检索结果-内蒙古大学图书馆

A hardware accelerator to support deep learning processor units in real-time image processing

ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE 2025年 145卷

作者： Cittadini, Edoardo Marinoni, Mauro Buttazzo, Giorgio Scuola Super Sant Anna Pisa Italy

deep neural networks are becoming crucial in many cyber-physical systems involving complex perceptual tasks. For those embedded systems requiring real-time interactions with dynamic environments, as autonomous robots and drones, it is of paramount importance that such algorithms are efficiently executed onboard on properly designed hardware accelerators to meet the required performance specifications. In particular, some neural network architectures for object detection and tracking, as You Only Look Once (YOLO), include heavy computational stages that need to be executed before and after the model inference. Such stages are typically not incorporated in traditional accelerators and are executed on general-purpose processors, thus introducing a bottleneck in the overall processing pipeline. To overcome such a problem, this paper presents a generalpurpose accelerator on a field-programmable gate array (FPGA) able to run pre-processing and post-processing operations typically required by vision tasks. The proposed solution has been tested in combination with a YOLO object detector accelerated on an Advanced Micro Devices (AMD) Xilinx Kria KR260 board mounting an UltraScale+ multiprocessor system-on-chip, achieving a significant improvement in terms of both timing performance and power consumption, and enabling onboard visual processing into drones. The proposed solution is able to boost the traditional object detection process by a factor of 4.4, allowing the execution of the full processing pipeline at 60 frames per second (fps), versus 13.6 fps reachable without the proposed accelerator. Asa result, this work enables the use of high-speed cameras for developing more reactive systems that can respond to incoming events with lower latency.

关键词： Hardware acceleration Object detection deep neural networks real-time processing image pre-processing acceleration deep processing unit

来源：评论

学校读者我要写书评

暂无评论

MobileMEF: fast and efficient method for real-time mobile multi-exposure fusion

引用

JOURNAL OF real-time image processing 2025年第1期22卷 1-15页

作者： Kirsten, Lucas Nedel Fu, Zhicheng Madhusudhana, Nikhil Ambha Motorola Mobil Comercio Prod Eletron Ltda Jaguariuna SP Brazil Lenovo Res Chicago IL USA

Recent advances in camera design and imaging technology have enabled the capture of high-quality images using smartphones. However, due to the limited dynamic range of digital cameras, the quality of photographs captured in environments with highly imbalanced lighting often results in poor-quality images. To address this issue, most devices capture multi-exposure frames and then use some multi-exposure fusion method to merge those frames into a final fused image. Nevertheless, most traditional and current deep learning approaches are unsuitable for real-time applications on mobile devices due to their heavy computational and memory requirements. We propose MobileMEF, a new method for multi-exposure fusion based on an encoder-decoder deep learning architecture with efficient building blocks tailored for mobile devices. This efficient design makes MobileMEF capable of processing 4K resolution images in less than 2 s on mid-range smartphones. MobileMEF outperforms state-of-the-art techniques regarding full-reference quality measures and computational efficiency (runtime and memory usage), making it ideal for real-time applications on hardware-constrained devices. Our code is available at: https://***/LucasKirsten/MobileMEF.

关键词： image fusion Multi-exposure image real-time methods Smartphone photography

来源：评论

学校读者我要写书评

暂无评论

deep learning image burst stacking to reconstruct high-resolution ground-based solar observations

引用

ASTRONOMY & ASTROPHYSICS 2025年 693卷

作者： Schirninger, C. Jarolim, R. Veronig, A. M. Kuckein, C. Karl Franzens Univ Graz Inst Phys Univ Pl 5 A-8010 Graz Austria Natl Ctr Atmospher Res High Altitude Observ 3080 Ctr Green Dr Boulder CO USA Karl Franzens Univ Graz Kanzelhohe Observ Solar & Environm Res Graz Austria Inst Astrofis Canarias IAC Via Lactea S-N E-38205 San Cristobal la Laguna Tenerife Spain Univ La Laguna Dept Astrofis E-38206 San Cristobal la Laguna Tenerife Spain Max Planck Inst Sonnensystemforsch Justus Von Liebig Weg 3 D-37077 Gottingen Germany

