Computers in Industry最新文献_第4页

A hybrid mechanism and data-driven optimization method of process parameters in laser cutting 激光切割工艺参数的混合优化机制与数据驱动方法

IF 9.1 1区计算机科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers in Industry

Pub Date : 2025-10-07 DOI: 10.1016/j.compind.2025.104394

Shuai Ma , Zhuyun Chen , Zehao Li , Jiewu Leng , Huitao Liu , Yixian Du , Xiaoji Zhang , Qiang Liu

Laser cutting quality is directly influenced by process parameters, which govern the formation of burrs and the extent of the heat-affected zone. Consequently, selecting and optimizing these parameters is crucial for achieving high-quality laser cutting results. Machine learning techniques have proven effective in process parameter optimization by establishing surrogate models that link process parameters with quality indicators. However, these models often overlook critical temperature field information generated during laser cutting, which provides valuable mechanistic insights. To overcome this limitation, a hybrid mechanism and data-driven optimization method is proposed. First, a laser cutting experimental platform is developed, and the full-factorial design with five factors at three levels is employed for data collection. Detailed laser-cutting physical models are then established to simulate key temperature field information, compensating for the scarcity of such data in real-world scenarios. Subsequently, a novel physics-informed neural network is designed with dual input branches to handle low-dimensional process parameters and high-dimensional temperature field data. Besides, the physics-informed neural network includes a focused fusion layer with an attention mechanism to selectively integrate the most relevant mechanistic features with process parameters. To further optimize the trained physics-informed neural network model, a clustering-assisted multi-objective evolutionary algorithm is developed, which leverages the clustering strategy to select and retrieve historical mechanistic data that best match candidate process parameters, ensuring valid surrogate model inputs and improving optimization efficiency. Experimental validation demonstrates that the proposed hybrid method significantly outperforms conventional machine learning approaches, delivering superior accuracy and reliability in laser cutting process parameter optimization.

激光切割质量直接受到工艺参数的影响，工艺参数决定着毛刺的形成和热影响区的范围。因此，选择和优化这些参数对于获得高质量的激光切割结果至关重要。通过建立将工艺参数与质量指标联系起来的代理模型，机器学习技术已被证明在工艺参数优化方面是有效的。然而，这些模型往往忽略了激光切割过程中产生的关键温度场信息，这些信息提供了有价值的机械见解。为了克服这一局限性，提出了一种混合机制和数据驱动优化方法。首先，搭建了激光切割实验平台，采用五因子三水平全因子设计进行数据采集。然后建立详细的激光切割物理模型来模拟关键的温度场信息，弥补了现实场景中此类数据的缺乏。随后，设计了一种具有双输入分支的物理信息神经网络来处理低维工艺参数和高维温度场数据。此外，物理信息神经网络还包括一个聚焦融合层，该融合层具有一个关注机制，可以选择性地将最相关的机械特征与过程参数相结合。为了进一步优化训练后的物理信息神经网络模型，开发了一种聚类辅助多目标进化算法，该算法利用聚类策略选择和检索与候选过程参数最匹配的历史机制数据，确保有效的代理模型输入，提高优化效率。实验验证表明，所提出的混合方法明显优于传统的机器学习方法，在激光切割工艺参数优化方面具有更高的精度和可靠性。

{"title":"A hybrid mechanism and data-driven optimization method of process parameters in laser cutting","authors":"Shuai Ma , Zhuyun Chen , Zehao Li , Jiewu Leng , Huitao Liu , Yixian Du , Xiaoji Zhang , Qiang Liu","doi":"10.1016/j.compind.2025.104394","DOIUrl":"10.1016/j.compind.2025.104394","url":null,"abstract":"<div><div>Laser cutting quality is directly influenced by process parameters, which govern the formation of burrs and the extent of the heat-affected zone. Consequently, selecting and optimizing these parameters is crucial for achieving high-quality laser cutting results. Machine learning techniques have proven effective in process parameter optimization by establishing surrogate models that link process parameters with quality indicators. However, these models often overlook critical temperature field information generated during laser cutting, which provides valuable mechanistic insights. To overcome this limitation, a hybrid mechanism and data-driven optimization method is proposed. First, a laser cutting experimental platform is developed, and the full-factorial design with five factors at three levels is employed for data collection. Detailed laser-cutting physical models are then established to simulate key temperature field information, compensating for the scarcity of such data in real-world scenarios. Subsequently, a novel physics-informed neural network is designed with dual input branches to handle low-dimensional process parameters and high-dimensional temperature field data. Besides, the physics-informed neural network includes a focused fusion layer with an attention mechanism to selectively integrate the most relevant mechanistic features with process parameters. To further optimize the trained physics-informed neural network model, a clustering-assisted multi-objective evolutionary algorithm is developed, which leverages the clustering strategy to select and retrieve historical mechanistic data that best match candidate process parameters, ensuring valid surrogate model inputs and improving optimization efficiency. Experimental validation demonstrates that the proposed hybrid method significantly outperforms conventional machine learning approaches, delivering superior accuracy and reliability in laser cutting process parameter optimization.</div></div>","PeriodicalId":55219,"journal":{"name":"Computers in Industry","volume":"173 ","pages":"Article 104394"},"PeriodicalIF":9.1,"publicationDate":"2025-10-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145261933","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Integrated label correction for e-commerce data: Boosting accuracy with baseline attention and enhanced Bayesian updating 电子商务数据的集成标签校正：通过基线关注和增强贝叶斯更新来提高准确性

IF 9.1 1区计算机科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers in Industry

