Robust real-time weld joint feature detection method

Abstract

The invention discloses a robust real-time weld joint feature detection method. The method comprises the steps that S1, collecting a weld joint image in real time; s2, cutting the acquired image intoan image with a certain size by taking a specific value as a center to serve as network input; s3, processing the cut image by using a multilayer convolutional neural network to obtain high-level semantic features of the image and local information and global information of the image; s4, splicing the multi-layer features, using a convolutional neural network to realize fusion of local informationand global information, and obtaining high-layer fusion features; s5, processing the high-level fusion features by using a recurrent neural network, and learning and obtaining the high-level fusion features containing time context information; s6, generating a series of candidate boxes on the feature map; and S7, processing the obtained high-level fusion features containing the time context information by using a convolutional neural network, and obtaining the position and the category of the weld joint by combining the candidate box. The method has a stable weld joint detection result when alarge amount of noise exists in the image.

Description

technical field [0001] The invention relates to a welding track automatic detection system, mainly relates to a welding seam detection based on a laser vision sensor, and specifically relates to a robust real-time welding seam feature detection method. Background technique [0002] Robot welding improves production efficiency and prolongs production time while reducing production costs and bringing greater profits to enterprises. However, limited by the "teaching-reproduction" working mode, the existing robot welding methods do not have the ability to adapt to changes in the weld trajectory in real time, thus placing extremely high requirements on the positioning and clamping accuracy of the workpiece, and it is difficult to achieve the level of flexible manufacturing . The robot weld seam real-time detection system based on laser vision sensor realizes real-time and accurate detection of weld seam through laser vision sensor and intelligent weld seam detection algorithm, a...

AI Technical Summary

Problems solved by technology

Aiming at the problem that it is difficult to realize weld location in images of strong arc light and spatter noise by using traditional image processing methods during weld detection, and the performance of the weld detection system is severely restricted by this, the present invention proposes a robust real-time weld feature detection method , using deep learning methods to achieve accurate detection of weld features under strong noise

Images