基于AlexNet的焊缝缺陷分类方法

Weld defect classification method based on AlexNet

X射线检测作为一种实用的无损检测(NDT)方法,在压力容器的焊缝缺陷检测中得到了广泛的应用。基于X射线图像的自动缺陷识别技术也随着人工智能(AI)的发展取得了飞速发展。本文将焊缝缺陷裁剪成像素小块作为神经网络的输入,并在AlexNet的基础上通过添加BN层改进了原网络,而后又选取了最优α值的LeakyReLU层代替了原有的ReLU层,使最终的AlexNet-BN-L模型取得了高达88.80%的5折交叉验证平均准确率。

As a practical non-destructive testing(NDT)method,X-ray testing has been widely applied in the detection of weld defects in pressure vessels.Automatic weld defect recognition technology based on X-ray film has also made great progress with the development of AI.In this research,the weld defect is cut into small pixel patches as the input of neural network.The original AlexNet is improved by adding BN layer,and the LeakyReLU layer with optimalαvalue replaces the original ReLU layer.The final AlexNet-BN-L model achieves the highest 5-fold cross validation average accuracy rate of 88.80%.