Pub Date : 2025-10-07 DOI: 10.1016/j.compind.2025.104392

Miao Zhang , Jingyuan Fu , Yuli Zhang , Qinghui Ma , Jinghong Chen

The rapid growth of the e-commerce industry has led to an explosion of product data, which e-commerce platforms increasingly leverage to enhance operational decision-making. However, the presence of label noise in product data poses a significant challenge, as mislabeled data can degrade decision-making performance, negatively impacting both platform efficiency and user experience. To tackle this challenge, this paper proposes an integrated label correction (ILC) method, featuring two key innovations: a baseline attention (BA) mechanism for noise detection and an enhanced Bayesian updating (EBU) strategy for label prediction. The proposed BA mechanism utilizes a dynamic attention scaling approach with learnable parameters to measure the similarity between samples and their labeled classes. The EBU strategy integrates sample features and noisy observations into a unified probabilistic framework, jointly optimizing the classifier and the label correction process. A novel adaptive noise transition learning approach is further introduced to refine the noise transition matrix. Extensive experiments on a real-world JD text dataset, a benchmark image dataset, and an automotive user review dataset validate the effectiveness of the proposed ILC method. It consistently outperforms state-of-the-art approaches on average, achieving a 2.78% improvement on the JD dataset, and gains of 0.80% on the image dataset and 1.73% on the review dataset across different noise types. These improvements demonstrate that the ILC method not only improves classification performance but also offers scalable solutions to reduce mislabeling errors in real-world e-commerce platforms, thereby enhancing recommendation systems, inventory management, and user satisfaction.

电子商务行业的快速增长导致了产品数据的爆炸式增长，电子商务平台越来越多地利用这些数据来增强运营决策。然而，产品数据中标签噪声的存在带来了重大挑战，因为错误标记的数据会降低决策性能，对平台效率和用户体验产生负面影响。为了应对这一挑战，本文提出了一种集成标签校正（ILC）方法，该方法具有两个关键创新：用于噪声检测的基线注意（BA）机制和用于标签预测的增强贝叶斯更新（EBU）策略。所提出的BA机制利用具有可学习参数的动态注意力缩放方法来度量样本与其标记类之间的相似性。EBU策略将样本特征和噪声观测值整合到统一的概率框架中，共同优化分类器和标签校正过程。进一步介绍了一种新的自适应噪声转移学习方法来细化噪声转移矩阵。在现实世界JD文本数据集、基准图像数据集和汽车用户评论数据集上进行的大量实验验证了所提出的ILC方法的有效性。平均而言，它始终优于最先进的方法，在JD数据集上实现了2.78%的改进，在不同噪声类型的图像数据集上实现了0.80%的增益，在评论数据集上实现了1.73%的增益。这些改进表明，ILC方法不仅提高了分类性能，而且提供了可扩展的解决方案，以减少现实世界电子商务平台中的错误标记，从而增强了推荐系统、库存管理和用户满意度。

{"title":"Integrated label correction for e-commerce data: Boosting accuracy with baseline attention and enhanced Bayesian updating","authors":"Miao Zhang , Jingyuan Fu , Yuli Zhang , Qinghui Ma , Jinghong Chen","doi":"10.1016/j.compind.2025.104392","DOIUrl":"10.1016/j.compind.2025.104392","url":null,"abstract":"<div><div>The rapid growth of the e-commerce industry has led to an explosion of product data, which e-commerce platforms increasingly leverage to enhance operational decision-making. However, the presence of label noise in product data poses a significant challenge, as mislabeled data can degrade decision-making performance, negatively impacting both platform efficiency and user experience. To tackle this challenge, this paper proposes an integrated label correction (ILC) method, featuring two key innovations: a baseline attention (BA) mechanism for noise detection and an enhanced Bayesian updating (EBU) strategy for label prediction. The proposed BA mechanism utilizes a dynamic attention scaling approach with learnable parameters to measure the similarity between samples and their labeled classes. The EBU strategy integrates sample features and noisy observations into a unified probabilistic framework, jointly optimizing the classifier and the label correction process. A novel adaptive noise transition learning approach is further introduced to refine the noise transition matrix. Extensive experiments on a real-world JD text dataset, a benchmark image dataset, and an automotive user review dataset validate the effectiveness of the proposed ILC method. It consistently outperforms state-of-the-art approaches on average, achieving a 2.78% improvement on the JD dataset, and gains of 0.80% on the image dataset and 1.73% on the review dataset across different noise types. These improvements demonstrate that the ILC method not only improves classification performance but also offers scalable solutions to reduce mislabeling errors in real-world e-commerce platforms, thereby enhancing recommendation systems, inventory management, and user satisfaction.</div></div>","PeriodicalId":55219,"journal":{"name":"Computers in Industry","volume":"173 ","pages":"Article 104392"},"PeriodicalIF":9.1,"publicationDate":"2025-10-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145261916","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Real-time fire temperature field reconstruction using multi-source acoustic wave data and a hybrid deep learning approach 基于多源声波数据和混合深度学习方法的实时火灾温度场重建

IF 9.1 1区计算机科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers in Industry

Pub Date : 2025-10-06 DOI: 10.1016/j.compind.2025.104389

Hengjie Qin , Pengyu Qu , Cheng Cheng , Haowei Yao , Zhen Lou , Zihe Gao , Donghao Li , Xiaoge Wei

Accurate reconstruction of temperature distributions in complex fire scenarios is essential for effective fire monitoring, early warning, and firefighting decision-making. Traditional methods often face challenges in both accuracy and computational efficiency due to the highly nonlinear and dynamic nature of fire environments. To address this issue, a novel reconstruction framework driven by multi-source acoustic wave data is presented. This approach integrates an Adaptive Weighted Hybrid Convolution–Dynamic Residual Attention-Aware Fusion Network (AWHC-DRAAFN) with an Integrated K-Nearest Neighbor (IKNN) model. The AWHC-DRAAFN facilitates efficient extraction and fusion of multi-scale features by combining various convolution operations with adaptive weighting mechanisms, thereby enhancing the network’s capacity to capture complex nonlinear relationships between acoustic wave propagation and temperature distribution. Meanwhile, the IKNN model transforms discrete temperature data into a continuous field through a locally weighted K-nearest neighbor interpolation strategy. Experimental results demonstrate that the proposed method achieves high prediction accuracy (MAE

<

5.3%, MSE

<

0.7%, RMSE

<

8.3%) and high computational efficiency (reconstruction time

<

0.54s), highlighting its potential as a reliable solution for real-time reconstruction of fire temperature fields.

