基于改进深度学习算法的铸件缺陷自动检测识别研究  

Research on Automatic Detection and Recognition of Casting Defects Based on Improved Deep Learning Algorithm

作　　者：王飞 张素兰[1] WANG Fei;ZHANG Sulan(College of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030024,China)

机构地区：[1]太原科技大学计算机科学与技术学院,太原030024

出　　处：《自动化与仪表》2023年第9期82-86,共5页Automation & Instrumentation

基　　金：山西省自然科学基金项目(202103021224285);太原科技大学教学改革项目(JG2021101)。

摘　　要：为改善多目标缺陷识别漏检问题,提高铸件缺陷检测精度,该文提出了基于改进深度学习算法的铸件缺陷自动检测识别方法。利用数字式辐射成像技术获取铸件DR图像,并采用引导滤波算法对其作平滑处理,在YOLOv3网络结构基础上,引入空间金字塔池化(SPP)结构,将优化后的YOLOv3网络与Faster RCNN、Cascade RCNN网络融合构建缺陷检测融合模型,将处理后的铸件DR图像作为模型输入,实现铸件缺陷的高精度识别。实验结果表明,该方法可识别铸件DR图像多个缺陷目标,平均检测精度达到97.5%以上,有效降低漏检缺陷数量。In order to improve the missing problem of multi-objective defect identification and improve the accuracy of casting defect detection,an automatic casting defect detection method based on improved deep learning algorithm was proposed.Digital radiation imaging technology is used to obtain casting DR images,and guided filtering algorithm is used to smooth them.Based on YOLOv3 network structure,space pyramid pool(SPP)structure is introduced.The optimized YOLOv3 network was fused with Faster RCNN and Cascade RCNN networks to build a defect detection fusion model,and the processed casting DR image was used as the model input to achieve high-precision casting defect identification.The experimental results show that the method can identify multiple defect targets in DR images of castings,and the average detection accuracy is more than 97.5%,which effectively reduces the number of missed defects.

关 键 词：深度学习算法 铸件缺陷 数字式辐射成像 引导滤波 YOLOv3网络 损失函数 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TG115.28[自动化与计算机技术—控制科学与工程]