Deep Learning for Asphalt Pavement Cracking Recognition Using Convolutional Neural Network

Abstract

Due to the complexity and diversity of pavement surfaces, pavement cracking detection is a challenging task even for human operators. The automation of pavement cracking detection generally requires robust algorithms with high level of intelligence. From such a perspective, Deep Learning, a promising branch of Artificial Intelligence, can serve as an advanced approach to high level of intelligence by learning from huge amount of historical data and enhancing the capability of behaving correctly under unforeseen and complex environments. This paper proposes a convolutional neural network (CNN) for recognizing cracks on asphalt surfaces at subdivided image cells. The proposed CNN consists of 3 convolution layers and 2 fully-connected layers with 1,246,240 parameters in total. Feeding the proposed CNN with 5,000 manually-processed example images recursively, the accuracy of the proposed CNN is progressively improved, and the generalization is also enhanced. It is demonstrated in the paper that the trained CNN can achieve high accuracies 96.32 and 94.29% on training data and testing data respectively.