Abstract: In this paper, we propose a data augmentation technique based on Convolutional Neural Networks (CNN or ConvNet) training to efficiently obtain a dataset of images containing concrete cracks.