检索结果-内蒙古大学图书馆

TinyCount: an efficient crowd counting network for intelligent surveillance

JOURNAL OF real-time image processing 2024年第4期21卷 153页

作者： Lee, Hyeonbeen Lee, Jangho Incheon Natl Univ Dept Comp Sci & Engn Incheon 22012 South Korea

Crowd counting, the task of estimating the total number of people in an image, is essential for intelligent surveillance. Integrating a well-trained crowd counting network into edge devices, such as intelligent CCTV systems, enables its application across various domains, including the prevention of crowd collapses and urban planning. For a model to be embedded in edge devices, it requires robust performance, reduced parameter count, and faster response times. This study proposes a lightweight and powerful model called TinyCount, which has only 60k parameters. The proposed TinyCount is a fully convolutional network consisting of a feature extract module (FEM) for robust and rapid feature extraction, a scale perception module (SPM) for scale variation perception and an upsampling module (UM) that adjusts the feature map to the same size as the original image. TinyCount demonstrated competitive performance across three representative crowd counting datasets, despite utilizing approximately 3.33 to 271 times fewer parameters than other crowd counting approaches. The proposed model achieved relatively fast inference times by leveraging the MobileNetV2 architecture with dilated and transposed convolutions. The application of SEblock and findings from existing studies further proved its effectiveness. Finally, we evaluated the proposed TinyCount on multiple edge devices, including the Raspberry Pi 4, NVIDIA Jetson Nano, and NVIDIA Jetson AGX Xavier, to demonstrate its potential for practical applications.

关键词： Crowd counting Lightweight network Intelligent surveillance Artificial intelligence deep learning Edge machine learning

来源：评论

学校读者我要写书评

暂无评论

Fast and efficient computing for deep learning-based defect detection models in lightweight devices

引用

JOURNAL OF INTELLIGENT MANUFACTURING 2024年 1-16页

作者： Fisne, Alparslan Kalay, Alperen Eken, Suleyman Aselsan Inc Ankara Turkiye Kocaeli Univ Informat Syst Engn TR-41001 Izmit Turkiye

Defect anomaly detection is beneficial in the production cycle of various industries. It is widely used in areas such as metal surface and fabric industries. This paper focuses on deep learning-driven defect detection models using energy-efficient computing. We concentrate on a segmentation-based defect detection model for metal surface anomaly detection, while we deal with a deconvolution-based defect detection model for fabric defects in this work. We propose a depth-wise convolution structure for the segmentation-based visual defect detection model. In addition, we apply the optimizations supported by the inference engine to two models. The segmentation-based defect detection model inference is approximately 10x faster than the original. Furthermore, the real-time requirement is achieved in a lightweight vision processing unit (VPU) device with a power consumption of only 1.5 Watts for the fabric defect detection model. The practical values of this work are multifaceted, offering substantial benefits in terms of cost reduction, product quality, real-time processing, energy efficiency, and scalability. These advancements not only improve operational efficiency but also contribute to sustainability efforts and provide a competitive advantage in the industry.

关键词： Defect detection Energy efficiency deep learning Edge computing

来源：评论

学校读者我要写书评

暂无评论

Multi-view image Fusion Using Ensemble deep learning Algorithm For MRI And CT images

引用

ACM TRANSACTIONS ON ASIAN AND LOW-RESOURCE LANGUAGE INFORMATION processing 2024年第3期23卷 1-24页

作者： Thenmoezhi, N. Perumal, B. Lakshmi, A. Kalasalingam Acad Res & Educ Dept Elect & Commun Engn Sch Elect Elect & Biomed Technol Krishnankoil 626126 Tamil Nadu India Ramco Inst Technol Dept Elect & Commun Engn Rajapalayam 626117 Tamil Nadu India

