基于Mask RCNN的绝缘子自爆缺陷检测
    点此下载全文
引用本文:汪 琦,刘向阳.基于Mask RCNN的绝缘子自爆缺陷检测[J].计算技术与自动化,2022,(1):52-58
摘要点击次数: 356
全文下载次数: 0
作者单位
汪 琦,刘向阳 (河海大学 理学院江苏 南京 211100) 
中文摘要:针对绝缘子自爆缺陷位置检测问题,提出了一种基于Mask RCNN的绝缘子自爆缺陷检测的方法。通过构造基于Mask RCNN的绝缘子串分割模型,在获取的掩模图像中引入最小外接矩形提取绝缘子串图像,从而搭建基于Mask RCNN的自爆缺陷检测模型,检测绝缘子串中的自爆位置。结合两个模型,将绝缘子串位置及其自爆缺陷位置映射到原图。该方法在绝缘子串分割模型的验证集上,平均Dice达到0.822,在自爆缺陷识别模型的验证集上,平均IOU达到0.835,最终模型对缺陷位置识别准确率达到94.12%。
中文关键词:绝缘子  深度学习  Mask RCNN  自爆缺陷
 
Detection of Self-explosion Defects of Insulators Based on Mask RCNN
Abstract:Aiming at the problem of detecting the position of insulator self-explosion defect, a method of insulator self-explosion defect detection based on Mask RCNN is proposed. By constructing a segmentation model of the insulator string based on Mask RCNN, the smallest bounding rectangle is introduced into the acquired mask image to extract the image of the insulator string, so as to build a self-explosion defect detection model based on Mask RCNN to detect the self-explosion position in the insulator string. Combining the two models, the location of the insulator string and its self-explosive defect location are mapped to the original image. This method has an average Dice of 0.822 on the validation set of the insulator string segmentation model, and an average IOU of 0.835 on the validation set of the self-explosive defect recognition model. The final model has an accuracy of 94.12% for defect location recognition.
keywords:insulator  deep learning  Mask RCNN  self-explosion defect
查看全文   查看/发表评论   下载pdf阅读器