准确重建复杂火灾情景下的温度分布对有效的火灾监测、预警和消防决策至关重要。由于火灾环境的高度非线性和动态性，传统方法在精度和计算效率方面都面临挑战。为了解决这一问题，提出了一种基于多源声波数据驱动的重构框架。该方法将自适应加权混合卷积-动态剩余注意感知融合网络（AWHC-DRAAFN）与集成k -最近邻（IKNN）模型相结合。AWHC-DRAAFN通过将各种卷积操作与自适应加权机制相结合，促进了多尺度特征的高效提取和融合，从而增强了网络捕捉声波传播与温度分布之间复杂非线性关系的能力。同时，IKNN模型通过局部加权k近邻插值策略将离散温度数据转换为连续场。实验结果表明，该方法具有较高的预测精度（MAE < 5.3%, MSE < 0.7%, RMSE < 8.3%）和较高的计算效率（重建时间<； 0.54s），具有作为火灾温度场实时重建可靠解决方案的潜力。

{"title":"Real-time fire temperature field reconstruction using multi-source acoustic wave data and a hybrid deep learning approach","authors":"Hengjie Qin , Pengyu Qu , Cheng Cheng , Haowei Yao , Zhen Lou , Zihe Gao , Donghao Li , Xiaoge Wei","doi":"10.1016/j.compind.2025.104389","DOIUrl":"10.1016/j.compind.2025.104389","url":null,"abstract":"<div><div>Accurate reconstruction of temperature distributions in complex fire scenarios is essential for effective fire monitoring, early warning, and firefighting decision-making. Traditional methods often face challenges in both accuracy and computational efficiency due to the highly nonlinear and dynamic nature of fire environments. To address this issue, a novel reconstruction framework driven by multi-source acoustic wave data is presented. This approach integrates an Adaptive Weighted Hybrid Convolution–Dynamic Residual Attention-Aware Fusion Network (AWHC-DRAAFN) with an Integrated K-Nearest Neighbor (IKNN) model. The AWHC-DRAAFN facilitates efficient extraction and fusion of multi-scale features by combining various convolution operations with adaptive weighting mechanisms, thereby enhancing the network’s capacity to capture complex nonlinear relationships between acoustic wave propagation and temperature distribution. Meanwhile, the IKNN model transforms discrete temperature data into a continuous field through a locally weighted K-nearest neighbor interpolation strategy. Experimental results demonstrate that the proposed method achieves high prediction accuracy (MAE <span><math><mo><</mo></math></span> 5.3%, MSE <span><math><mo><</mo></math></span> 0.7%, RMSE <span><math><mo><</mo></math></span> 8.3%) and high computational efficiency (reconstruction time <span><math><mo><</mo></math></span> 0.54s), highlighting its potential as a reliable solution for real-time reconstruction of fire temperature fields.</div></div>","PeriodicalId":55219,"journal":{"name":"Computers in Industry","volume":"173 ","pages":"Article 104389"},"PeriodicalIF":9.1,"publicationDate":"2025-10-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145261917","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A fuzzy feature integration-enhanced network for surface defect detection of no-service rails 基于模糊特征集成的停运轨道表面缺陷检测网络

IF 9.1 1区计算机科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers in Industry

Pub Date : 2025-10-06 DOI: 10.1016/j.compind.2025.104382

Liming Huang , Aojun Gong

Surface defect detection on no-service rails is crucial for ensuring the safety and reliability of industrial manufacturing processes. Vision-based detection methods have seen significant progress in this domain. However, this task still faces significant challenges from the following aspects: (1) The great amounts of noise in rail surface defect images; (2) Great similarity between the foreground and background of defect images. Fuzzy logic, a significant technique in the automatic control, is effective for handling uncertain and imprecise features to improve the stability of detection systems. In this paper, we propose a novel approach by integrating fuzzy logic into deep neural networks for rail surface defect detection. First, we introduce a fuzzy logic-based feature enhancement module, where a Gaussian-based fuzzy strategy is utilized to improve feature representation. Next, we devise a fuzzy logic-based loss function tailored for fuzzy features which ensures that the fuzzy representation of features is beneficial for defect segmentation. Experimental validation using both RGB and RGB-depth images demonstrates the competitive and promising performance of our proposed approach compared to state-of-the-art models. Extensive validation on strip steel surface defect detection and salient object detection in natural images further confirms the effectiveness of our model and the application of fuzzy logic. Furthermore, this paper discusses the research significance and potential applications of the proposed methodology across various domains.