Medical image fusions are crucial elements in image-based health care diagnostics or therapies and generic applications of computer visions. However, the majority of existing methods suffer from noise distortion that affects the overall output. When pictures are distorted by noises, classical fusion techniques perform badly. Hence, fusion techniques that properly maintain information comprehensively from multiple faulty pictures need to be created. This work presents Enhanced Lion Swarm Optimization (ESLO) with Ensemble deep learning (EDL) to address the aforementioned issues. The primary steps in this study include image fusions, segmentation, noise reduction, feature extraction, picture classification, and feature selection. Adaptive Median Filters are first used for noise removal in sequence to enhance image quality by eliminating noises. The MRIs and CT images are then segmented using the Region Growing-based k-Means Clustering (RKMC) algorithm to separate the images into their component regions or objects. images in black and white are divided into image. In the white image, the RKMC algorithm successfully considered the earlier tumour probability. The next step is feature extraction, which is accomplished by using the Modified Principal Component Analysis (MPCA) to draw out the most informative aspects of the images. Then the ELSO algorithm is applied for optimal feature selection, which is computed by best fitness values. After that, multi-view image fusions of multi modal images derive lower-, middle-, and higher-level image contents. It is done by using deep Convolution Neural Network (DCNN) and the Tissue-Aware Conditional Generative Adversarial Network (TAcGAN) algorithm, which fuses the multi-view features and relevant image features, and it is used for real-time applications. ELSO +EDL algorithm gives better results in terms of accuracy, Peak Signal-To-Noise Ratio (PSNR), and lower Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE)

关键词： Multi-view image fusions Magnetic Resonance Imaging (MRI) images Computed Tomography (CT) images Enhanced Lion Swarm Optimization (ELSO) Ensemble deep learning (EDL) algorithm

来源：评论

学校读者我要写书评

暂无评论

Meta learning based Object Tracking Technology: A Survey

引用

KSII TRANSACTIONS ON INTERNET AND INFORMATION SYSTEMS 2024年第8期18卷 2067-2081页

作者： Baek, Ji-Won Chung, Kyungyong Kyonggi Univ Dept Comp Sci 154-42 Gwanggyosan Ro Suwon 16227 Gyeonggi Do South Korea Kyonggi Univ Div AI Comp Sci & Engn 154-42 Gwanggyosan Ro Suwon 16227 Gyeonggi Do South Korea

Recently, image analysis research has been actively conducted due to the accumulation of big image data and the development of deep learning. image analytics research has different characteristics from other data such as data size, real-time, image quality diversity, structural complexity, and security issues. In addition, a large amount of data is required to effectively analyze images with deep-learning models. However, in many fields, the data that can be collected is limited, so there is a need for meta learning based image analysis technology that can effectively train models with a small amount of data. This paper presents a comprehensive survey of meta-learning-based object-tracking techniques. This approach comprehensively explores object tracking methods and research that can achieve high performance in datalimited situations, including key challenges and future directions. It provides useful information for researchers in the field and can provide insights into future research directions.

关键词： Convolution Neural Network Object Tracking Meta learning deep learning Object Detection

来源：评论

学校读者我要写书评

暂无评论

Slim-neck by GSConv: a lightweight-design for real-time detector architectures

引用

JOURNAL OF real-time image processing 2024年第3期21卷 62-62页

作者： Li, Hulin Li, Jun Wei, Hanbing Liu, Zheng Zhan, Zhenfei Ren, Qiliang Chongqing Jiaotong Univ Coll Traff & Transportat Chongqing 400074 Peoples R China Chongqing Jiaotong Univ Sch Mechatron & Vehicle Engn Chongqing 400074 Peoples R China Univ British Columbia Okanagan Sch Engn Kelowna BC V1V 1V7 Canada

real-time object detection is significant for industrial and research fields. On edge devices, a giant model is difficult to achieve the real-time detecting requirement, and a lightweight model built from a large number of the depth-wise separable convolutional could not achieve the sufficient accuracy. We introduce a new lightweight convolutional technique, GSConv, to lighten the model but maintain the accuracy. The GSConv accomplishes an excellent trade-off between the accuracy and speed. Furthermore, we provide a design suggestion based on the GSConv, slim-neck (SNs), to achieve a higher computational cost-effectiveness of the real-time detectors. The effectiveness of the SNs was robustly demonstrated in over twenty sets comparative experiments. In particular, the real-time detectors of ameliorated by the SNs obtain the state-of-the-art (70.9% AP(50) for the SODA10M at a speed of similar to 100 FPS on a Tesla T4) compared with the baselines. Code is available at https://***/alanli1997/slim-neck-by-gsconv.

关键词： GSConv deep learning CNNs real-time detection Lightweight Edge computing

来源：评论

学校读者我要写书评

暂无评论

FIRESTART: Fire Ignition Recognition with Enhanced Smoothing Techniques and real-time Tracking 22nd

FIRESTART: Fire Ignition Recognition with Enhanced Smoothing...

