基于Deeplab-V3的焊缝缺陷检测应用研究

Study on weld defect detection method based on Deeplad-V3

首先,采用三维激光线扫相机采集焊接部位的三维点云数据投影至3个平面获取深度图,更好地将焊缝缺陷呈现出来;其次,采用深度图制作训练数据集,利用循环一致性对抗网络对样本数据进行扩充,并基于该数据集提出两种焊缝缺陷检测方法,分别是基于特征和基于滑动窗口的Deeplab-V3模型深度学习焊缝缺陷检测方法;再次,采用AHP(层次分析法,Analytic hierarchy process)和FCE(模糊综合评价方法,Fuzzy comprehensive evaluation)系统评价方法对提出的两种焊缝缺陷检测方法进行评价,得出基于深度学习的焊缝缺陷检测方法在检测效果上更好;最后,开发了焊缝缺陷检测系统,进行测试和校验,验证了基于Deeplab-V3模型的焊缝缺陷检测模型的可行性和有效性。

The 3D point cloud data of the welding part collected by the 3D laser line scanning camera are projected onto three planes to obtain the depth map,and the weld defects can be better presented.The training data set is developed based on the depth map,and it is expanded by the cyclic consistency adversarial network.Two weld defect detection methods are proposed based on the training data set,which are feature-based method and sliding window deeplab-v3 model based deep learning weld defect detection method.The AHP(Analytic hierarchy process)and FCE(Fuzzy comprehensive evaluation)system evaluation methods are used to evaluate the two proposed weld defect detection methods.It can be concluded that the weld defect detection method based on deep learning has better detection performance.Finally,a weld defect detection system is devoleped for testing and verification.The feasibility and effectiveness of the weld defect detection model based on Deeplad-V3 model are verified by the experimental results.