Scholarlink

<p>Effective pavement management systems rely on accurate pavement condition assessments to inform timely rehabilitation measures and allocate resources that extend the pavement’s service life. Traditionally, pavement distress assessment has relied on visual inspections and manual data collection, which are time-consuming, labor-intensive, prone to human error, and often ineffective in capturing the detailed characteristics of distress types, such as variations in crack morphology, severity, and dimensions. Over the past decade, significant technological advancements have been made in data collection and analysis, with a growing focus on smart systems and machine learning models, which still require a thorough understanding. This paper presents an approach to integrating data collection using smartphone systems and analysis based on machine learning models to classify and evaluate the severity of different pavement cracks. High-resolution images of local and arterial roads in Arizona, United States, were collected using a smartphone setup. These images were used to train the models. The performance of machine learning models, including Logistic Regression (LR), Support Vector Machines (SVM), Decision Trees (DT), Random Forests (RF), Convolutional Neural Networks (CNNs), VGG16, and ResNet50—in classifying the severity of Alligator Cracks (AC) and Block Cracks (BC) is evaluated. Achieving the highest accuracies in severity classification, 91.67% for AC and 85.63% for BC, indicates that the deep learning models, especially ResNet50 and VGG16, outperform conventional machine learning methodologies. In contrast, conventional models such as LR, SVM, DT, and RF achieved lower accuracies, with LR attaining 64.74% for AC and 46.88% for BC, and DT attaining 60.26% for AC and 45.62% for BC. These models effectively captured complex patterns in pavement images, providing robust classification across various types of distress and their corresponding severities. Adopting smartphone-based data collection enables the development of rapid data-collection systems and machine learning (ML) models that transportation agencies can use to inform their pavement management systems.</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

Classification and Severity Assessment of Pavement Cracks with Low-Cost Data Collection Systems and Machine Learning Models

  • 期刊论文

  • 发表日期：

    2025年12月21日

查看全文
  • Syed Yad Morshed,
  • Tejo V. Bheemasetti

摘要

Effective pavement management systems rely on accurate pavement condition assessments to inform timely rehabilitation measures and allocate resources that extend the pavement’s service life. Traditionally, pavement distress assessment has relied on visual inspections and manual data collection, which are time-consuming, labor-intensive, prone to human error, and often ineffective in capturing the detailed characteristics of distress types, such as variations in crack morphology, severity, and dimensions. Over the past decade, significant technological advancements have been made in data collection and analysis, with a growing focus on smart systems and machine learning models, which still require a thorough understanding. This paper presents an approach to integrating data collection using smartphone systems and analysis based on machine learning models to classify and evaluate the severity of different pavement cracks. High-resolution images of local and arterial roads in Arizona, United States, were collected using a smartphone setup. These images were used to train the models. The performance of machine learning models, including Logistic Regression (LR), Support Vector Machines (SVM), Decision Trees (DT), Random Forests (RF), Convolutional Neural Networks (CNNs), VGG16, and ResNet50—in classifying the severity of Alligator Cracks (AC) and Block Cracks (BC) is evaluated. Achieving the highest accuracies in severity classification, 91.67% for AC and 85.63% for BC, indicates that the deep learning models, especially ResNet50 and VGG16, outperform conventional machine learning methodologies. In contrast, conventional models such as LR, SVM, DT, and RF achieved lower accuracies, with LR attaining 64.74% for AC and 46.88% for BC, and DT attaining 60.26% for AC and 45.62% for BC. These models effectively captured complex patterns in pavement images, providing robust classification across various types of distress and their corresponding severities. Adopting smartphone-based data collection enables the development of rapid data-collection systems and machine learning (ML) models that transportation agencies can use to inform their pavement management systems.