无服务轨道表面缺陷检测对于保证工业制造过程的安全性和可靠性至关重要。基于视觉的检测方法在这一领域取得了重大进展。然而，这项任务仍然面临着以下几个方面的重大挑战：(1)钢轨表面缺陷图像中存在大量的噪声；(2)缺陷图像的前景与背景相似性大。模糊逻辑是自动控制中的一项重要技术，它可以有效地处理不确定和不精确的特征，从而提高检测系统的稳定性。本文提出了一种将模糊逻辑与深度神经网络相结合的钢轨表面缺陷检测方法。首先，我们引入了一个基于模糊逻辑的特征增强模块，其中利用基于高斯的模糊策略来改进特征表示。其次，我们针对模糊特征设计了一个基于模糊逻辑的损失函数，以确保特征的模糊表示有利于缺陷分割。使用RGB和RGB深度图像的实验验证表明，与最先进的模型相比，我们提出的方法具有竞争力和有前景的性能。对带钢表面缺陷检测和自然图像中显著目标检测的大量验证进一步证实了该模型的有效性以及模糊逻辑的应用。此外，本文还讨论了该方法在各个领域的研究意义和潜在应用。

{"title":"A fuzzy feature integration-enhanced network for surface defect detection of no-service rails","authors":"Liming Huang , Aojun Gong","doi":"10.1016/j.compind.2025.104382","DOIUrl":"10.1016/j.compind.2025.104382","url":null,"abstract":"<div><div>Surface defect detection on no-service rails is crucial for ensuring the safety and reliability of industrial manufacturing processes. Vision-based detection methods have seen significant progress in this domain. However, this task still faces significant challenges from the following aspects: (1) The great amounts of noise in rail surface defect images; (2) Great similarity between the foreground and background of defect images. Fuzzy logic, a significant technique in the automatic control, is effective for handling uncertain and imprecise features to improve the stability of detection systems. In this paper, we propose a novel approach by integrating fuzzy logic into deep neural networks for rail surface defect detection. First, we introduce a fuzzy logic-based feature enhancement module, where a Gaussian-based fuzzy strategy is utilized to improve feature representation. Next, we devise a fuzzy logic-based loss function tailored for fuzzy features which ensures that the fuzzy representation of features is beneficial for defect segmentation. Experimental validation using both RGB and RGB-depth images demonstrates the competitive and promising performance of our proposed approach compared to state-of-the-art models. Extensive validation on strip steel surface defect detection and salient object detection in natural images further confirms the effectiveness of our model and the application of fuzzy logic. Furthermore, this paper discusses the research significance and potential applications of the proposed methodology across various domains.</div></div>","PeriodicalId":55219,"journal":{"name":"Computers in Industry","volume":"173 ","pages":"Article 104382"},"PeriodicalIF":9.1,"publicationDate":"2025-10-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145261941","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

SymmFlow: Unsupervised anomaly detection via symmetric normalizing flow SymmFlow：通过对称规范化流进行无监督异常检测

IF 9.1 1区计算机科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers in Industry

Pub Date : 2025-10-04 DOI: 10.1016/j.compind.2025.104393

Zeyu Zhang , Danqing Kang , Biaohua Ye , Jianhuang Lai

Anomaly detection in industrial imaging has attracted significant research interest due to its critical applications. Recent advancements have demonstrated the potential of normalizing flows for unsupervised anomaly detection. However, conventional approaches often face challenges with the degeneracy of transformed distributions, especially in scenarios where anomaly samples are both scarce and subtle. To overcome these challenges, we propose a symmetrically structured normalizing flow model called SymmFlow. SymmFlow addresses the degeneracy of transformed distributions by maintaining the positive definiteness of the covariance matrix within multivariate Gaussian distributions. A novel two-stage training strategy is also proposed to stabilize training initially with regularization and subsequently reinforce the model’s robustness through symmetrical design. Extensive experiments on MVTec, VisA, and BTAD datasets demonstrate that the proposed SymmFlow outperforms existing methods, delivering superior detection accuracy both at the image and pixel levels. The source code is available at: https://github.com/Ace-blue/SymmFlow.

工业成像中的异常检测由于其重要的应用而引起了人们极大的研究兴趣。最近的进展已经证明了在无监督异常检测中规范化流程的潜力。然而，传统的方法常常面临变换分布退化的挑战，特别是在异常样本稀缺和微妙的情况下。为了克服这些挑战，我们提出了一个对称结构的规范化流模型，称为SymmFlow。SymmFlow通过在多元高斯分布中保持协方差矩阵的正确定性来解决变换分布的退化问题。提出了一种新的两阶段训练策略，通过正则化初始稳定训练，然后通过对称设计增强模型的鲁棒性。在MVTec、VisA和BTAD数据集上进行的大量实验表明，所提出的SymmFlow优于现有方法，在图像和像素级别上都提供了卓越的检测精度。源代码可从https://github.com/Ace-blue/SymmFlow获得。

{"title":"SymmFlow: Unsupervised anomaly detection via symmetric normalizing flow","authors":"Zeyu Zhang , Danqing Kang , Biaohua Ye , Jianhuang Lai","doi":"10.1016/j.compind.2025.104393","DOIUrl":"10.1016/j.compind.2025.104393","url":null,"abstract":"<div><div>Anomaly detection in industrial imaging has attracted significant research interest due to its critical applications. Recent advancements have demonstrated the potential of normalizing flows for unsupervised anomaly detection. However, conventional approaches often face challenges with the degeneracy of transformed distributions, especially in scenarios where anomaly samples are both scarce and subtle. To overcome these challenges, we propose a symmetrically structured normalizing flow model called SymmFlow. SymmFlow addresses the degeneracy of transformed distributions by maintaining the positive definiteness of the covariance matrix within multivariate Gaussian distributions. A novel two-stage training strategy is also proposed to stabilize training initially with regularization and subsequently reinforce the model’s robustness through symmetrical design. Extensive experiments on MVTec, VisA, and BTAD datasets demonstrate that the proposed SymmFlow outperforms existing methods, delivering superior detection accuracy both at the image and pixel levels. The source code is available at: <span><span>https://github.com/Ace-blue/SymmFlow</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":55219,"journal":{"name":"Computers in Industry","volume":"173 ","pages":"Article 104393"},"PeriodicalIF":9.1,"publicationDate":"2025-10-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145221261","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

