检索结果-内蒙古大学图书馆

Real-Time Multi-Scale Barcode image Deblurring Based on Edge Feature Guidance

ELECTRONICS 2025年第7期14卷 1298-1298页

作者： Shi, Chenbo Jiang, Xin Zhang, Xiangyu Zhu, Changsheng Hu, Xiaowei Zhang, Guodong Li, Yuejia Zhang, Chun Shandong Univ Sci & Technol Coll Intelligent Equipment Tai An 271019 Peoples R China Suzhou Lamberv Intelligent Technol Dept Artificial Intelligence Suzhou 215000 Peoples R China

Barcode technology plays a crucial role in automatic identification and data acquisition systems, with extensive applications in retail, warehousing, healthcare, and industrial automation. However, barcode images often suffer from blurriness due to lighting conditions, camera quality, motion blur, and noise, adversely affecting their readability and system performance. This paper proposes a multi-scale real-time deblurring method based on edge feature guidance. Our designed multi-scale deblurring network integrates an edge feature fusion module (EFFM) to restore image edges better. Additionally, we introduce a feature filtering mechanism (FFM), which effectively suppresses noise interference by precisely filtering and enhancing critical signal features. Moreover, by incorporating wavelet reconstruction loss, the method significantly improves the restoration of details and textures. Extensive experiments on various barcode datasets demonstrate that our method significantly enhances barcode clarity and scanning accuracy, especially in noisy environments. Furthermore, our algorithm ensures robustness and real-time performance. The research results indicate that our method holds significant promise for enhancing barcode image processing, with potential applications in retail, logistics, inventory management, and industrial automation.

关键词： barcode deblurring multi-scale strategy neural networks edge guidance

来源：评论

学校读者我要写书评

暂无评论

An Efficient Sparse Blocks Inference Method for image Editing Based on Diffusion Models

An Efficient Sparse Blocks Inference Method for Image Editin...

引用

International Joint conference on neural networks (IJCNN)

作者： Yang, Zhuochao Liu, Jingjing Guo, Aiying Zhang, Jianhua Shanghai Univ Sch Microelect Shanghai Peoples R China

ISBN: (纸本)9798350359329;9798350359312

image editing methods based on diffusion models are significantly superior to traditional methods. However, due to their slow sampling speed, high computational complexity, and weak data generalization ability, these have encountered certain limitations in practical applications. This paper proposes an efficient sparse blocks inference method for diffusion models to address this issue. It also compensates and trains the sampled feature maps by reusing low-frequency information and introduces Lp-norm to replace Euclidean distance to calculate the loss function, thereby enhancing the reconstruction effect of highfrequency image features. This method takes low-frequency sparse block data inputs as constraints, using the masks of the difference and converting them into indices to achieve the reproduction of high-resolution data. Experiments on the LSUN and CelebAHQ dataset, our method improves the inference speed of DDIM by 2.68 x, PD by 1.53 x and SDEdit by 5.5 x, reduces the computational complexity of DDIM by 4.1 x, PD by 1.7 x and SDEdit by 4.8 x.

关键词： image editing Diffusion models Sparse blocks Indices Computational complexity

来源：评论

学校读者我要写书评

暂无评论

How image distortions affect inference accuracy

How image distortions affect inference accuracy

引用

International Joint conference on neural networks (IJCNN)

作者： Dvoracek, Petr Univ Ostrava Inst Res & Applicat Fuzzy Modeling Ctr Excellence IT4Innovat Ostrava Czech Republic

