Crack Identification Method of Steel Fiber Reinforced Concrete Based on Deep Learning: A Comparative Study and Shared Crack Database

Abstract

As a common disease of concrete structure in engineering, cracks mainly lead to durability problems such as steel corrosion, rain erosion, and protection layer peeling, and then the building gets destroyed. In order to detect the cracks of concrete structure in time, the bending test of steel fiber reinforced concrete is carried out, and the pictures of concrete cracks are obtained. Furthermore, the crack database is expanded by the migration learning method and the crack database is shared on the Baidu online disk. Finally, a concrete crack identification model based on YOLOv4 and Mask R-CNN is established. In addition, the improved Mask R-CNN method is proposed in order to improve the prediction accuracy based on the Mask R-CNN. The results show that the average prediction accuracy of concrete crack identification is 82.60% based on the YOLO v4 method. The average prediction accuracy of concrete crack identification is 90.44% based on the Mask R-CNN method. The average prediction accuracy of concrete crack identification is 96.09% based on the improved Mask R-CNN method.