CASIA-Net: An indoor work site smoking detection framework CASIA-Net：一个室内工作场所吸烟检测框架

IF 9.1 1区计算机科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers in Industry

Pub Date : 2025-10-04 DOI: 10.1016/j.compind.2025.104383

Meng Wang , Mei Li

Detecting smoking behaviors in indoor work sites poses significant challenges due to the small scale of targets, poor visibility, and cluttered environments. These factors significantly heighten the risk of fire hazards. We propose the Context-Aware Small-Item Attention Net (CASIA-Net), a novel detection framework to address these issues. CASIA-Net incorporates a Deformable Feature Extraction (DFE) module to tackle the non-salient characteristics of smoking targets. It adaptively adjusts the convolution kernel size according to the target scale. An Adaptive Feature Attention (AFA) module is proposed to extract small objects. It enhances the attention to critical features of smoking from complex indoor work site backgrounds. To address the issue of attention drift in complex environments, a Smoker Feature Integration module is proposed to integrate the features extracted by DFE and AFA. Additionally, a dedicated dataset for indoor work site smoking detection is constructed. Experimental results demonstrate that the proposed model achieves an mAP50 of 0.917 on the dataset with a compact weight of 6MB. The proposed model demonstrates outstanding accuracy, robustness, and lightweight design. It is highly suitable for deployment in complex indoor work sites and industrial applications.

由于目标范围小、能见度低和环境杂乱，在室内工作场所检测吸烟行为面临着重大挑战。这些因素大大增加了发生火灾的风险。我们提出了上下文感知小项目注意网络（CASIA-Net），这是一个新的检测框架来解决这些问题。CASIA-Net结合了一个可变形特征提取（DFE）模块来处理吸烟目标的非显著特征。它根据目标尺度自适应调整卷积核的大小。提出了一种自适应特征注意（AFA）模块来提取小目标。它增强了人们对复杂的室内工作场所背景中吸烟的关键特征的关注。为了解决复杂环境下的注意力漂移问题，提出了烟民特征集成模块，将DFE和AFA提取的特征进行集成。此外，构建了室内工作场所吸烟检测专用数据集。实验结果表明，该模型在精简权值为6MB的数据集上的mAP50值为0.917。所提出的模型具有出色的准确性、鲁棒性和轻量级设计。它非常适合在复杂的室内工作场所和工业应用中部署。

{"title":"CASIA-Net: An indoor work site smoking detection framework","authors":"Meng Wang , Mei Li","doi":"10.1016/j.compind.2025.104383","DOIUrl":"10.1016/j.compind.2025.104383","url":null,"abstract":"<div><div>Detecting smoking behaviors in indoor work sites poses significant challenges due to the small scale of targets, poor visibility, and cluttered environments. These factors significantly heighten the risk of fire hazards. We propose the Context-Aware Small-Item Attention Net (CASIA-Net), a novel detection framework to address these issues. CASIA-Net incorporates a Deformable Feature Extraction (DFE) module to tackle the non-salient characteristics of smoking targets. It adaptively adjusts the convolution kernel size according to the target scale. An Adaptive Feature Attention (AFA) module is proposed to extract small objects. It enhances the attention to critical features of smoking from complex indoor work site backgrounds. To address the issue of attention drift in complex environments, a Smoker Feature Integration module is proposed to integrate the features extracted by DFE and AFA. Additionally, a dedicated dataset for indoor work site smoking detection is constructed. Experimental results demonstrate that the proposed model achieves an mAP50 of 0.917 on the dataset with a compact weight of 6MB. The proposed model demonstrates outstanding accuracy, robustness, and lightweight design. It is highly suitable for deployment in complex indoor work sites and industrial applications.</div></div>","PeriodicalId":55219,"journal":{"name":"Computers in Industry","volume":"173 ","pages":"Article 104383"},"PeriodicalIF":9.1,"publicationDate":"2025-10-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145221260","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Securing additive manufacturing with blockchain-based cryptographic anchoring and dual-lock integrity auditing 通过基于区块链的加密锚定和双锁完整性审计来保护增材制造

IF 9.1 1区计算机科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers in Industry

Pub Date : 2025-10-04 DOI: 10.1016/j.compind.2025.104395

Mahender Kumar, Gregory Epiphaniou, Carsten Maple

The Additive Manufacturing (AM) industry is a multibillion-dollar sector offering numerous benefits, such as customisation and efficiency. However, the rise in cyber–physical attacks significantly threatens its growth and security. These attacks target the integrity of digital design files, intellectual property (IP), and interconnected AM systems’ security. To address these security challenges, we propose Securing AM with Blockchain-based Cryptographic Anchoring and Dual-lock Integrity Auditing (SAM-BCADA), a system that ensures the security, integrity, and authenticity of 3D objects within the AM supply chain. SAM-BCADA utilises a dual-lock approach to connect the physical attributes of a product with its digital counterpart, improving traceability and security. By leveraging private permissioned blockchain technology, the system enables robust tracking, authentication, and verification of a product within the AM supply chain. Additionally, SAM-BCADA has developed a distributed off-chain storage for G-code data files that includes an advanced integrity auditing system using homomorphic verifiable tags. This allows for secure and efficient verification of data integrity without compromising confidentiality. This integrated approach addresses critical vulnerabilities in the AM process, providing a reliable framework for safeguarding digital and physical assets. The security analysis shows that SAM-BCADA is secure against product counterfeiting and IP theft and outperforms state-of-the-art schemes. Performance analyses have revealed that SAM-BCADA is computationally and communication-efficient and scalable in a distributed AM environment.