ISBN: (纸本)9798350359329;9798350359312

This research thoroughly analyses the effects of various image distortions on different neural network architectures, utilizing the imageNet validation dataset. The study reveals that distortions such as blur, noise, JPEG compression, and grid patterns significantly elevate error rates. Contrast change yielded inconsistent impacts, only occasionally increasing error rates, while rescaling the image drastically (when target size was more than five times smaller than original) led to substantial model performance degradation, although the images became difficult for even humans to recognize at such scales. Among the architectures tested, EfficientNet models demonstrated superior robustness, which can be attributed to their scalable input sizes. The Vision Transformer (ViT), trained on the extensive JFT-300M dataset, also showcased notable resilience to distortions. Other architectures like ConvNeXt, Swin, and ResNetD were more susceptible to the tested distortions. MobileNetV3-based models seem to provide an exceptionally good ratio between the robustness and model size, but this could also be caused by the fact that models from this family have inherently higher base error rates, which intrinsically allows for a larger margin of error.

关键词： image processing neural network inference adversarial machine learning

来源：评论

学校读者我要写书评

暂无评论

Impact of the Preprocessing Steps in Deep Learning-Based image Classifications

引用

NATIONAL ACADEMY SCIENCE LETTERS-INDIA 2024年第6期47卷 645-647页

作者： Koresh, H. James Deva KPR Inst Engn & Technol Dept Elect & Commun Engn Coimbatore India

Deep learning softwares are designed using artificial neural networks for various applications by training and testing them with an appropriate dataset. The raw image samples available in the dataset may contain noisy and unclear information due to radiation, heat and poor lighting conditions. Therefore, the researchers are trying to filter and enhance such noisy images through preprocessing steps for providing a valid feature information to the neural network layers included in the deep learning software. However, there are certain claims that roam around the researchers such as an image may lose some useful information when it is not preprocessed with an appropriate filter or enhancement technique. Hence, the work reviews the efficacy of the methodologies that are designed with and without a preprocessing step. Also, the work summarizes the common reasons and statements highlighted by the researchers for using and avoiding the preprocessing steps on designing a deep learning approach. The study is conducted to provide a clarity toward the requirement and non-requirement of preprocessing step in a deep learning software.

关键词： image enhancement Filters Feature processing Classification Computation reduction

来源：评论

学校读者我要写书评

暂无评论

image denoising based on Swin Transformer Residual Conv U-Net 27

Image denoising based on Swin Transformer Residual Conv U-Ne...

引用

27th IEEE/ACIS International conference on Software Engineering, artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD)

作者： Gan, Yong Zhou, Shaohui Chen, Haonan Wang, Yuefeng Zhengzhou Univ Technol Zhengzhou Peoples R China Zhengzhou Univ Light Ind Coll Comp Sci & Technol Zhengzhou Peoples R China

ISBN: (纸本)9798350391961;9798350391954

In the field of computer vision, image denoising remains a fundamental and challenging problem, playing a crucial role in the preprocessing of various image processing tasks. The introduction of Convolutional neural networks (CNNs) into the image denoising domain has yielded significant improvements across different levels of visual tasks. In recent years, models based on the Swin Transformer have also been applied to the image denoising field, demonstrating superior denoising performance that surpasses CNN-based methods, thus becoming advanced techniques in current image denoising research. This paper proposes a Swin-Conv module that combines the local modeling capabilities of residual convolutional layers with the non-local modeling capabilities of the Swin Transformer and integrates this module into the UNet architecture for image denoising. For the dataset used in the model training process, data augmentation techniques were employed to randomly enhance the dataset, thereby improving overall robustness. The results indicate that the proposed Swin Transformer Residual Conv U-Net model shows improvement over current advanced networks, achieving PSNR and SSIM values of 36.09 and 0.963 at sigma = 15, 33.87 and 0.915 at sigma = 25, and 28.96 and 0.810 at sigma = 50.

关键词： image Denoising Deep Learning Transformer Convolutional neural networks U-Net

来源：评论

学校读者我要写书评

暂无评论

BP neural Network Optimization Considering Computational Resource Constraints 2

BP Neural Network Optimization Considering Computational Res...

