Classification of Asphalt Pavement Cracks Using Laplacian Pyramid-Based Image Processing and a Hybrid Computational Approach

Abstract

To improve the efficiency of the periodic surveys of the asphalt pavement condition, this study puts forward an intelligent method for automating the classification of pavement crack patterns. The new approach relies on image processing techniques and computational intelligence algorithms. The image processing techniques of Laplacian pyramid and projection integral are employed to extract numerical features from digital images. Least squares support vector machine (LSSVM) and Differential Flower Pollination (DFP) are the two computational intelligence algorithms that are employed to construct the crack classification model based on the extracted features. LSSVM is employed for data classification. In addition, the model construction phase of LSSVM requires a proper setting of the regularization and kernel function parameters. This study relies on DFP to fine-tune these two parameters of LSSVM. A dataset consisting of 500 image samples and five class labels of alligator crack, diagonal crack, longitudinal crack, no crack, and transverse crack has been collected to train and verify the established approach. The experimental results show that the Laplacian pyramid is really helpful to enhance the pavement images and reveal the crack patterns. Moreover, the hybridization of LSSVM and DFP, named as DFP-LSSVM, used with the Laplacian pyramid at the level 4 can help us to achieve the highest classification accuracy rate of 93.04%. Thus, the new hybrid approach of DFP-LSSVM is a promising tool to assist transportation agencies in the task of pavement condition surveying.

Efficient data dimensionality reduction method for improving road crack classification algorithms

“…To carry out a performance analysis of our proposal, we have used the five datasets generated from the previous section and employed the classification algorithms detailed in Section 4. In addition, in order to compare results, we have used the data reduction methods of Section 3 and the proposal of Cubero-Fernandez et al ( 2017), Hoang (2018), and Rodriguez-Lozano et al ( 2020) under the name "PP" on the same datasets. However, to set a fair comparison of our proposal with the methods presented in Section 3 and PP, all images in each dataset are set to the same resolution: 640 × 480.…”

“…Zhang & Yuen 2021). These proposals range from algorithms based on image enhancement (Hoang, 2018), illumination correction (C. Chen et al, 2021), and edge detection (Cubero-Fernandez et al, 2017) to others based on more sophisticated methods, such as minimal path estimation (Amhaz et al, 2016), shadow removal (Palomar et al, 2010), or even classic machine learning algorithms (Y. Shi et al, 2016).…”

“…The second one is the creation of a binary image (typical values between 0 and 255) that emphasizes the pixels that correspond to the damaged surface (Y. Zhang & Yuen 2021). These proposals range from algorithms based on image enhancement (Hoang, 2018), illumination correction (C. Chen et al., 2021), and edge detection (Cubero‐Fernandez et al., 2017) to others based on more sophisticated methods, such as minimal path estimation (Amhaz et al., 2016), shadow removal (Palomar et al., 2010), or even classic machine learning algorithms (Y. Shi et al., 2016). Furthermore, the emerging growth of deep learning algorithms is being used widely, and we will review it in Section 2.…”

Revisión de métodos para la clasificación de fallas superficiales en pavimentos flexibles

“…Las svm son un método de clasificación donde la idea principal es construir hiperplanos como superficies de decisión, de tal manera que el margen de separación entre los ejemplos positivos y negativos se maximice [65]. Algunas publicaciones, como [19], [21], [23], [25], [43], [35] y [62], muestran que las svm arrojan resultados satisfactorios en la detección de fallas en pavimentos. Dentro de las fortalezas de las svm tenemos que su modelamiento no necesita la totalidad de puntos disponibles del conjunto de entrenamiento para hallar la separación entre clases, lo cual representa una ventaja frente a otros métodos que utilizan un porcentaje alto de las muestras del conjunto de entrenamiento.…”

Crack Detection and Comparison Study Based on Faster R-CNN and Mask R-CNN

“…In the previous ten years, many scholars have conducted in-depth examinations of road crack recognition primarily based on digital image processing. Hoang [ 1 ] proposed a smart method of automatically classifying road cracks to enhance the effectiveness of periodic surveys of asphalt pavement conditions. The new method depends on algorithms of computational intelligence and methods of image processing.…”