增材制造（AM）行业是一个价值数十亿美元的行业，提供了许多好处，如定制和效率。然而，网络物理攻击的增加严重威胁着其增长和安全。这些攻击的目标是数字设计文件的完整性、知识产权（IP）和互联AM系统的安全性。为了应对这些安全挑战，我们提出使用基于区块链的加密锚定和双锁完整性审计（SAM-BCADA）来保护AM，这是一个确保AM供应链中3D对象的安全性、完整性和真实性的系统。SAM-BCADA利用双锁方法将产品的物理属性与其数字对应物连接起来，提高了可追溯性和安全性。通过利用私人许可的区块链技术，该系统可以对AM供应链中的产品进行可靠的跟踪、认证和验证。此外，SAM-BCADA还为g代码数据文件开发了分布式链下存储，其中包括使用同态可验证标签的高级完整性审计系统。这允许在不损害机密性的情况下安全有效地验证数据完整性。这种综合方法解决了增材制造过程中的关键漏洞，为保护数字和物理资产提供了可靠的框架。安全分析表明，SAM-BCADA是安全的，防止产品假冒和知识产权盗窃，并优于最先进的方案。性能分析表明，SAM-BCADA在分布式AM环境中具有计算和通信效率以及可扩展性。

{"title":"Securing additive manufacturing with blockchain-based cryptographic anchoring and dual-lock integrity auditing","authors":"Mahender Kumar, Gregory Epiphaniou, Carsten Maple","doi":"10.1016/j.compind.2025.104395","DOIUrl":"10.1016/j.compind.2025.104395","url":null,"abstract":"<div><div>The Additive Manufacturing (AM) industry is a multibillion-dollar sector offering numerous benefits, such as customisation and efficiency. However, the rise in cyber–physical attacks significantly threatens its growth and security. These attacks target the integrity of digital design files, intellectual property (IP), and interconnected AM systems’ security. To address these security challenges, we propose Securing AM with Blockchain-based Cryptographic Anchoring and Dual-lock Integrity Auditing (SAM-BCADA), a system that ensures the security, integrity, and authenticity of 3D objects within the AM supply chain. SAM-BCADA utilises a dual-lock approach to connect the physical attributes of a product with its digital counterpart, improving traceability and security. By leveraging private permissioned blockchain technology, the system enables robust tracking, authentication, and verification of a product within the AM supply chain. Additionally, SAM-BCADA has developed a distributed off-chain storage for G-code data files that includes an advanced integrity auditing system using homomorphic verifiable tags. This allows for secure and efficient verification of data integrity without compromising confidentiality. This integrated approach addresses critical vulnerabilities in the AM process, providing a reliable framework for safeguarding digital and physical assets. The security analysis shows that SAM-BCADA is secure against product counterfeiting and IP theft and outperforms state-of-the-art schemes. Performance analyses have revealed that SAM-BCADA is computationally and communication-efficient and scalable in a distributed AM environment.</div></div>","PeriodicalId":55219,"journal":{"name":"Computers in Industry","volume":"173 ","pages":"Article 104395"},"PeriodicalIF":9.1,"publicationDate":"2025-10-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145228758","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A lightweight metal surface defect detection network with hierarchical multi-branch feature extraction and group decision attention 基于分层多分支特征提取和群体决策关注的轻量化金属表面缺陷检测网络

IF 9.1 1区计算机科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers in Industry

Pub Date : 2025-10-03 DOI: 10.1016/j.compind.2025.104390

Meng Zhang, Yanzhu Hu, Lisha Luo, Binbin Xu, Song Wang, Yingjian Wang

Metal materials are widely used in aerospace, bridge construction, and other critical applications. Surface defects such as cracks and scratches can severely undermine structural integrity and material performance, making defect detection on metal surfaces an essential industrial task. Unlike general object detection, metal surface defect detection faces unique challenges including significant scale diversity, pronounced feature ambiguity, severe data imbalance, and stringent computational resource constraints in industrial environments. To address these challenges, this study introduces LHMB-Net, a novel detection algorithm built around four key components including a hierarchical multi-branch feature extraction (HMBFE) module, a partial group decision attention (PGDA) mechanism, an HMB-head detection head, and a BoundaryIoU loss function. The lightweight hierarchical multi-scale architecture of the HMBFE module captures defect features across scales, mitigating scale diversity and feature ambiguity. The PGDA mechanism applies adaptive weighting and group decision strategies to emphasize critical features and substantially alleviate the impact of data imbalance. The HMB-Head replaces conventional convolutional structures with the HMB module to reduce model complexity while enhancing feature representation. Finally, the BoundaryIoU loss introduces boundary point distance constraints for precise localization across scales. Experimental results demonstrate that LHMB-Net outperforms current state-of-the-art methods in both detection accuracy and computational efficiency, highlighting its strong potential for practical industrial deployment.

金属材料广泛应用于航空航天、桥梁建设和其他关键应用领域。裂纹和划痕等表面缺陷会严重破坏结构完整性和材料性能，使得金属表面缺陷检测成为一项重要的工业任务。与一般的物体检测不同，金属表面缺陷检测面临着独特的挑战，包括工业环境中显著的尺度多样性、明显的特征模糊性、严重的数据不平衡以及严格的计算资源限制。为了解决这些挑战，本研究引入了LHMB-Net，这是一种围绕四个关键组件构建的新型检测算法，包括分层多分支特征提取（HMBFE）模块、部分群体决策注意（PGDA）机制、HMB-head检测头和BoundaryIoU损失函数。HMBFE模块的轻量级分层多尺度体系结构捕获了跨尺度的缺陷特征，减轻了尺度多样性和特征模糊性。PGDA机制采用自适应加权和群体决策策略，突出关键特征，大大减轻数据不平衡的影响。HMB- head用HMB模块取代了传统的卷积结构，降低了模型的复杂度，同时增强了特征表示。最后，borderyiou损失引入边界点距离约束，用于跨尺度精确定位。实验结果表明，LHMB-Net在检测精度和计算效率方面都优于当前最先进的方法，突出了其在实际工业部署中的强大潜力。