引用

2nd IEEE International conference on image processing and Computer applications, ICIPCA 2024

作者： Luo, Weihan Li, Zhiyi Peng, Fangwei School of Science Harbin Institute of Technology Shenzhen Shenzhen China Sichuan University Chengdu China East China University of Technology Nanchang China

ISBN: (纸本)9798350360240

BP neural networks have a wide range of applications and can cope with a variety of problems. In the past, many scholars have applied neural networks to various problems. However, BP neural networks require a large number of resources for computation during training, which leads to problems such as long computation time when the problem is more complex. In order to solve this problem, this paper limits the number of running iterations of the neural network by considering the condition of limited computational resources. To enhance prediction accuracy, when employing the genetic algorithm for optimizing the BP neural network, the forward propagation process of the neural network is utilized to compute the fitness of each iteration. This approach serves to enhance the optimization efficiency of the algorithm. Experiments show that the IGABP algorithm designed in this paper has a better prediction performance in all three indexes: MSE, MAE and MAPE. © 2024 IEEE.

关键词： Optimization algorithms

来源：评论

学校读者我要写书评

暂无评论

Machine learning algorithms for manufacturing quality assurance: A systematic review of performance metrics and applications

ARRAY

引用

ARRAY 2025年 26卷

作者： Kausik, Ashfakul Karim Bin Rashid, Adib Baki, Ramisha Fariha Maktum, Md Mifthahul Jannat Mil Inst Sci & Technol MIST Ind & Prod Engn Dept Dhaka Bangladesh Mil Inst Sci & Technol MIST Comp Sci & Engn Dept Dhaka Bangladesh Mil Inst Sci & Technol MIST Naval Architecture & Marine Engn Dept Dhaka Bangladesh

Adopting Machine Learning (ML) in manufacturing quality assurance (QA) has accelerated with Industry 4.0, enabling automated defect detection, predictive maintenance, and real-time process optimization. However, selecting the most effective ML model remains challenging due to performance variability, scalability constraints, and inconsistent evaluation metrics across manufacturing sectors. This systematic review analyzes over 300 peerreviewed studies over the last two decades (mostly analyzing the recent works) to evaluate the effectiveness of widely used ML algorithms-artificial neural networks (ANNs), Support Vector Machines (SVMs), Random Forests (RFs), Decision Trees (DTs), and K-Nearest Neighbors (KNN)-in QA applications. Performance metrics include accuracy, precision, speed, recall, computational efficiency, scalability, and real-time processing capabilities. Findings reveal that ANNs outperform other models in image-based defect detection, while SVMs and RFs excel in predictive maintenance and process parameter optimization. DTs provide better interpretability for process control, and KNN is effective for small-scale QA implementations. In specific case scenarios, RF models showed particular strength in handling high-dimensional sensor data in fault detection in manufacturing quality assurance operations. The study presents a comparative assessment framework, guiding algorithm selection based on industry-specific requirements and operational constraints. This review provides the latest implementation of ML in QA along with quantitative evidence on which algorithm offers the most optimization in specific industrial settings, which would help in algorithm selection in manufacturing quality assurance in future for both researchers and industrial experts. Also, it offers an overview of the major and minor algorithms based on their performance metrics.

关键词： Machine learning Quality assurance artificial intelligence artificial neural networks Predictive analytics Process optimization Decision trees Object detection KNN SVM Clustering algorithms

来源：评论

学校读者我要写书评

暂无评论

Using artificial neural networks for graffiti detection 8

Using artificial neural networks for graffiti detection

引用

8th Scientific School "Dynamics of Complex networks and their applications", DCNA 2024

作者： Savinov, Vladimir Kamyshov, Gleb Sapunov, Viktor Shusharina, Natalia Baltic Center for Neurotechnology and Artificial Intelligence Immanuel Kant Baltic Federal University Kaliningrad Russia

ISBN: (纸本)9798350390834

To solve the problem of detecting graffiti (tags), an artificial neural network with the YOLOv5 architecture was used. This architecture is known for its high operating speed and low computing power requirements, enabling the quick processing of video streams from a large number of city cameras. To account for the regional characteristics of graffiti, a dataset consisting of 1927 photographs was collected. This dataset is open and available at https://***/KGD/. The accuracy of graffiti detection on the test sample was 0.82. © 2024 IEEE.

