Automatic classification of pavement crack using deep convolutional neural network

ABSTRACT

The classification of pavement crack heavily relies on the engineers’ experience or the hand-crafted algorithms. Convolutional Neural Network (CNN) has demonstrated to be useful for image classification, which provides an alternative to traditional imaging classification algorithms. This paper proposes a novel method using deep CNN to automatically classify image patches cropped from 3D pavement images. In all, four supervised CNNs with different sizes of receptive field are successfully trained. The experimental results demonstrate that all the proposed CNNs can perform the classification with a high accuracy. Overall classification accuracy of each proposed CNN is above 94%. Upon the evaluation of these neural networks with respect to accuracy and training time, we find that the size of receptive field has a slight effect on the classification accuracy. However, the CNNs with smaller size of receptive field require more training times than others.

KEYWORDS:

• Pavement
• crack classification
• deep learning
• convolutional neural network
• receptive field

Acknowledgements

The authors would like to thank Dr. Wang’s team for providing pavement 3D data.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

Funding for this research was provided by the National Natural Science Foundation of China (Grant number U153420027, 51478398). The authors gratefully acknowledge these supports