{"title":"A lightweight metal surface defect detection network with hierarchical multi-branch feature extraction and group decision attention","authors":"Meng Zhang, Yanzhu Hu, Lisha Luo, Binbin Xu, Song Wang, Yingjian Wang","doi":"10.1016/j.compind.2025.104390","DOIUrl":"10.1016/j.compind.2025.104390","url":null,"abstract":"<div><div>Metal materials are widely used in aerospace, bridge construction, and other critical applications. Surface defects such as cracks and scratches can severely undermine structural integrity and material performance, making defect detection on metal surfaces an essential industrial task. Unlike general object detection, metal surface defect detection faces unique challenges including significant scale diversity, pronounced feature ambiguity, severe data imbalance, and stringent computational resource constraints in industrial environments. To address these challenges, this study introduces LHMB-Net, a novel detection algorithm built around four key components including a hierarchical multi-branch feature extraction (HMBFE) module, a partial group decision attention (PGDA) mechanism, an HMB-head detection head, and a BoundaryIoU loss function. The lightweight hierarchical multi-scale architecture of the HMBFE module captures defect features across scales, mitigating scale diversity and feature ambiguity. The PGDA mechanism applies adaptive weighting and group decision strategies to emphasize critical features and substantially alleviate the impact of data imbalance. The HMB-Head replaces conventional convolutional structures with the HMB module to reduce model complexity while enhancing feature representation. Finally, the BoundaryIoU loss introduces boundary point distance constraints for precise localization across scales. Experimental results demonstrate that LHMB-Net outperforms current state-of-the-art methods in both detection accuracy and computational efficiency, highlighting its strong potential for practical industrial deployment.</div></div>","PeriodicalId":55219,"journal":{"name":"Computers in Industry","volume":"173 ","pages":"Article 104390"},"PeriodicalIF":9.1,"publicationDate":"2025-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145221259","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Operation time and rack stability optimisation in tier-to-tier shuttle-based storage and retrieval systems with multiple retrieval locations 具有多个检索位置的层对层梭式存储和检索系统的操作时间和机架稳定性优化

IF 9.1 1区计算机科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers in Industry

Pub Date : 2025-10-03 DOI: 10.1016/j.compind.2025.104385

Zhun Xu , Liyun Xu , Huan Shao , Timo Lehmann , Andrea Matta

Tier-to-tier shuttle-based storage and retrieval systems (t-SBS/RSs) are increasingly recognised in the industry for their enhanced flexibility, reduced costs, and improved shuttle utilisation. However, the ineffectiveness of task scheduling (TSc) and retrieval location assignment (LA) schemes diminishes retrieval efficiency and compromises rack stability, thereby elevating retrieval costs and jeopardising system operational safety. Moreover, a potential conflict has been identified between total operation time and rack stability during retrieval operations. Consequently, this study proposes a joint optimisation approach for the batch retrieval TSc and LA challenges in t-SBS/RS, accommodating multiple location distributions for identical cargo types. This issue is modelled as a multi-objective optimisation problem aimed at minimising total operation time while maximising rack stability. To address this, a non-dominated sorting genetic algorithm II (NSGA-II) enhanced by a prior knowledge-based initialisation strategy was proposed. Employing a real case study, several benchmark scenarios were established to assess the effectiveness of the proposed algorithm against five other algorithms, highlighting the superiority of NSGA-II. Additionally, the effectiveness of this method is validated by comparison with classical scheduling policies. The derived Pareto fronts verify the inherent conflict between the two objectives, suggesting that a trade-off is necessary. Finally, managerial insights are offered to highlight the practical application of these findings for warehouse managers.

基于层对层航天飞机的存储和检索系统（t-SBS/RSs）因其增强的灵活性、降低的成本和提高的航天飞机利用率而越来越受到业界的认可。然而，任务调度（TSc）和检索位置分配（LA）方案的有效性降低了检索效率，影响了机架的稳定性，从而增加了检索成本，危及系统的运行安全。此外，在检索过程中，总操作时间和机架稳定性之间存在潜在的冲突。因此，本研究提出了一种联合优化方法，用于t-SBS/RS中批量检索TSc和LA挑战，以适应相同货物类型的多个位置分布。该问题被建模为一个多目标优化问题，旨在最小化总操作时间，同时最大化机架稳定性。为了解决这一问题，提出了一种基于先验知识初始化策略的非支配排序遗传算法II （NSGA-II）。通过实际案例研究，建立了几个基准场景来评估所提出算法与其他五种算法的有效性，突出了NSGA-II的优越性。通过与经典调度策略的比较，验证了该方法的有效性。推导出的帕累托前沿验证了两个目标之间的内在冲突，表明权衡是必要的。最后，提供了管理见解，以突出这些发现对仓库管理人员的实际应用。

