Detection and classification of pavement damages using wavelet scattering transform, fractal dimension by box-counting method and machine learning algorithms

Pavement condition assessment is essential for roadway maintenance and rehabilitation processes. Image-based inspection methods provide information regarding the surface of the pavement and allow quantitative analyses of pavement conditions. A methodology for the detection of damages in the pavement, by applying pattern recognition and image analysis and machine learning algorithms, is presented in the paper. This methodology consists of image acquisition, image processing using the wavelet scattering transform (WST), feature extraction employing the fractal dimension by box-counting method, and finally classification. The methodology was applied for the detection of three common types of damages: potholes, longitudinal and alligator cracks. Two different supervised learning algorithms, Artificial Neural Networks (ANN) and Support Vector Machine (SVM), were used for classification and results are compared training Convolutional Neural Networks (CNN). The multilayer ANN an overall accuracy of 98.36% and an F-score of 98.33%, while the SVM an accuracy of 97.22% and an F-score of 97.22%.

Automated summary

This paper presents a methodology for the detection of damages in pavement using pattern recognition and image analysis with machine learning algorithms. The methodology includes image acquisition, processing with wavelet scattering transform, feature extraction using fractal dimension, and classification. Results showed high accuracy in detecting three common types of damages using Artificial Neural Networks and Support Vector Machine.