Context. Large aperture ground-based solar telescopes allow the solar atmosphere to be resolved in unprecedented detail. However, ground-based observations are inherently limited due to Earth's turbulent atmosphere, requiring image correction techniques. Aims. Recent post-image reconstruction techniques are based on using information from bursts of short-exposure images. Shortcomings of such approaches are the limited success, in case of stronger atmospheric seeing conditions, and computational demand. real-time post-image reconstruction is of high importance to enabling automatic processing pipelines and accelerating scientific research. In an attempt to overcome these limitations, we provide a deep learning approach to reconstruct an original image burst into a single high-resolution high-quality image in real time. Methods. We present a novel deep learning tool for image burst reconstruction based on image stacking methods. Here, an image burst of 100 short-exposure observations is reconstructed to obtain a single high-resolution image. Our approach builds on unpaired image-to-image translation. We trained our neural network with seeing degraded image bursts and used speckle reconstructed observations as a reference. With the unpaired image translation, we aim to achieve a better generalization and increased robustness in case of increased image degradations. Results. We demonstrate that our deep learning model has the ability to effectively reconstruct an image burst in real time with an average of 0.5 s of processing time while providing similar results to standard reconstruction methods. We evaluated the results on an independent test set consisting of high- and low-quality speckle reconstructions. Our method shows an improved robustness in terms of perceptual quality, especially when speckle reconstruction methods show artifacts. An evaluation with a varying number of images per burst demonstrates that our method makes efficient use of the combined image info

关键词： atmospheric effects techniques: image processing telescopes Sun: atmosphere Sun: photosphere

来源：评论

学校读者我要写书评

暂无评论

Syflo: augmenting yolo for real-time health monitoring of electric assets in power transmission lines

引用

JOURNAL OF real-time image processing 2025年第1期22卷 1-15页

作者： Sankuri, Raja Sekhar Sristy, Nagesh Bhattu Karri, Sri Phani Krishna Natl Inst Technol Andhra Pradesh Dept Elect Engn Tadepalligudem India Natl Inst Technol Andhra Pradesh Dept Comp Sci & Engn Tadepalligudem India

Sustainable transmission of electrical energy to consumers across regions relies heavily on the integrity of power transmission lines and continuous monitoring of assets is crucial for maintaining system reliability. Unmanned aerial vehicles have revolutionized defect identification in real-time and accessibility, even in difficult-to-reach geographical landscapes, thereby improving image-based inspections. This work introduces semisupervised Yolo with focal loss function (SYFLo), a novel method that augments YOLO for real-time health monitoring of electric assets in power transmission lines. SYFLo integrates the focal loss function with semi-supervised learning to effectively address the lack of abundant labeled data, data imbalances and enhance detection accuracy. Additionally, it improves data generalizability across a wide range of images, ensuring robust performance despite varied image backgrounds. By leveraging YOLOv8, SYFLo significantly improves fault identification, achieving a detection accuracy of 96.5% and an FPS of 16.39. Experimental results demonstrate the impact of the proposed approach, highlighting its potential to enhance the reliability of power transmission line monitoring. These findings underscore the importance of integrating advanced deep learning techniques with innovative loss functions to address common challenges in real-time health monitoring systems.

关键词： deep learning Power grid inspection Defect detection Semi-supervised learning Focal loss function

来源：评论

学校读者我要写书评

暂无评论

A novel hybrid Bayesian-optimized CNN-SVM deep learning model for real-time surface roughness classification and prediction based on in-process machined surface image analysis

引用

INTERNATIONAL JOURNAL OF INTERACTIVE DESIGN AND MANUFACTURING - IJIDEM 2025年 1-18页

作者： Arif, Abdul Rao, Ponugoti Gangadhara Prasad, Kalapala JNTUK Univ Coll Engn A Kakinada 533003 India Aditya Univ Dept Mech Engn Surampalem 533437 AP India JNTUK Dept Mech Engn Univ Coll Engn A Kakinada 533003 India

This study presents a novel hybrid Bayesian-optimized CNN-SVM deep learning model for real-time surface roughness classification and prediction based on in-process machined surface image analysis. The hybrid deep learning model, achieves unprecedented accuracy in real-time surface roughness classification and prediction from image data of Inconel 716 superalloy machined surface, eliminating human intervention for feature extraction and selection procedures. The Bayesian optimization algorithm (BOA) was employed to fine-tune the CNN model's hyperparameters. The derived optimal hyperparameters were used to train the CNN model, which achieved a 92.37% classification accuracy on a test dataset. The classification performance was further enhanced by integrating an SVM classifier achieving an accuracy of 98.21%. Additionally, the model's performance was compared with five state-of-the-art deep learning architectures (AlexNet, ResNet50, VGGNet, Faster R-CNN ResNet50, and CNN), with the proposed model showing superior performance in terms of precision (98.02%), sensitivity (98.40%), and F1-score (98.20%). For surface roughness estimation, the hybrid model demonstrated a 96.89% accuracy in predicting roughness values, highlighting its potential for real-time in-process quality monitoring in industrial applications. These results indicate the proposed hybrid approach demonstrated substantially better performance than other popular object detection models. Hence, the presented work demonstrates an efficient end-to-end approach based on the in-process acquired machined surface image analysis by eliminating human intervention in the process of image processing, feature extraction, and feature selection for the real-time classification and prediction of the surface roughness process.

