Research on Automatic Identification Methods for Apparent Damage of Concrete Structure Bridge Based on YOLOv4

After rapid development for almost twenty years, the highway infrastructures of China's transportation industry, especially the concrete structure bridges, reached a peak of maintenance. Currently, the bridge apparent damage has mainly depended on manual detection, and it is characterized by low efficiency, large labor intensity and great susceptibility to subjective factor. Aiming at the automatic identification of apparent damage, the YOLOv4 is introduced for detecting apparent damage (flaking, exposed bars, honeycombs and holes, etc.) of concrete structure bridge. Through the experiment, it is proved that YOLOv4-based detection method can accurately identify the common apparent damage, and its detection accuracy rate is superior to that of other single-stage target detection algorithms such as YOLOv3 and SSD, etc. While he threshold value of IoU is set to 0.5, the detection precision of YOLOv4, namely the mAP value, reaches 71.3%, which is able to well meet the demand for real-time identification apparent damage of concrete structure bridge.