基于改进Faster R-CNN的轴承表面缺陷检测

Bearing Surface Defect Detection Based on Improved Faster R-CNN

针对轴承表面缺陷小、几何形状多变以及低对比度的特点,提出了一种改进的Faster R-CNN算法,对轴承表面缺陷进行检测。首先,以ResNet-50结合特征金字塔网络对轴承表面缺陷进行特征提取;其次,在改进的特征提取网络中引入可变形卷积,通过卷积学习偏移量自适应调整感受野,提高了缺陷的提取能力;最后,针对ROI Pooling因二次量化而导致的区域不匹配问题,采用基于双线插值的ROI Align改进ROI Pooling。实验结果表明,在采集的轴承表面缺陷数据集上,改进的Faster R-CNN平均精度均值为97.6%,与改进前相比,提高了11.76%,可以实现对轴承表面各类缺陷更为准确的检测,具有较强的实用性。

An improved Faster R-CNN algorithm is proposed for the detection of bearing surface defects with small size,variable geometry and low contrast.Firstly,ResNet-50 combined with feature pyramid network is used for feature extraction of bearing surface defects;secondly,the deformable convolutional is introduced into the improved feature extraction network,and the perceptual field is adaptively adjusted by the convolutional learning offsets to improve the extraction capability of defects;finally,to address the problem of region mismatch caused by quadratic quantization of ROI Pooling,a bilinear interpolation-based ROI Align is used to improve ROI Pooling.The experimental results show that the mean average precision of the improved Faster R-CNN was 97.6%on the collected bearing surface defect dataset,which is an improvement of 11.76%compared with that before the improvement,and can achieve more accurate detection of various types of defects on the bearing surface,which has strong practicality.