Enhancing Pavement Crack Classification with Bidirectional Cascaded Neural Networks

2025-03-27

AI Summary

To address low fine-grained classification accuracy and poor robustness in pavement crack detection, this paper proposes a Bidirectional Cascaded Neural Network (BCNN), the first to integrate bidirectional information flow with cascaded progressive optimization for jointly modeling local texture and global structural features. The method incorporates U-Net50 for image enhancement and is evaluated on a real-world dataset of 599 road images, achieving precise classification of three defect types: linear cracks, potholes, and fatigue cracks. Experimental results show an overall accuracy of 87%, with per-class F1-scores of 0.85, 0.93, and 0.85, respectively; macro- and weighted-average F1-scores both reach 0.88. This work significantly enhances discriminative robustness and classification reliability under complex crack patterns, providing a practical, deployable technical foundation for intelligent pavement maintenance decision-making.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Planning, Routing, and Scheduling: Activity and Plan Recognition

Application Category

• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Pavement distress, such as cracks and potholes, is a significant issue affecting road safety and maintenance. In this study, we present the implementation and evaluation of Bidirectional Cascaded Neural Networks (BCNNs) for the classification of pavement crack images following image augmentation. We classified pavement cracks into three main categories: linear cracks, potholes, and fatigue cracks on an enhanced dataset utilizing U-Net 50 for image augmentation. The augmented dataset comprised 599 images. Our proposed BCNN model was designed to leverage both forward and backward information flows, with detection accuracy enhanced by its cascaded structure wherein each layer progressively refines the output of the preceding one. Our model achieved an overall accuracy of 87%, with precision, recall, and F1-score measures indicating high effectiveness across the categories. For fatigue cracks, the model recorded a precision of 0.87, recall of 0.83, and F1-score of 0.85 on 205 images. Linear cracks were detected with a precision of 0.81, recall of 0.89, and F1-score of 0.85 on 205 images, and potholes with a precision of 0.96, recall of 0.90, and F1-score of 0.93 on 189 images. The macro and weighted average of precision, recall, and F1-score were identical at 0.88, confirming the BCNN's excellent performance in classifying complex pavement crack patterns. This research demonstrates the potential of BCNNs to significantly enhance the accuracy and reliability of pavement distress classification, resulting in more effective and efficient pavement maintenance and management systems.

Problem

• Classify pavement cracks using Bidirectional Cascaded Neural Networks
• Improve accuracy of crack detection with image augmentation
• Enhance road maintenance via precise distress classification

Innovation

• Bidirectional Cascaded Neural Networks for crack classification
• U-Net 50 used for image augmentation
• Cascaded structure refines output progressively