Scholarlink

<p>Concrete pavement fatigue cracking is one of the most significant challenges of transportation infrastructure, and it has a direct impact on transportation quality. However, machine learning and deep learning models are unable to forecast asphalt pavement fatigue due to internal progressive damage within the structure, and this continues to be a major issue for pavement experts. In order to tackle the aforementioned difficulties, a Multi-View Spatial–Temporal Graph Convolutional Network (MV-STGCN) for pavement crack classification is proposed. A Pavement Crack Data is collected and then preprocessed using Multi-stage Progressive Denoising Network (MSPNet) and Adaptive Learning Attention Network (LANet) for noise reduction and to enhance contrast in the crack image. The next phase is segmenting the portions of an image that have cracks using a modified S Mask Region-based Convolutional Neural Networks&#xa0;(R-CNN)&#xa0;with InceptionResNetV2 as the backbone. Using the Two-Branch Multiscale Context Feature Extraction Module (TMCFM), which extracts several image scales to capture the texture of fracture patterns for feature extraction. The extracted features are then fed into a Multi-View Spatial–Temporal Graph Convolutional Network (MV-STGCN) in order to identify the input image as either Crack or Non-Crack, depending on the pavement state. The proposed strategy predicts fatigue pavement cracks with an accuracy of 96.80%, a TNR of 97.80%, and an FDR of 6.10%, respectively. Therefore, the proposed approach is more accurate and efficient at predicting pavement cracks with large amounts of 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

Two-Branch Multiscale Context with Multi-View Spatial–Temporal Graph Convolutional Networks for Pavement Fatigue Cracking Prediction

  • 期刊论文

  • 发表日期：

    2025年11月4日

查看全文
  • Santosh Kumar,
  • Amit Madhukar,
  • Suvir Kumar,
  • Shruti Shriya

摘要

Concrete pavement fatigue cracking is one of the most significant challenges of transportation infrastructure, and it has a direct impact on transportation quality. However, machine learning and deep learning models are unable to forecast asphalt pavement fatigue due to internal progressive damage within the structure, and this continues to be a major issue for pavement experts. In order to tackle the aforementioned difficulties, a Multi-View Spatial–Temporal Graph Convolutional Network (MV-STGCN) for pavement crack classification is proposed. A Pavement Crack Data is collected and then preprocessed using Multi-stage Progressive Denoising Network (MSPNet) and Adaptive Learning Attention Network (LANet) for noise reduction and to enhance contrast in the crack image. The next phase is segmenting the portions of an image that have cracks using a modified S Mask Region-based Convolutional Neural Networks (R-CNN) with InceptionResNetV2 as the backbone. Using the Two-Branch Multiscale Context Feature Extraction Module (TMCFM), which extracts several image scales to capture the texture of fracture patterns for feature extraction. The extracted features are then fed into a Multi-View Spatial–Temporal Graph Convolutional Network (MV-STGCN) in order to identify the input image as either Crack or Non-Crack, depending on the pavement state. The proposed strategy predicts fatigue pavement cracks with an accuracy of 96.80%, a TNR of 97.80%, and an FDR of 6.10%, respectively. Therefore, the proposed approach is more accurate and efficient at predicting pavement cracks with large amounts of data.