摘要
针对焊缝X射线图像缺陷识别传统方法的计算量大与准确度差的问题,提出了基于MobileNet的识别方法。首先对样本图像进行预处理和数量上的增强;然后引入MobileNet结构以解决传统深度卷积神经网络中对计算资源要求高的问题,引入残差结构与ELU激活函数以解决原始MobileNet网络中出现的退化问题与权重偏置更新失效的问题,在训练时应用迁移学习方法,解决小数据集容易过拟合与训练效率低的问题;最后,针对相同数据集,与改进前的网络、AlexNet网络和VGG-16网络进行对比,表明该文方法具备更优的识别准确率和相比传统网络拥有更小的计算量,相比传统网络的缺陷识别方法拥有更大的应用范围。
Aiming at the large amount of calculation and poor accuracy of the traditional method of welding seam Xray image defect recognition,a recognition method based on MobileNet is proposed. First,the sample images are preprocessed and quantitatively enhanced;then the MobileNet structure is introduced to solve the problem of high computational resources in the traditional deep convolutional neural network,and the residual structure and ELU activation function are introduced to solve the original MobileNet network. The degradation problem and the problem of weight bias update failure,apply the transfer learning method during training to solve the problem of small data set easy to overfit and low training efficiency. Finally,for the same data set,compared with the network before the improvement,the AlexNet network and the VGG-16 network,it shows that the method in this paper has a better recognition accuracy and a smaller amount of calculation than the traditional network. The defect identification method has a wider range of applications.
作者
陈艳菲
彭洪晟
王俊涛
王洪伟
CHEN Yan-fei;PENG Hong-sheng;WANG Jun-tao;WANG Hong-wei(School of Electrical Information Engineering,Wuhan Institute of Technology,Wuhan 430073,China;Research Institute of Nuclear Power Operation,Wuhan 430223,China)
出处
《自动化与仪表》
2022年第1期49-54,共6页
Automation & Instrumentation
基金
湖北省教育厅科学研究计划重点项目(D20171502)
智能机器人湖北省重点实验室开发基金项目(HBIR201706)
武汉工程大学科学研究基金项目(K201810)。
