Identification of Concrete Crack Using Deep Learning Based Approach
Ziyan Bian *
School of Civil and Transportation, North China University of Water Resources and Electric Power, Zhengzhou-450045, Henan, China.
*Author to whom correspondence should be addressed.
Abstract
Cracks reflect the safety and durability of concrete structures, and the existing artificial crack detection has the disadvantages of low efficiency and large error. However, the use of deep learning of images to identify cracks has the advantages of high efficiency, small error and low cost. This paper systematically discusses the deep learning in the identification of concrete structure cracks, expounds the deep learning technology, studies the SqueezeNet network model and YOLO (You Only Look Once) network model in the field of concrete structure crack identification, and improves the model according to the test results. In this paper, the classification accuracy based on crack trend is improved by increasing the width and depth of the Fire module of the SqueezeNet network. Secondly, the target detection accuracy is improved by the transfer learning of the YOLO model. The results show that the classification accuracy of the improved SqueezeNet model is more than 82% and the recognition accuracy of YOLO model is more than 92%.
Keywords: Deep learning, crack recognition, SqueezeNet model, YOLO model