基于视觉技术和Mask R-CNN的法兰盘表面缺陷检测研究

Research on Surface Defects Detection of Flange Based on Vision Technology and Mask R-CNN Networks

在法兰表面缺陷检测任务中,为了对缺陷进行精准定位和分类,根据缺陷位置和类别综合判定法兰是否符合特定标准的质量要求,结合缺陷特征和相应标准,提出一种基于视觉技术和Mask R-CNN的法兰表面缺陷检测与评估方法。依据法兰的适用原则和缺陷判定标准对法兰表面进行区域划分。通过搭建图像采集平台,采集图像并对其进行预处理操作后添加至网络训练集中。采用Mask R-CNN作为缺陷检测网络的基本框架,结合法兰表面缺陷特点改进Mask R-CNN骨干网络和颈部网络,并对网络性能进行验证。最后,根据检测标准,使用边缘检测算法对模型检测结果进行复检。结果表明:改进后的方法能够实现精确的定位并进行质量评估,满足法兰表面缺陷检测的要求。

In the task of flange surface defect detection,in order to accurately locate and classify the defect,and comprehensively determine whether the flange meets the quality requirements of the specific standard according to the defect location and category,taking the characteristics of the defects and corresponding standards into consideration,a method of flange surface defect detection and evaluation based on Mask R-CNN and vision technology was proposed.The flange surface region was divided according to the applicable principles and defect criteria of the flange.An image acquisition platform was constructed to capture images,the images were collected and pre-processed,then added to the network training set.Mask R-CNN was used as the basic framework of the defect detection network,and the backbone network and neck network of Mask R-CNN network were modified by considering the characteristics of surface defects,and the network performance was validated.Ultimately,according to quality criteria,the detection results of the model were reexamined using edge detection algorithms for the flange contour information extraction.The results show that the improved method can realize precise localization and quality evaluation,and satisfy the requirements of flange surface defect detection.