Aiming at the problem of fast detection,identification and numerical extraction of surface cracks,an improved lightweight algorithm ConvCrackDet based on YOLOx target detection frame-work was proposed.The numerical extraction method was integrated to realize rapid detection and cal-culate the specific information of cracks.The results show that among the 980 data sets integrated by four data sets,the results of AP50=90.9,parameter quantity=7.06 and reasoning speed=3.23ms are obtained,which verifies the advantages of this method in improving accuracy,reducing complexity and reasoning time,and accurately obtains numerical information such as fracture length and width.
关键词
深度学习/计算机视觉/混凝土表面裂缝/裂缝检测/目标检测
Key words
deep learning/computer vision/concrete surface crack/crack detection/object detection
维普
万方数据