Deep Machine Learning Approach to Develop a New Asphalt Pavement Condition Index (2004.13314v1)

Abstract: Automated pavement distress detection via road images is still a challenging issue among pavement researchers and computer-vision community. In recent years, advancement in deep learning has enabled researchers to develop robust tools for analyzing pavement images at unprecedented accuracies. Nevertheless, deep learning models necessitate a big ground truth dataset, which is often not readily accessible for pavement field. In this study, we reviewed our previous study, which a labeled pavement dataset was presented as the first step towards a more robust, easy-to-deploy pavement condition assessment system. In total, 7237 google street-view images were extracted, manually annotated for classification (nine categories of distress classes). Afterward, YOLO (you look only once) deep learning framework was implemented to train the model using the labeled dataset. In the current study, a U-net based model is developed to quantify the severity of the distresses, and finally, a hybrid model is developed by integrating the YOLO and U-net model to classify the distresses and quantify their severity simultaneously. Various pavement condition indices are developed by implementing various machine learning algorithms using the YOLO deep learning framework for distress classification and U-net for segmentation and distress densification. The output of the distress classification and segmentation models are used to develop a comprehensive pavement condition tool which rates each pavement image according to the type and severity of distress extracted.

• The paper proposes a deep learning framework integrating YOLO and U-net models to automate asphalt pavement condition assessment using Google Street View images.
• The study created a new labeled Pavement Image Dataset (PID) and demonstrated high precision (YOLO precision ~0.93) for distress detection and quantification.
• The validated approach offers a more precise, automated pavement condition index compared to traditional methods, enabling cost-effective road maintenance decisions.

Deep Machine Learning Approach to Develop a New Asphalt Pavement Condition Index

The paper presents a compelling strategy for improving pavement condition assessment by leveraging deep learning models that classify and quantify pavement distress using image analysis. Conventionally, pavement distress detection relies on costly and inefficient methods dependent on human intervention, such as sophisticated survey vehicles. The authors introduce an alternative approach through the application of deep convolutional neural networks (DCNNs) to analyze pavement images sourced from Google Street View.

The paper initiates with the development of a labeled Pavement Image Dataset (PID), consisting of 7,237 images annotated for nine categories of pavement distress. It acts as the cornerstone for training the YOLO-based classification model and a U-net model for segmentation and density quantification. By integrating these models into a hybrid framework, researchers propose a novel methodology to assess pavement condition indexes using image data, thereby reducing human reliance in the process.

Some pivotal aspects and contributions of the paper include:

1. Data and Model Development: The PID dataset offers a diverse set of road images enabling the classification and severity quantification of multiple distress types. The usage of the YOLO v2 framework ensures high precision in object detection, while the U-net architecture adeptly quantifies the density of cracks amidst challenges such as lighting and shadows.
2. Hybrid Model: The integration of YOLO and U-net models overcomes limitations inherent in using each model individually. It allows simultaneous distress classification and density estimation, consequently enhancing the reliability and accuracy of pavement condition assessment.
3. Comparative Analysis: Rich comparative analyses juxtapose the proposed indices against traditional PASER ratings, emphasizing the precision improvements rendered by computerized models. The precision and recall for the YOLO model averaged around 0.93 and 0.77, respectively, showcasing robustness.
4. Machine Learning Methodology: Genetic Expression Programming (GEP), linear regression, and weight-based index prediction models are employed to rate pavement conditions using distress classification and density outputs. High determination coefficients (R²) across training and testing datasets indicate model efficacy.
5. Implications and Validation: The paper validates models using imagery from additional road sections, illustrating consistent alignment with human-assessed PASER values. The fluctuations in predicted values underscore the precision achievable with the automated model, overcoming inherent rigidity in PASER’s manual ratings.

The paper establishes a paradigm shift toward automated and cost-effective pavement condition monitoring using readily available street-view imagery, culminating in an evaluation tool of paramount importance in timely road maintenance decisions. Future studies are envisaged to expand dataset breadth for model robustness, explore 3D imagery analysis, and develop comprehensive, deployable software solutions integrating the modeling framework.

The research demonstrates a significant advance in computer vision applications within civil infrastructure, setting a benchmark in both theoretical innovation and practical applicability. As the technology evolves, expanding collaborative efforts to enhance dataset diversity and developing adaptive models with real-time updating capabilities present exciting opportunities for further exploration and refinement.