Article
Engineering, Civil
Yuki Inoue, Hiroto Nagayoshi
Summary: Pixel-level crack segmentation is a highly studied research topic with significant impact in building and road inspections. Previous approaches heavily rely on time-consuming pixel-level crack annotations, which limits their practicality. This work proposes a weakly-supervised approach to reduce annotation cost by refining inference with per-pixel brightness values and exploiting similar local visual features between falsely annotated cracks and the background. Experimental results on multiple datasets demonstrate the effectiveness of the proposed method in speeding up the annotation process while maintaining comparable detection accuracy.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2023)
Article
Chemistry, Analytical
Chungho Park, Donghyeon Kim, Hanseok Ko
Summary: This paper addresses the limitations in weakly labeled sound event detection by constructing a more efficient model with the use of GLU and dilated convolution, as well as proposing pseudo-label-based learning. Experimental results demonstrate significant performance improvement of the proposed SED model at various SNR levels.
Article
Computer Science, Artificial Intelligence
Jiabin Zhang, Hu Su, Wei Zou, Xinyi Gong, Zhengtao Zhang, Fei Shen
Summary: The proposed weakly supervised learning method, CADN, is designed for surface defect detection. It can be trained with image tag annotations only and performs image classification and defect localization simultaneously, with good real-time performance and high accuracy achieved.
PATTERN RECOGNITION
(2021)
Article
Computer Science, Artificial Intelligence
Paul Bergmann, Kilian Batzner, Michael Fauser, David Sattlegger, Carsten Steger
Summary: The detection of anomalous structures in natural image data is crucial for various tasks in the field of computer vision. Unsupervised anomaly detection methods require data for training and evaluating new approaches. The MVTec anomaly detection dataset includes high-resolution color images of different objects and textures, with normal and abnormal images for training and testing purposes.
INTERNATIONAL JOURNAL OF COMPUTER VISION
(2021)
Article
Environmental Sciences
Dilli Paudel, Diego Marcos, Allard de Wit, Hendrik Boogaard, Ioannis N. Athanasiadis
Summary: A weakly supervised deep learning framework is proposed to produce high resolution crop yield forecasts using high resolution inputs and low resolution labels. The framework is calibrated by comparing aggregated high resolution forecasts with low resolution crop statistics. Experimental results in Europe and the US show that the weakly supervised models outperform the strongly supervised models in crop yield forecasting.
ENVIRONMENTAL RESEARCH LETTERS
(2023)
Article
Computer Science, Information Systems
Yongbin Guo, Xinjian Kang, Junfeng Li, Yuanxun Yang
Summary: Facing detection problems caused by complex textile texture backgrounds, different sizes, and different types of defects, commonly used object detection networks have limitations in handling target sizes and weaker stability and anti-jamming capabilities. To meet the stringent requirements of textile defect detection, we propose a novel AC-YOLOv5-based method that fully considers the optical properties, texture distribution, imaging properties, and detection requirements specific to textiles.
Article
Materials Science, Characterization & Testing
Baoxin Zhang, Jinhan Cui, Yan Li, Jinhang Li, Yunhua Tan, Xiaoming Chen, Wenliang Wu, Xinghua Yu
Summary: Radiographic non-destructive evaluation is crucial for understanding defects in welds. Recent deep learning techniques, particularly semantic segmentation, can enhance welding defect detection and classification. Our work utilizes weakly supervised semantic segmentation based on the Cut-Cascade RCNN model to classify defects, achieving an accuracy of 90.15% on our x-ray dataset.
NDT & E INTERNATIONAL
(2023)
Article
Ecology
Mohammad Jahanbakht, Mostafa Rahimi Azghadi, Nathan J. Waltham
Summary: Fish play a crucial role in marine ecosystems and are important for a healthy human diet. Deep Neural Networks (DNNs) can be used to detect fish in underwater videos, but training them requires large, labeled datasets which are time-consuming to create. To address this, the researchers collected a dataset called FishInTurbidWater and used it to develop a semi-supervised contrastive learning fish detection model. They also created a weakly-supervised ensemble DNN for fish detection in turbid waters. Both models showed faster turnaround time and reasonably high accuracy compared to traditional fully-supervised models. The dataset and code are publicly available.