{"title":"Operation time and rack stability optimisation in tier-to-tier shuttle-based storage and retrieval systems with multiple retrieval locations","authors":"Zhun Xu , Liyun Xu , Huan Shao , Timo Lehmann , Andrea Matta","doi":"10.1016/j.compind.2025.104385","DOIUrl":"10.1016/j.compind.2025.104385","url":null,"abstract":"<div><div>Tier-to-tier shuttle-based storage and retrieval systems (t-SBS/RSs) are increasingly recognised in the industry for their enhanced flexibility, reduced costs, and improved shuttle utilisation. However, the ineffectiveness of task scheduling (TSc) and retrieval location assignment (LA) schemes diminishes retrieval efficiency and compromises rack stability, thereby elevating retrieval costs and jeopardising system operational safety. Moreover, a potential conflict has been identified between total operation time and rack stability during retrieval operations. Consequently, this study proposes a joint optimisation approach for the batch retrieval TSc and LA challenges in t-SBS/RS, accommodating multiple location distributions for identical cargo types. This issue is modelled as a multi-objective optimisation problem aimed at minimising total operation time while maximising rack stability. To address this, a non-dominated sorting genetic algorithm II (NSGA-II) enhanced by a prior knowledge-based initialisation strategy was proposed. Employing a real case study, several benchmark scenarios were established to assess the effectiveness of the proposed algorithm against five other algorithms, highlighting the superiority of NSGA-II. Additionally, the effectiveness of this method is validated by comparison with classical scheduling policies. The derived Pareto fronts verify the inherent conflict between the two objectives, suggesting that a trade-off is necessary. Finally, managerial insights are offered to highlight the practical application of these findings for warehouse managers.</div></div>","PeriodicalId":55219,"journal":{"name":"Computers in Industry","volume":"173 ","pages":"Article 104385"},"PeriodicalIF":9.1,"publicationDate":"2025-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145221258","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

An industrial defect detection method based on mixed noise synthesis 基于混合噪声合成的工业缺陷检测方法

IF 9.1 1区计算机科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers in Industry

Pub Date : 2025-10-02 DOI: 10.1016/j.compind.2025.104388

Aihua Ke , Jian Luo , Yaoxiang Yu , Le Li , Bo Cai

Deep learning-based methods have significantly reduced the cost of traditional manual quality inspection while enhancing accuracy and efficiency in industrial defect detection. As a result, these methods have become a prominent research focus in computer vision for intelligent manufacturing. They are increasingly applied in various production and operational contexts, including automated inspection, intelligent monitoring, and quality control. This paper presents a novel method called mixed noise synthesized defect detection, designed to identify multiple types of defects in industrial products. The proposed method employs a generative adversarial network architecture composed of a defect synthesizer, a defect discriminator, a feature extractor, and a multi-scale patch adaptor. By leveraging the feature extractor and multi-scale adaptor, the method effectively captures normal feature distributions and synthesizes realistic defect features through mixed noise synthesis, thereby significantly reducing reliance on labeled data. In addition, the defect discriminator uses a dual evaluation strategy that combines adversarial loss with Kullback–Leibler divergence to assess input features and quantify defect severity. Comprehensive experiments on benchmark anomaly detection datasets demonstrate that the method achieves high performance, with image-level and pixel-level area under the receiver operating characteristic curve scores of 99.8% and 99.4% for texture categories, and 96.7% and 98.3% for object categories, substantially outperforming state-of-the-art methods. The source code is publicly available at https://github.com/ah-ke/MNS-Defect.git.

基于深度学习的方法大大降低了传统人工质量检测的成本，同时提高了工业缺陷检测的准确性和效率。因此，这些方法已成为智能制造计算机视觉领域的一个突出研究热点。它们越来越多地应用于各种生产和操作环境，包括自动检测、智能监控和质量控制。本文提出了一种新的混合噪声综合缺陷检测方法，用于识别工业产品中多种类型的缺陷。该方法采用由缺陷综合器、缺陷鉴别器、特征提取器和多尺度贴片适配器组成的生成式对抗网络结构。该方法利用特征提取器和多尺度适配器，有效捕获正态特征分布，并通过混合噪声合成合成真实缺陷特征，从而显著降低对标记数据的依赖。此外，缺陷鉴别器采用对抗性损失与Kullback-Leibler散度相结合的双重评估策略来评估输入特征并量化缺陷严重程度。在基准异常检测数据集上进行的综合实验表明，该方法取得了较好的性能，纹理类和对象类的图像级和像素级接收器工作特征曲线下的面积得分分别为99.8%和99.4%，目标类的得分分别为96.7%和98.3%，大大优于现有方法。源代码可在https://github.com/ah-ke/MNS-Defect.git上公开获得。

{"title":"An industrial defect detection method based on mixed noise synthesis","authors":"Aihua Ke , Jian Luo , Yaoxiang Yu , Le Li , Bo Cai","doi":"10.1016/j.compind.2025.104388","DOIUrl":"10.1016/j.compind.2025.104388","url":null,"abstract":"<div><div>Deep learning-based methods have significantly reduced the cost of traditional manual quality inspection while enhancing accuracy and efficiency in industrial defect detection. As a result, these methods have become a prominent research focus in computer vision for intelligent manufacturing. They are increasingly applied in various production and operational contexts, including automated inspection, intelligent monitoring, and quality control. This paper presents a novel method called mixed noise synthesized defect detection, designed to identify multiple types of defects in industrial products. The proposed method employs a generative adversarial network architecture composed of a defect synthesizer, a defect discriminator, a feature extractor, and a multi-scale patch adaptor. By leveraging the feature extractor and multi-scale adaptor, the method effectively captures normal feature distributions and synthesizes realistic defect features through mixed noise synthesis, thereby significantly reducing reliance on labeled data. In addition, the defect discriminator uses a dual evaluation strategy that combines adversarial loss with Kullback–Leibler divergence to assess input features and quantify defect severity. Comprehensive experiments on benchmark anomaly detection datasets demonstrate that the method achieves high performance, with image-level and pixel-level area under the receiver operating characteristic curve scores of 99.8% and 99.4% for texture categories, and 96.7% and 98.3% for object categories, substantially outperforming state-of-the-art methods. The source code is publicly available at <span><span>https://github.com/ah-ke/MNS-Defect.git</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":55219,"journal":{"name":"Computers in Industry","volume":"173 ","pages":"Article 104388"},"PeriodicalIF":9.1,"publicationDate":"2025-10-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145220527","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0