat Univ Dept Software Engn Elazig Turkiye Knowledge Univ Coll Sci Dept Comp Sci Erbil 44001 Iraq

The advent of deep learning has revolutionized computer vision, enabling real-time analysis crucial for traffic management and vehicle identification. This research introduces a system combining vehicle make and model detection with Automatic Number Plate Recognition (ANPR), achieving a groundbreaking 97.5% accuracy rate. Unlike traditional methods, which focus on either make and model detection or ANPR independently, this study integrates both aspects into a single, cohesive system, providing a more holistic and efficient solution for vehicle identification, ensuring robust performance even in adverse weather conditions. The paper explores the use of deep learning techniques, including OpenCV, in combination with Python programming language. Leveraging MobileNet-V2 and YOLOx (You Only Look Once) for vehicle identification, and YOLOv4-tiny, Paddle OCR (optical character recognition), and SVTR-tiny for ANPR, the system was rigorously tested at Firat University's entrance with a thousand images captured under various conditions such as fog, rain, and low light. The system's exceptional success rate in these tests highlights its robustness and practical applicability. Additionally, experiments evaluate the system's accuracy and effectiveness, using Gradient-weighted Class Activation Mapping (GradCam) technology to gain insights into neural networks' decision-making processes and identify areas for improvement, particularly in misclassifications. The implications of this research for computer vision are significant, paving the way for advanced applications in autonomous driving, traffic management, stolen vehicles, and security surveillance. Achieving real-time, high-accuracy vehicle identification, the integrated Vehicle Make and Model Recognition (VMM R) and ANPR system sets a new standard for future research in the field.

关键词： License plate recognition Computational modeling image recognition deep learning Computer vision Training Convolutional neural networks Automobiles YOLO Car model plate detection deep learning computer vision OpenCV MobileNet-V2 YOLOv4 GradCam Firat University

来源：评论

学校读者我要写书评

暂无评论

Adaptive discrete wavelet transform and optimized residual-based deep CNN for image dehazing with a new meta-heuristic algorithm

引用

MULtimeDIA TOOLS AND APPLICATIONS 2024年第28期83卷 71335-71358页

作者： Kumar, R. Prakash Naik, N. Manaja Univ B D T Coll Engn Dept Elect & Commun Engn Davangere Karnataka India Visvesvaraya Technol Univ Belgavi Karnataka India

image dehazing is said to be an emerging research area in the platform of computer vision and image processing. Due to the cruel fog, air dispersion, and haze around the environment, the hazes images are resulted in different challenges in retrieving the actual information of the original image. On the other hand, the conventional approaches are ensured with the huge computational complexity and also with the distortion of actual images like over-saturation and halos. The recent methods are used for restoring the haze-free images however they are worked with the physical models and along with the learning methods. It is a very challenging task to maintain the detailed details of the image at the time of reducing the fog in the single-image dehazing. With an advanced development deep structured strategy, mostly Convolutional Neural Network (CNN)-aided dehazing approaches are developed for processing the single image dehazing. However, haze residual and slow training of the convergence rate are considered as the two main drawbacks in these conventional dehazing networks. To deal with these problems, the latest approach is proposed for the restoration of haze-free images. The hazy images are gathered from the standard datasets. At first, Adaptive Discrete Wavelet Transform (ADWT) is utilized for decomposing the images, where the ADWT is implemented by Hybrid African Vultures Fire Fly Optimization (HAVFFO). Further, image dehazing is designed by Optimized Residual-Based deep CNN (OR-deep CNN), where the hyperparameters of the Residual-Based deep CNN are optimized by the same HAVFFO. Finally, the restoration of haze-free images is carried out through adaptive inverse DWT. Through the performance analysis, our recommended model is better in quantitative visual and performances on online resources.

关键词： image dehazing Adaptive discrete wavelet transform Hybrid African vultures fire fly optimization Optimized residual-based deep convolutional neural network deep learning

来源：评论

学校读者我要写书评

暂无评论

RL-LOGO: deep REINFORCEMENT learning LOCALIZATION FOR LOGO RECOGNITION 49

RL-LOGO: DEEP REINFORCEMENT LEARNING LOCALIZATION FOR LOGO R...

引用

49th IEEE International Conference on Acoustics, Speech, and Signal processing (ICASSP)

作者： Fujitake, Masato Fast Accounting Co Ltd FA Res Tokyo Japan

ISBN: (纸本)9798350344868;9798350344851

This paper proposes a novel logo image recognition approach incorporating a localization technique based on reinforcement learning. Logo recognition is an image classification task identifying a brand in an image. As the size and position of a logo vary widely from image to image, it is necessary to determine its position for accurate recognition. However, because there is no annotation for the position coordinates, it is impossible to train and infer the location of the logo in the image. Therefore, we propose a deep reinforcement learning localization method for logo recognition (RL-LOGO). It utilizes deep reinforcement learning to identify a logo region in images without annotations of the positions, thereby improving classification accuracy. We demonstrated a significant improvement in accuracy compared with existing methods in several published benchmarks. Specifically, we achieved an 18-point accuracy improvement over competitive methods on the complex dataset Logo-2K+. This demonstrates that the proposed method is a promising approach to logo recognition in real-world applications.

关键词： Logo Recognition image Classification deep Reinforcement 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：