ECOLOGICAL INFORMATICS
(2023)
Article
Engineering, Multidisciplinary
Lin Wang, Xiangjun Wang, Feng Liu, Mingyang Li, Xin Hao, Nianfu Zhao
Summary: This paper proposes an anomaly detection method based on weakly supervised learning, which transforms the weakly supervised learning problem into a fully supervised learning problem through a pseudo-label generation technique. The network combines attention and guidance augmentation modules to improve the model's spatial localization capability. Experimental results show that the method achieves high AUC values on datasets with different scales and scene complexity.
Article
Engineering, Electrical & Electronic
Kuikui He, Xiaotao Liu, Jing Liu, Peng Wu
Summary: In this article, a novel weakly supervised DL-based method is proposed for accurately segmenting and locating defects on textured surfaces. Utilizing multitask learning with a unique encoder and two decoders has improved the performance of the deep neural network. Experimental results demonstrate that the method outperforms other state-of-the-art methods in defect detection.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2021)
Article
Materials Science, Multidisciplinary
Yichuan Shao, Shuo Fan, Haijing Sun, Zhenyu Tan, Ying Cai, Can Zhang, Le Zhang
Summary: Defect classification is crucial in steel surface defect detection. Traditional methods using convolutional neural networks (CNNs) improve accuracy by increasing network depth and parameter count. However, this approach overlooks the memory overhead and diminishing accuracy gains. To address these issues, a multi-scale lightweight neural network model (MM) is proposed, which uses a fusion encoding module and Gaussian difference pyramid. Experimental results show that MM network achieves 98.06% accuracy in defect classification, surpassing other networks in both parameter reduction and accuracy.
Article
Computer Science, Artificial Intelligence
Hui Lv, Chuanwei Zhou, Zhen Cui, Chunyan Xu, Yong Li, Jian Yang
Summary: This paper proposes a Weakly Supervised Anomaly Localization (WSAL) method that focuses on temporally localizing anomalous segments within anomalous videos, achieving state-of-the-art performance on the UCF-Crime and TAD datasets. The method utilizes a high-order context encoding model and an enhancement strategy to improve accuracy, and also introduces a new traffic anomaly (TAD) dataset to diversify anomaly detection benchmarks.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2021)
Article
Computer Science, Artificial Intelligence
Hongbing Shang, Chuang Sun, Jinxin Liu, Xuefeng Chen, Ruqiang Yan
Summary: Surface defect detection plays a crucial role in intelligent manufacturing and product life-cycle management. Existing methods mainly rely on convolutional architectures, but the limited receptive field poses challenges for performance improvement. Transformer-based models, with their ability to model long-range dependencies, have achieved success in computer vision. However, using Transformer models without modification lacks defect awareness and relevance.
ADVANCED ENGINEERING INFORMATICS
(2023)
Article
Engineering, Multidisciplinary
Chao Zhao, Xin Shu, Xi Yan, Xin Zuo, Feng Zhu
Summary: Steel surfaces may have defects due to imperfect manufacturing techniques and external factors, which significantly impact their lifespan and usability. Therefore, surface defect detection is crucial in industrial production. However, traditional detection algorithms suffer from low accuracy and speed. In this study, we propose a model called RDD-YOLO, which is based on YOLOv5, for steel surface defect detection. Our model utilizes Res2Net blocks for feature extraction, a double feature pyramid network for enhanced representations, and a decoupled head for improved detection precision. Experimental results demonstrate that RDD-YOLO achieves accuracies of 81.1 mAP on NEU-DET and 75.2 mAP on GC10-DET, surpassing YOLOv5 by 4.3% and 5.8% respectively. Our proposed model shows comprehensive performance in steel surface defect detection.
