A weld seam feature extraction method, device and electronic equipment
By combining generative adversarial networks and deep convolutional neural networks, a high-quality labeled dataset is generated, which solves the problem of noise interference in the welding environment, achieves efficient weld feature extraction, adapts to different welding scenarios, and improves the level of welding intelligence.
CN116229087BActive Publication Date: 2026-06-19SHANGHAI JIAOTONG UNIV
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI JIAOTONG UNIV
- Filing Date
- 2023-03-13
- Publication Date
- 2026-06-19
Smart Images
Figure CN116229087B_ABST
Abstract
This invention discloses a method, apparatus, and electronic device for weld seam feature extraction. The method includes: step S1, designing a generative adversarial network (GAN) for generating realistic welding images; step S2, designing a deep convolutional neural network (DNN) for weld seam feature extraction; step S3, acquiring welding images, weld seam images, and laser stripe images to obtain a first dataset; step S4, inputting the first dataset into the GAN for training to obtain several realistic welding images; step S5, processing the laser stripe images generated by the GAN using image processing methods to obtain corresponding sub-pixel level annotations, obtaining a batch of annotated realistic welding images to form a second dataset; step S6, inputting the annotated realistic welding images from the second dataset into the DNN for training; and step S7, inputting real welding images into the trained DNN model to obtain weld seam features.
Need to check novelty before this filing date? Find Prior Art