基于改进Faster R-CNN的铝型材表面缺陷检测研究

Research on surface defect detection of aluminum profiles based on improved Faster R-CNN

针对传统目标检测算法在铝型材表面缺陷检测中检测精度低、微小缺陷难以识别等问题,提出一种改进的Faster R-CNN算法。该算法对特征提取网络、感兴趣区域池化和锚框尺寸进行了优化,并在此基础上引入了特征金字塔和可变形卷积,以提升检测精度。实验表明,改进算法的平均精确度均值可达到86.73%,相较于原算法提升了8.70%,其对微小缺陷识别效果好,满足了工业上对缺陷检测的需求。

Aiming at the issues of low detection accuracy and difficulty in identifying small defects on the surface of aluminum profiles using traditional object detection algorithms,an improved Faster R-CNN algorithm is proposed.It optimizes the feature extraction network,region of interest pooling and anchor box size,and then introduces feature pyramid network and deformable convolution to improve the detection accuracy.Experimental results show that the improved algorithm achieves a mean average precision of 86.73%,which is 8.70%higher than the original algorithm,and performs well in identifying small defects,meeting the industrial requirements for defect detection.