Scholarlink

<p>This paper presents an explainable hybrid framework for pavement crack detection to extend road service life and ensure public safety. The key objective of this study is to develop an automated crack detection system by integrating deep learning-based feature extraction with stacked ensemble learning classification. Four state-of-the-art pre-trained convolutional neural networks such as Xception, NASNetMobile, MobileNetV2, and EfficientNetB0 are utilized as feature extractors to capture high-level crack characteristics from pavement images. Extracted deep features are subsequently fed into stacked ensemble model. This stacking model consists of two stages; the first stage uses the Support Vector Machine (SVM), Random Forest (RF), AdaBoost, and Extra Trees. The second stage operates on Logistic Regression using the first stage predictions for the final accurate binary classification of cracked versus non-cracked surfaces. To improve model transparency and interpretability, Gradient-weighted Class Activation Mapping (GradCAM) is integrated as an explainable artificial intelligence (XAI) component, producing visual heatmap overlays that highlight image regions most influential in crack detection decisions. The proposed Xception with the Stacked-Ensemble explained framework performed the best with an accuracy, precision, recall, and F1-Score of 90.00%, 90.99%, 90.00%, and 89.94%, respectively. These metrics improve by approximately 2.22%, 1.64%, 2.22%, and 2.24% respectively in comparison to the best performing classifier. Its performance makes it dependable for real-time deployment and support for pavement crack detection in the context of smart roads. From the results, it can be seen that this hybrid model performs much better than conventional models by reducing the need for subjective manual evaluation while providing real-time intelligent pavement inspection. Hence, the proposed system enables timely maintenance, lowers costs, and promotes sustainable smart city infrastructure management.</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

An Explainable Hybrid Deep Learning Framework for Automated Pavement Crack Detection Using GradCAM and Stacked Ensemble Classifiers in Smart Road Monitoring System

  • 期刊论文

  • 发表日期：

    2026年7月22日

查看全文
  • Ezz El-Din Hemdan,
  • M. E. Al-Atroush

摘要

This paper presents an explainable hybrid framework for pavement crack detection to extend road service life and ensure public safety. The key objective of this study is to develop an automated crack detection system by integrating deep learning-based feature extraction with stacked ensemble learning classification. Four state-of-the-art pre-trained convolutional neural networks such as Xception, NASNetMobile, MobileNetV2, and EfficientNetB0 are utilized as feature extractors to capture high-level crack characteristics from pavement images. Extracted deep features are subsequently fed into stacked ensemble model. This stacking model consists of two stages; the first stage uses the Support Vector Machine (SVM), Random Forest (RF), AdaBoost, and Extra Trees. The second stage operates on Logistic Regression using the first stage predictions for the final accurate binary classification of cracked versus non-cracked surfaces. To improve model transparency and interpretability, Gradient-weighted Class Activation Mapping (GradCAM) is integrated as an explainable artificial intelligence (XAI) component, producing visual heatmap overlays that highlight image regions most influential in crack detection decisions. The proposed Xception with the Stacked-Ensemble explained framework performed the best with an accuracy, precision, recall, and F1-Score of 90.00%, 90.99%, 90.00%, and 89.94%, respectively. These metrics improve by approximately 2.22%, 1.64%, 2.22%, and 2.24% respectively in comparison to the best performing classifier. Its performance makes it dependable for real-time deployment and support for pavement crack detection in the context of smart roads. From the results, it can be seen that this hybrid model performs much better than conventional models by reducing the need for subjective manual evaluation while providing real-time intelligent pavement inspection. Hence, the proposed system enables timely maintenance, lowers costs, and promotes sustainable smart city infrastructure management.