关键词： Surface roughness prediction image-based classification Regression deep learning Convolutional neural networks Hyperparameters Bayesian optimization Support vector machine

来源：评论

学校读者我要写书评

暂无评论

Bubble evolution identification method based on U-Net deep learning

引用

SHIPS AND OFFSHORE STRUCTURES 2025年

作者： Gou, Yuxin Shi, Dongyan Wang, Jiuqiang Zhang, Haifeng Cui, Xiongwei Harbin Engn Univ Coll Mech & Elect Engn Harbin Peoples R China Harbin Engn Univ Coll Shipbldg Engn 145 Nantong St Harbin 150001 Heilongjiang Peoples R China

This study addresses the challenge of automated analysis of underwater bubble dynamics, as traditional methods for bubble detection often struggle with precision and real-time processing, requiring labour-intensive manual analysis or inefficient image processing techniques. To overcome these limitations, we propose a U-Net deep learning-based method for bubble segmentation and dynamic analysis, utilising high-speed photography (62420 fps) to capture bubble formation and dissipation. The proposed method achieves accurate bubble segmentation through deep feature extraction and upsampling using the U-Net model, with a proportional estimation method applied to calculate bubble size, expansion rate, and movement rate. Validation through manual measurements and CFD simulations confirms the method's reliability. The results demonstrate that the U-Net model enhances the efficiency and accuracy of bubble identification, providing a robust tool for studying underwater bubble dynamics in real-time.

关键词： U-net deep learning bubble evolution identification underwater electric spark experiment expansion and movement rates automated image segmentation

来源：评论

学校读者我要写书评

暂无评论

Efficient NPU-GPU scheduling for real-time deep learning inference on mobile devices

引用

JOURNAL OF real-time image processing 2025年第2期22卷 1-13页

作者： Yu, Chengwu Wang, Meng Chen, Shan Wang, Wanqi Fang, Weiwei Chen, Yanming Xiong, Neal N. Beijing Jiaotong Univ Sch Comp Sci & Technol Beijing 100044 Peoples R China Hubei Engn Res Ctr Intelligent Detect & Identifica Wuhan 430205 Hubei Peoples R China Anhui Univ Sch Comp Sci & Technol Hefei 230601 Anhui Peoples R China Sul Ross State Univ Dept Comp Sci & Math Alpine TX 79830 USA

As the need for on-device artificial intelligence (AI) has increased in recent years, mobile devices tend to be equipped with multiple heterogeneous processors, including CPU, GPU, and Neural processing Unit (NPU). While NPUs can offer low-cost and real-time AI processing capabilities for deep Neural Network (DNN) inference, its limited resources often lead to a trade-off between performance and accuracy, potentially resulting in a non-trivial accuracy drop. To address this problem, we propose a new NPU-GPU Scheduling (NGS) framework for DNN-based video analytics. The main challenge lies in determining when and how to execute inference on the NPU/GPU to satisfy the performance objectives. To make more precise scheduling decisions, we first propose a new image complexity assessment model to replace the existing normalized edge density metric. Then, we formulate the scheduling problem with the objective of maximizing inference accuracy under the given latency constraint, and introduce an adaptive solution based on dynamic programming to determine which frames should be processed on the GPU and when to exit from inference for each of them. Extensive experiments conducted on a real mobile device show that our NGS framework substantially outperforms other solutions, and achieves a close-to-oracle performance.

关键词： Edge computing DNN inference Heterogeneous processors Task scheduling Dynamic Programming

来源：评论

学校读者我要写书评

暂无评论

real-time Facial Expression Recognition Based on image processing in Virtual reality

引用

INTERNATIONAL JOURNAL OF COMPUTATIONAL INTELLIGENCE SYSTEMS 2025年第1期18卷 1-16页

作者： Gong, Qingzhen Liu, Xuefang Ma, Yongqiang Jining Normal Univ Sch Phys & Elect Informat Engn Ulanqab 012000 Inner Mongolia Peoples R China Jingdezhen Univ Sch Informat Engn Jingdezhen 333000 Jiangxi Peoples R China Jining Normal Univ Sch Comp & Big Data Ulanqab 012000 Inner Mongolia Peoples R China

