Surface Identification of Top-Down, Bottom-Up, and Cement-Treated Reflective Cracks Using Convolutional Neural Network and Artificial Neural Networks

Abstract

Transportation agencies often need to differentiate between top-down and bottom-up cracking in the field to set up an effective and targeted schedule and budget for the repair of these cracks. The objective of this study was to formulate a convolutional neural networks (CNN) model and to develop a decision-making tool using artificial neural networks (ANN) to identify top-down, bottom-up, and cement-treated (CT) reflective cracking for in-service flexible pavements. The CNN model was developed by modifying a pretrained network, which was fitted, tested, and validated using 200 pavement images. The CNN’s architecture consisted of five convolutional layers with three max-pooling layers and three fully connected layers. Input variables for the ANN model were pavement age, asphalt concrete (AC) thickness, annual average daily traffic (AADT), type of base, crack orientation, and crack location. The ANN network architecture consisted of three layers: an input layer of six neurons, a hidden layer of ten neurons, and a target layer of three neurons. In-service pavement sections were selected for validation and testing of the ANN model based on the parameters identified for these sites. The developed CNN model was found to achieve an accuracy of 88.9% and 86.7% in the testing and validation phases, respectively. The ANN-based decision-making tool achieved an overall accuracy of 89.3%, indicating its effectiveness in crack identification and classification.