引用

22nd International Conference on image Analysis and processing (ICIAP)

作者： Zedda, Luca Loddo, Andrea Di Ruberto, Cecilia Univ Cagliari Dept Math & Comp Sci Cagliari Italy

ISBN: (纸本)9783031510229;9783031510236

Fires can potentially cause significant harm to both people and the environment. Recently, there has been a growing interest in real-time fire and smoke detection to provide practical assistance. Detecting fires in outdoor areas is crucial to safeguard human lives and the environment. This is especially important in situations where more than traditional smoke detectors may be required. In this work, we propose FIRESTART, which aims to achieve accurate and robust ignition detection for prompt identification and response to fire incidents. The proposed framework utilizes a lightweight deep learning architecture and post-processing techniques for fire-starting interval detection. Its evaluation was conducted on the ONFIRE dataset, comparing it with several state-of-the-art methods. The results are encouraging, particularly from computational and real-time use perspectives.

关键词： deep learning Computer Vision image processing Vision Transformers Fire Detection

来源：评论

学校读者我要写书评

暂无评论

Urban traffic monitoring based on deep learning on an embedded GPU

引用

EXPERT SYSTEMS WITH APPLICATIONS 2025年 273卷

作者： Nocua, M. Fredy Perez-Holguin, Wilson-Javier Pardo-Beainy, Camilo Univ Pedag & Tecnol Colombia UPTC Grp GIRA Sogamoso Colombia Tunja Colombia Univ Santo Tomas Grp GIDINT Tunja Colombia Bogota Colombia Fdn Univ San Gil Unisangil COMUNIT Yopal Colombia Santander Colombia

This paper presents a deep learning-based system for urban traffic monitoring, focusing on the detection and tracking of motorcycles using embedded hardware, due to the high accident rates of this type of vehicle. Different convolutional neural network (CNN) models were evaluated, including MobileNet-v1-SSD, YOLOv5, and Faster R-CNN, implemented on an NVIDIA Graphics processing Units (GPUs) board as the Jetson Xavier NX (R). The MobileNet-v1-SSD model stands out for its balance between precision (90 %), recall (66 %), and latency (similar to 10 ms), making it ideal for real-time applications. Additionally, a tracking algorithm based on optical flow using the Lucas-Kanade method was developed, complemented with logic for creating and deleting identities (IDs), enabling object tracking in dynamic scenarios with partial occlusions. The system includes a methodology for calculating key traffic variables such as speed and direction by correlating pixels with real-world distances through camera calibration. This approach demonstrates the feasibility of developing complex image-processing applications based on resource-constrained platforms by leveraging the features of efficient embedded systems such as General Purpose GPUs.

关键词： deep learning Computer vision Object detection Object tracking Embedded system

来源：评论

学校读者我要写书评

暂无评论

Detection method for weld defects in time-of-flight diffraction images based on multi-image fusion and feature hybrid enhancement

引用

ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE 2024年 138卷

作者： Yang, Deyan Jiang, Hongquan Ai, Song Yang, Tianlun Zhi, Zelin Jing, Deqiang Gao, Jianmin Yue, Kun Cheng, Huyue Xu, Yongjun Xi An Jiao Tong Univ State Key Lab Mfg Syst Engn Xian 710049 Peoples R China Dongfang Turbine Co Ltd State Key Lab Clean & Efficient Turbomachinery Pow Deyang 618000 Peoples R China Shaanxi Special Equipment Inspect & Testing Inst Xian 710048 Peoples R China

