基于新型CNN优化方法的混凝土表面裂缝实时检测与分析

Real-time Detection and Measurement of Concrete Surface Crack Based on Improved CNN Method

针对表面裂缝的快速检测识别和数值提取问题,提出了基于YOLOx目标检测框架的改进性轻量算法ConvCrackDet,并集成数值提取方法,实现快速检测的同时可计算出裂缝的具体信息.结果表明:在所使用的由四个数据集集成的980个数据中,取得了AP_(50)=90.9,参数量=7.06,推理速度=3.23 ms的结果,验证了该方法在提高精度、减小复杂度,以及推理时间方面的优势,并准确地获取裂缝的长度宽度等数值信息.

Aiming at the problem of fast detection,identification and numerical extraction of surface cracks,an improved lightweight algorithm ConvCrackDet based on YOLOx target detection framework was proposed.The numerical extraction method was integrated to realize rapid detection and calculate the specific information of cracks.The results show that among the 980 data sets integrated by four data sets,the results of AP_(50)=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.