关键词： neural networks

来源：评论

学校读者我要写书评

暂无评论

artificial neural networks with Soft Attention: Natural Language processing for Phishing Email Detection Optimized with Modified Metaheuristics 4th

Artificial Neural Networks with Soft Attention: Natural Lang...

引用

4th International conference on Advanced Network Technologies and Intelligent Computing-ANTIC-Annual

作者： Lakicevic, Bojana Spalevic, Zaklina Volas, Igor Jovanovic, Luka Zivkovic, Miodrag Zivkovic, Tamara Bacanin, Nebojsa Univ Montenegro Fac Law Cetinjski Put 2 Podgorica 81000 Montenegro Singidunum Univ Danijelova 32 Belgrade 11010 Serbia Univ Belgrade Sch Elect Engn Bulevar Kralja Aleksandra 73 Belgrade 11000 Serbia

ISBN: (纸本)9783031837890;9783031837906

This study presents a novel approach to phishing email detection, leveraging artificial neural networks (ANN) with soft attention in natural language processing (NLP) through the integration of BERT encoders. Addressing the critical need for effective aritifical intelligence (AI)-driven cybersecurity solutions, this research combines BERT's NLP capabilities with a modified crayfish optimization algorithm (COA) to fine-tune the hyperparameters of deep neural network models, enhancing classification accuracy. Experimental results show that our optimized model achieves a phishing detection accuracy of 92.5%, outperforming several high-performing optimizers. Comparative analysis demonstrates that this approach offers superior detection capabilities, underscoring its potential for real-world applications. This work advances the field by refining BERT's application with optimization algorithms and provides a valuable framework for future cybersecurity research.

关键词： ANN soft attention NLP BERT metaheuristics optimization crayfish optimization algorithm phishing detection cybersecurity

来源：评论

学校读者我要写书评

暂无评论

A Design Framework for Hardware-Efficient Logarithmic Floating-Point Multipliers

引用

IEEE TRANSACTIONS ON EMERGING TOPICS IN COMPUTING 2024年第4期12卷 991-1001页

作者： Zhang, Tingting Niu, Zijing Han, Jie Univ Alberta Dept Elect & Comp Engn Edmonton AB T6G 1H9 Canada

The symbiotic use of logarithmic approximation in floating-point (FP) multiplication can significantly reduce the hardware complexity of a multiplier. However, it is difficult for a limited number of logarithmic FP multipliers (LFPMs) to fit in a specific error-tolerant application, such as neural networks (NNs) and digital signal processing, due to their unique error characteristics. This article proposes a design framework for generating LFPMs. We consider two FP representation formats with different ranges of mantissas, the IEEE 754 Standard FP Format and the Nearest Power of Two FP Format. For both logarithm and anti-logarithm computation, the applicable regions of inputs are first evenly divided into several intervals, and then approximation methods with negative or positive errors are developed for each sub-region. By using piece-wise functions, different configurations of approximation methods throughout applicable regions are created, leading to LFPMs with various trade-offs between accuracy and hardware cost. The variety of error characteristics of LFPMs is discussed and the generic hardware implementation is illustrated. As case studies, two LFPM designs are presented and evaluated in applications of JPEG compression and NNs. They do not only increase the classification accuracy, but also achieve smaller PDPs compared to the exact FP multiplier, while being more accurate than a recent logarithmic FP design.

关键词： Hardware Standards Transform coding Training Costs artificial neural networks image coding Floating-point multiplier logarithmic multiplier neural networks JPEG compression error tolerance approximate computing approximate multiplier

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：