Scholarlink

<p>Cracks in various structures pose significant threats to their integrity and can lead to catastrophic failures if not detected and addressed promptly. Traditional crack detection methods often rely on manual inspections, which are time-consuming, labour-intensive and prone to human error. To overcome these limitations, researchers have turned to developing an automated crack detection system. With the advancements in the field of ML and DL methods, there has been a significant shift towards automated systems that leverage these technologies to enhance accuracy, efficiency and reliability. This review paper aims to provide a comprehensive overview of the different approaches used for crack detection and segmentation through ML and DL methods. The paper begins by discussing the fundamentals of crack detection and the limitations of traditional approaches. It then delves into the key concepts and algorithms of ML and DL such as Random Forest (RF), Convolutional Neural Network (CNN), Support Vector Machine (SVM) and Recurrent Neural Network (RNN). This paper compares approximately 50 technical and review papers and at the end, the comparative results are shown. The result has been shown according to the method classification, detection and segmentation which shows that Visual Geometry Group (VGG-16) with an accuracy of 99.83%, Fully Convolutional Network- Region-based Convolutional Neural Network (FCN-RCNN) and U-shaped Encoder-Decoder Network (UNET) with an accuracy of 99.52% and encoder-decoder with 25% higher performance result respectively. In a study which shows CNN approach for vibration-based damage gives up to 99.9% accuracy with noisy data.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A review of image-based deep learning methods for crack detection

  • 期刊论文

  • 发表日期：

    2025年3月12日

查看全文
  • Vindhyesh Pandey,
  • Shambhu Sharan Mishra

摘要

Cracks in various structures pose significant threats to their integrity and can lead to catastrophic failures if not detected and addressed promptly. Traditional crack detection methods often rely on manual inspections, which are time-consuming, labour-intensive and prone to human error. To overcome these limitations, researchers have turned to developing an automated crack detection system. With the advancements in the field of ML and DL methods, there has been a significant shift towards automated systems that leverage these technologies to enhance accuracy, efficiency and reliability. This review paper aims to provide a comprehensive overview of the different approaches used for crack detection and segmentation through ML and DL methods. The paper begins by discussing the fundamentals of crack detection and the limitations of traditional approaches. It then delves into the key concepts and algorithms of ML and DL such as Random Forest (RF), Convolutional Neural Network (CNN), Support Vector Machine (SVM) and Recurrent Neural Network (RNN). This paper compares approximately 50 technical and review papers and at the end, the comparative results are shown. The result has been shown according to the method classification, detection and segmentation which shows that Visual Geometry Group (VGG-16) with an accuracy of 99.83%, Fully Convolutional Network- Region-based Convolutional Neural Network (FCN-RCNN) and U-shaped Encoder-Decoder Network (UNET) with an accuracy of 99.52% and encoder-decoder with 25% higher performance result respectively. In a study which shows CNN approach for vibration-based damage gives up to 99.9% accuracy with noisy data.