基于深度学习的不锈钢棒材表面螺纹缺陷检测

Detection of Thread Defects on Stainless Steel Bar Surface Based on Deep Learning

不锈钢棒材表面的螺纹是棒材磨制过程中造成的缺陷,严重影响棒材的验收与后续使用,目前针对该类缺陷多采用人眼观察、手指感知等人工方式进行判断,漏检率较高,且螺纹缺陷的图像极具多样性,传统的特征提取方法不能很好地表征螺纹,检测率较低,无法满足工业现场的需求。据此,采用深度网络对螺纹进行检测,并建立大型不锈钢棒材图像的螺纹缺陷数据集,采用迁移学习的方法对螺纹图像进行训练,得到分类器。实验结果表明,文章提出的深度学习方法在保持较高检测速度的同时,有效提升了检测螺纹缺陷的正确率。

The thread on the surface of stainless steel bar is a defect brought by the bar grinding process,which seriously affects the acceptance check and subsequent use of the bar.However,at present,such defects are mostly judged by manual methods such as binocular observation and finger perception,lea-ding to the high missed detection rate.Besides,the images of thread defects are very diverse.The traditional feature extraction methods can not characterize the thread well,and the detection rate is not high,which cannot meet the needs of industrial field.Therefore,the depth network is used to detect the thread,and a thread defect data set of large stainless steel bar images is established.The thread image is then trained by migration learning method to obtain the classifier.The experimental results show that the proposed deep learning method not only maintains a high detection speed,but also effectively improves the detection accuracy of thread defects.