Article
Computer Science, Interdisciplinary Applications
Meijian Ren, Rulin Shen, Yanling Gong
Summary: This paper proposes a defect detection method based on convolutional neural network, which extracts features through a backbone network, improves prediction accuracy through feature fusion, and further enhances pixel classification accuracy through an encoding module.
JOURNAL OF COMPUTING AND INFORMATION SCIENCE IN ENGINEERING
(2022)
Article
Computer Science, Interdisciplinary Applications
Francesco Pistolesi, Michele Baldassini, Beatrice Lazzerini
Summary: More than one in four workers worldwide suffer from back pain, resulting in the loss of 264 million work days annually. In the U.S., it costs $50 billion in healthcare expenses each year, rising up to $100 billion when accounting for decreased productivity and lost wages. The impending Industry 5.0 revolution emphasizes worker well-being and their rights, such as privacy, autonomy, and human dignity. This paper proposes a privacy-preserving artificial intelligence system that monitors the posture of assembly line workers. The system accurately assesses upper-body and lower-body postures while respecting privacy, enabling the detection of harmful posture habits and reducing the likelihood of musculoskeletal disorders.
COMPUTERS IN INDUSTRY
(2024)
Article
Computer Science, Interdisciplinary Applications
Xavier Boucher, Camilo Murillo Coba, Damien Lamy
Summary: This paper explores the new business strategies of digital servitization and smart PSS delivery, and develops conceptual prototypes of smart PSS value offers for early stages of the design process. It presents the development and experimentation of a modelling language and toolkit, and applies it to the design of a smart PSS in the field of heating appliances.
COMPUTERS IN INDUSTRY
(2024)
Article
Computer Science, Interdisciplinary Applications
Dieudonne Tchuente, Jerry Lonlac, Bernard Kamsu-Foguem
Summary: Artificial Intelligence (AI) is becoming increasingly important in various sectors of society. However, the black box nature of most AI techniques such as Machine Learning (ML) hinders their practical application. This has led to the emergence of Explainable artificial intelligence (XAI), which aims to provide AI-based decision-making processes and outcomes that are easily understood, interpreted, and justified by humans. While there has been a significant amount of research on XAI, there is currently a lack of studies on its practical applications. To address this research gap, this article proposes a comprehensive review of the business applications of XAI and a six-step framework to improve its implementation and adoption by practitioners.
COMPUTERS IN INDUSTRY
(2024)
Article
Computer Science, Interdisciplinary Applications
Francois-Alexandre Tremblay, Audrey Durand, Michael Morin, Philippe Marier, Jonathan Gaudreault
Summary: Continuous high-frequency wood drying, integrated with a traditional wood finishing line, improves the value of lumber by correcting moisture content piece by piece. Using reinforcement learning for continuous drying operation policies outperforms current industry methods and remains robust to sudden disturbances.
COMPUTERS IN INDUSTRY
(2024)
Article
Computer Science, Interdisciplinary Applications
Luyao Xia, Jianfeng Lu, Yuqian Lu, Wentao Gao, Yuhang Fan, Yuhao Xu, Hao Zhang
Summary: Efficient assembly sequence planning is crucial for enhancing production efficiency, ensuring product quality, and meeting market demands. This study proposes a dynamic graph learning algorithm called assembly-oriented graph attention sequence (A-GASeq), which optimizes the assembly graph structure to guide the search for optimal assembly sequences. The algorithm demonstrates superiority and broad utility in real-world scenarios.
COMPUTERS IN INDUSTRY
(2024)
Article
Computer Science, Interdisciplinary Applications
Mutahar Safdar, Padma Polash Paul, Guy Lamouche, Gentry Wood, Max Zimmermann, Florian Hannesen, Christophe Bescond, Priti Wanjara, Yaoyao Fiona Zhao
Summary: Metal-based additive manufacturing can achieve fully dense metallic components, and the application of machine learning in this field has been growing rapidly. However, there is a lack of framework to manage these machine learning models and guidance on the fundamental requirements for a cross-disciplinary platform to support process-based machine learning models in industrial metal AM.
COMPUTERS IN INDUSTRY
(2024)