The accurate recognition of defects in the time-of-flight diffraction (TOFD) images of welds is important to improve the capability and efficiency of defect detection. The existing deep learning-based defect detection methods take a single image as input, without considering the fact that technicians need to observe the image "dynamically" during its evaluation, resulting in low accuracy and credibility of the defect detection results. To address these issues, combining deep learning techniques with TOFD inspection domain knowledge, this article proposes a multi-image fusion and feature hybrid enhancement-based weld defect detection method for TOFD images, comprising three parts: a single-to-multiple image decomposition module based on gain preprocessing, multi-image feature extraction module, and weld defect detection module based on feature hybrid enhancement. The developed method can realize a "dynamically changing" feature extraction and target detection of weld defects in TOFD images. The proposed method was experimentally verified using TOFD images of welds in largescale spherical pressure tanks. This method greatly surpassed the current state-of-the-art approaches, including You Only Look Once (YOLO) v9, YOLOv10, and real-time DEtection TRansformer (RT-DETR), achieving a mean average precision of 82.0%, average precision for small-size targets of 45.2%, and average recall for small-size targets of 70.9%. The detection time for a single TOFD image with a resolution of 500 x 1350 pixels is 0.1287 s, satisfying the real-time requirements for weld TOFD inspection in practical engineering applications. The proposed method can also be extended to engineering applications such as intelligent detection of weld defects based on X-ray images.

关键词： time-of-flight diffraction deep learning Defect detection image decomposition Feature hybrid enhancement

来源：评论

学校读者我要写书评

暂无评论

Research on Mainlobe Jamming Suppression Algorithm for Airborne Bistatic Radar Based on deep Neural Networks 24

Research on Mainlobe Jamming Suppression Algorithm for Airbo...

引用

2024 International Conference on Virtual reality, image and Signal processing, ICVISP 2024

作者： Chen, Lu Xia, Deping Liu, Deshun Nanjing Research Institute of Electronic Technology Nanjing China

ISBN: (纸本)9798400710926

To address issues such as mainlobe distortion and sidelobe elevation in airborne monostatic radar for mainlobe jamming suppression, this paper proposes a mainlobe jamming suppression algorithm for airborne bistatic radar based on deep neural networks. This method leverages the real-time computational capabilities and powerful learning abilities of deep neural networks, using unsupervised learning to dynamically compute weights. These weights are then multiplied with the received signals from the auxiliary station to accurately estimate the jamming signal at the main station, achieving mainlobe jamming cancellation for both the primary and auxiliary radars. Simulation results show that using deep neural networks to calculate weight vectors for mainlobe jamming cancellation not only effectively suppresses mainlobe jamming but also reduces the signal-to-noise ratio loss of the target to about 3dB, significantly enhancing the system's real-time performance and anti-jamming capabilities. © 2024 Copyright held by the owner/author(s).

关键词： Signal to noise ratio

来源：评论

学校读者我要写书评

暂无评论

ADDITION: Detecting Adversarial Examples With image-Dependent Noise Reduction

引用

IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING 2024年第3期21卷 1139-1154页

作者： Wang, Yuchen Li, Xiaoguang Yang, Li Ma, Jianfeng Li, Hui Xidian Univ Sch Comp Sci & Technol Xian 710071 Peoples R China Xidian Univ Sch Cyber Engn State Key Lab Integrated Serv Networks Xian 710071 Peoples R China Purdue Univ Dept Comp & Informat Technol W Lafayette IN 47907 USA Xidian Univ Sch Cyber Engn Xian 710071 Peoples R China

Notwithstanding the tremendous success of deep neural networks in a range of realms, previous studies have shown that these learning models are exposed to an inherent hazard called adversarial example - images to which an elaborate perturbation is maliciously added could deceive a network, which entails the study of countermeasures urgently. However, existing solutions suffer from some weaknesses, e.g., parameters are usually determined empirically in some processing-based detection methods might result in a sub-optimal effect, and the directly performed processing on images might affect the classification of benign samples, leading to increment of false positive. In this paper, we propose a novel image-DepenDent noIse reducTION (ADDITION) model based on deep learning for adversarial detection. The ADDITION model can adaptively convert the adversarial perturbation in each image to approximate Gaussian noise by injecting image-dependent additional noise, then perform noise reduction to eliminate the adversarial perturbation, and finally detect adversarial examples by examining the classification inconsistency between the input image and its denoised version. The ADDITION model is trained end-to-end on benign samples without any prior knowledge of adversarial attacks, and thus avoid time-consuming task of generating adversarial examples in practical use. We generate more than 220,000 adversarial examples based on six attack algorithms for evaluation and present state-of-the-art comparisons on three real-word datasets. Extensive experiments demonstrate that our proposed method achieves improved performance in both detection accuracy rate and false positive rate.

关键词： Perturbation methods Noise reduction Gaussian noise Adaptation models Neural networks Face recognition Robustness Adversarial attack adversarial detection deep neural network noise reduction

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：