基于YOLOv3的H型钢表面缺陷检测系统

为了提升H型钢的表面质量和安全系数,设计了一种基于YOLOv3算法的型钢表面缺陷检测系统。设计的检测系统由硬件系统和软件系统组成:硬件系统包括八角架式图像采集装置、电动控制系统和通信系统;软件系统采用YOLOv3目标检测算法。现场测试结果表明:1)检测系统可实时采集图像,并根据H型钢的规格型号能够自动调节图像采集装置结构,准确快速地跟踪被检测目标,获得H型钢的高清全景图像;2)检测系统可对H型钢表面缺陷进行在线检测、分类和定位,并兼顾检测精度与检测速度,检测精度为81.25%,检测速度为30.78帧/s; 3)检测系统能够准确识别H型钢的结疤、凹坑、划伤和击伤等4类典型缺陷,满足生产过程中表面缺陷检测的实际需求。开发的型钢表面缺陷检测系统为H型钢表面质量智能化检测工作提供了新的选择。

H-beam surface defect detection system based on YOLOv3 algorithm

In order to improve the surface quality and safety factor of H-beam, an H-beam surface defect detection system based on YOLOv3 (you only look once) algorithm was designed. The designed detection system was composed of hardware system and software system. Octagonal frame image acquisition device, electric control system and communication system were included in the hardware system. The YOLOv3 target detection algorithm was used in the software system. The field test results show that: 1) the detection system can capture the surface image of H-beam in real time, and automatically adjust the structure of the image acquisition device according to the types of H-beam, accurately and quickly track the detected targets, and obtain high-definition panoramic images of H-beam; 2) the online detection, classification and location of H-beam surface defects can be carried out by the detection system, and both the detection accuracy and the detection speed are taken into account. The detection accuracy is 81.25%, and the detection speed is 30.78 frames/s; 3) the detection system can accurately identify four types of typical defects of H-beam, such as scar, pit, scratch and hit, which can meet the actual requirements of surface defect detection in the production process. The developed surface defect detection system provides a new choice for the intelligent detection of H-beam surface quality.