More virtual reality (VR) scenarios have become more prevalent in recent years. More and more people are getting into VR, meaning that objective physiological measures to assess a user's emotional state automatically are becoming more critical. Individuals' emotional states impact their behaviour, opinions, emotions, and decisions. They may be used to analyze VR experiences and make systems react to and engage with the user's emotions. VR environments require users to wear head-mounted displays (HMDs), blocking off their upper faces. That makes traditional Facial Expression Recognition (FER) approaches very limited in their usefulness. Thus, a deep learning (DL) solution combined with image processing is utilized to classify universal emotions: sadness, happiness, disgust, anger, fear and surprise. Hence, this paper suggests the deep Automatic Facial Expression Recognition Model (DAFERM) for interactive virtual reality (VR) applications such as intelligent education, social networks, and virtual training. Two main parts comprise the system: one that uses deep neural networks (DNNs) for facial emotion identification and another that automatically tracks and segments faces. The system begins by following a marker on the front of the head-mounted display (HMD). With the help of the spatial data that has been retrieved, the positions and rotations of the face are estimated to segment the mouth. Finally, the system interacts with DNN using the pixels processed by the lips. It obtains the facial expression results in real time using an adaptive method for histogram-based mouth segmentation.

关键词： Facial expression recognition deep learning Virtual reality image processing deep neural network

来源：评论

学校读者我要写书评

暂无评论

Uncertainty-Aware real-time Visual Anomaly Detection With Conformal Prediction in Dynamic Indoor Environments

引用

IEEE ROBOTICS AND AUTOMATION LETTERS 2025年第5期10卷 4468-4475页

作者： Saboury, Arya Uyguroglu, Mustafa Kemal Eastern Mediterranean Univ Dept Elect & Elect Engn TR-99628 Mersin Turkiye

This letter presents an efficient visual anomaly detection framework designed for safe autonomous navigation in dynamic indoor environments, such as university hallways. The approach employs an unsupervised autoencoder method within deep learning to model regular environmental patterns and detect anomalies as deviations in the embedding space. To enhance reliability and safety, the system integrates a statistical framework, conformal prediction, that provides uncertainty quantification with probabilistic guarantees. The proposed solution has been deployed on a real-time robotic platform, demonstrating efficient performance under resource-constrained conditions. Extensive hyperparameter optimization ensures the model remains dynamic and adaptable to changes, while rigorous evaluations confirm its effectiveness in anomaly detection. By addressing challenges related to real-time processing and hardware limitations, this work advances the state-of-the-art in autonomous anomaly detection. The probabilistic insights offered by this framework strengthen operational safety and pave the way for future developments, such as richer sensor fusion and advanced learning paradigms. This research highlights the potential of uncertainty-aware deep learning to enhance safety monitoring frameworks, thereby enabling the development of more reliable and intelligent autonomous systems for real-world applications.

关键词： Anomaly detection Autoencoders image reconstruction Robots Training Safety Uncertainty Reliability real-time systems Probabilistic logic deep learning for visual perception probability and statistical methods anomaly detection conformal prediction

来源：评论

学校读者我要写书评

暂无评论

Key Technologies for Machine Vision for Picking Robots:Review and Benchmarking

引用

Machine Intelligence Research 2025年第1期22卷 2-16页

作者： Xu Xiao Yiming Jiang Yaonan Wang College of Electrical and Information Engineering Hunan UniversityChangsha 410082China National Engineering Research Center for Robot Vision Perception and Control Technology Hunan UniversityChangsha 410082China

The increase in precision agriculture has promoted the development of picking robot technology,and the visual recognition system at its core is crucial for improving the level of agricultural *** paper reviews the progress of visual recognition tech-nology for picking robots,including image capture technology,target detection algorithms,spatial positioning strategies and scene *** article begins with a description of the basic structure and function of the vision system of the picking robot and em-phasizes the importance of achieving high-efficiency and high-accuracy recognition in the natural agricultural ***-sequently,various image processing techniques and vision algorithms,including color image analysis,three-dimensional depth percep-tion,and automatic object recognition technology that integrates machine learning and deep learning algorithms,were *** the same time,the paper also highlights the challenges of existing technologies in dynamic lighting,occlusion problems,fruit maturity di-versity,and real-time processing *** paper further discusses multisensor information fusion technology and discusses methods for combining visual recognition with a robot control system to improve the accuracy and working rate of *** the same time,this paper also introduces innovative research,such as the application of convolutional neural networks(CNNs)for accurate fruit detection and the development of event-based vision systems to improve the response speed of the *** the end of this paper,the future development of visual recognition technology for picking robots is predicted,and new research trends are proposed,including the refinement of algorithms,hardware innovation,and the adaptability of technology to different agricultural *** purpose of this paper is to provide a comprehensive analysis of visual recognition technology for researchers and practitioners in the field of agricul-tural rob

关键词： Picking robot visual system perception technology image processing machine learning deep 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：