一种基于YOLOv3算法的3D打印点阵结构缺陷识别方法  

A Method for Detecting Internal Defects of Metal Lattice Structure Based on YOLOv3 Algorithm

作　　者：张玉燕[1,2] 任腾飞 温银堂[1,2] ZHANG Yu-yan;REN Teng-fei;WEN Yin-tang(School of Electrical Engineering,Yanshan University,Qinhuangdao,Hebei 066004,China;Hebei Province Key Laboratory of Measuring and Testing Technologies and Instruments,Yanshan University,Qinhuangdao,Hebei 066004,China)

机构地区：[1]燕山大学电气工程学院,河北秦皇岛066004 [2]燕山大学测试计量技术及仪器河北省重点实验室,河北秦皇岛066004

出　　处：《计量学报》2022年第1期7-13,共7页Acta Metrologica Sinica

基　　金：河北省科技计划项目(20310401D,20312202D,216Z1704G)。

摘　　要：针对3D打印点阵结构中缺陷目标因尺寸小、缺陷特征微弱而难以准确自动识别的问题,提出了一种基于YOLOv3算法的点阵结构缺陷智能识别新方法。该方法利用深度学习网络模型在特征提取方面的优势,采用多尺度网络进行预测,将缺陷的分类和定位问题作为回归问题处理。实验结果表明,所提算法实现了一种3D打印点阵结构内部典型缺陷的识别,缺陷检测回召率为96.6%,准确率为93.2%,模型平均精度均值为0.957,为进一步精确表征缺陷并分析缺陷对点阵结构性能的影响提供了依据。To solve the problem that the defect in the lattice structure is difficult to accurately identify due to the small size and weak feature,an intelligent defect recognition method based on YOLOv3 algorithm is proposed.This method takes advantage of the deep learning network model in feature extraction,uses a multi-scale network to predict and treats the classification and location of defects as regression problems.The proposed algorithm realizes the identification of internal defects in a 3D printed lattice structure.And the detect recall is 96.6%,the accuracy is 93.2%,and the mean average precision value of model is 0.957.It provides a basis for further accurate characterization of defects and analysis of the effects of defects on the performance of lattice structures.

关 键 词：计量学 3D打印 缺陷检测 YOLOv3 点阵结构 深度学习 CT图像 

分 类 号：TB96[机械工程—光学工程] TG115.28[一般工业技术—计量学]