摘要
为了提高缺陷识别效率,提出利用基于深度学习网络进行焊缝缺陷识别。在分析X射线焊缝缺陷图像特征的基础上,构建一种基于模拟视觉感知原理的深度学习网络结构,并对卷积神经网络的卷积模板大小及层数进行了分析,对卷积神经网络隐藏层中2种不同的激活函数进行了实验验证,针对性地提出优化方法。该深度学习神经网络可以避免对焊缝缺陷图像特征的提取,直接判断疑似缺陷图像是否为缺陷。对580张图像进行了实验,结果表明,本文所提方法对SDR图像的识别准确率超过98%,优于传统方法。且所设计系统具有自动学习X射线焊缝缺陷图像中复杂的深度特征的特点,实用性较强。
In order to improve the efficiency of defect recognition,the recognition of X-ray weld defects based on deep learning is proposed.A deep learning network structure based on the principle of simulated visual perception is established based on the analysis of the characteristics of X-ray weld defect images.The convolution template size and layernumber of convolutional neural network are analyzed.Two different activation functions in the hidden layer of convolutional neural network are verified,and an optimization method for them is proposed.To use the deep learning neural network can avoid the step of extracting the features of the weld defect images and directly determine whether the suspected defect image is a defect.Experiments on 580 images show that the recognitionaccuracy of the proposed method to SDR images is over 98%,which shows that the proposed weld defect recognition method based on deep learning network is better than the traditional method.And the designed system has the ability of automatically learning the complex features in X-ray weld defect images,and it has strong practicability.
作者
李清格
高炜欣
LI Qingge;GAO Weixin(Key Laboratory of Shaanxi Province for Measurement and Control Technology for Oil and Gas Wells,Xi'an Shiyou University,Xi'an 710065,Shaanxi,China;Key Laboratory of MOE for Photoelectric Oil and Gas Logging and Detection,Xi'an Shiyou University,Xi'an 710065,Shaanxi,China)
出处
《西安石油大学学报（自然科学版）》
CAS
北大核心
2019年第4期74-81,共8页
Journal of Xi’an Shiyou University（Natural Science Edition）
基金
陕西省教育厅重点实验室科研计划项目(14JS079)
西安石油大学研究生创新与实践能力培养项目(YCS18213082)
