计算机与现代化 ›› 2021, Vol. 0 ›› Issue (06): 35-40.

• 图像处理 • 上一篇    下一篇

深度学习在手机数据接口缺陷检测中的应用

  

  1. (江苏理工学院电气信息工程学院,江苏常州213001)
  • 出版日期:2021-07-05 发布日期:2021-07-05
  • 作者简介:刘亚东(1994—),男,河南周口人,硕士研究生,研究方向:深度学习,计算机视觉,E-mail: aden_l@126.com; 通信作者:罗印升(1964—),男,陕西武功人,教授,博士,研究方向:智能控制理论方法与应用,计算机视觉,模式识别,E-mail: dxlys@jsut.edu.cn; 曹阳阳(1995—),女,江苏南通人,硕士研究生,研究方向:图像处理,计算机视觉,E-mail: 1241078114@qq.com。
  • 基金资助:
    江苏省研究生实践创新计划项目(SJCX19_0691)

Application of Deep Learning in Defect Detection of Mobile Phone Data Interface

  1. (School of Electrical & Information Engineering, Jiangsu University of Technology, Changzhou 213001, China)
  • Online:2021-07-05 Published:2021-07-05

摘要: 为了能更好地对手机数据接口的缺陷进行检测,提出一种基于Faster R-CNN的检测算法。将Faster R-CNN检测架构中的RoIPooling替换成RoIAlign,解决RoIPooling计算过程中2次量化造成的目标回归位置的偏差。使用ResNet50融合FPN的网络作为特征提取网络,提高模型对小型目标缺陷的检测效果。最后使用测试集进行预测,实验表明本文提出算法的均值平均精度(mAP)达到了91.17%,比使用VGG为特征提取网络时的mAP提高了24.72个百分点,比单独使用ResNet50为特征提取网络的mAP提高了2.58个百分点,因此,本文提出的算法对手机数据接口缺陷检测有显著的效果提升,为手机数据接口缺陷检测提供了一种更有效的检测方法。

关键词: 深度学习, 残差网络, 特征提取, 手机数据接口缺陷检测, 特征金字塔网络

Abstract: In order to better detect the defects of the mobile phone data interface, this paper proposes a detection algorithm based on Faster R-CNN. The specific research method is to replace RoIPooling in the Faster R-CNN detection architecture with RoIAlign to solve the deviation of the target return position caused by the two quantifications in the RoIPooling calculation process. The ResNet50 fusion FPN network is used as a feature extraction network to improve the model’s detection effect on small target defects. Finally, the test set is used for prediction. Experiments show that the mean average accuracy (mAP) of the proposed algorithm in this paper has reached 91.17%, which is 24.72 percent points higher than mAP when VGG is used as the feature extraction network, and is 2.58 percent points higher than mAP when ResNet50 is used alone as the feature extraction network. Therefore, the algorithm proposed in this paper has a significant effect on mobile phone data interface defect detection, and provides a more effective detection method for mobile phone data interface defect detection.

Key words: deep learning, residual networks, feature extraction, mobile phone data interface defect detection, feature pyramid networks