摘要
针对现有航空发动机制件缺陷检测所存在的检测准确率低、速度慢等问题,提出一种基于改进Faster R-CNN算法的缺陷检测方法。该算法使用深度残差网络提取缺陷特征,采用含有内容感知重组的特征金字塔模型融合各层次特征图,并根据检测框尺度选取相应层次的特征图进行检测和识别,在RCNN部分使用分层采样实现挖掘难例,增强模型对难例样本的学习。实验结果表明:所提算法具有较高的检测准确率,而且能够有效提升检测速度。
Aiming at the problems of low detection accuracy and inefficient detection in defect detection of aeroengine parts, an algorithm based on improved Faster R-CNN was proposed.In this algorithm, deep residual network was used to extract features of defects.The pyramid model with content-aware reassembly of features(CARAFE) was used to integrate the feature maps of each level, and the corresponding level feature maps were selected for detection and recognition according to the scale of anchor boxes.In the part of RCNN,stratified sampling was used to mine hard samples to enhance its learning.The experimental results show that the proposed algorithm possesses higher detection accuracy and the detection speed is improved effectively.
作者
唐嘉鸿
黄颀
田春岐
TANG Jiahong;HUANG Qi;TIAN Chunqi(AECC Shanghai Commercial Aircraft Engine Manufacturing Co.,Ltd.,Shanghai 201306,China;College of Electronics and Information Engineering,Tongji University,Shanghai 201804,China)
出处
《机床与液压》
北大核心
2022年第23期93-98,共6页
Machine Tool & Hydraulics
基金
上海市工业互联网资助项目(2018-GYHLW-02043)
上海市人工智能产学研专项资助项目(PKX2020-R13)
上海市信息化发展专项资金项目(201901010)。
