- 期刊
Research on Motion Deblurring Algorithm of Weld Image based on Improved DeblurGAN
摘要
In the existing welding robot workstation operation process, currently all use the visual positioning method, but in the actual operation process, due to camera shake and field environmental factors, will lead to the image blur phenomenon, affect the accuracy of the algorithm positioning and welding efficiency. Therefore, it is necessary to preprocess blurred images. In previous studies, image deblurring is based on traditional visual restoration methods, and there is no research on blurred image processing of welding seams. To solve this problem, this paper proposes an improved DeblurGAN method to preprocess blurred images of weld movement. In this method, Inverted residual blocks are used instead of Resnet blocks to make the model more lightweight. Finally, experimental tests are carried out on the weld motion fuzzy image data set constructed by ourselves. The results show that the proposed method has higher speed and robustness than the original DeblurGAN method in processing weld motion fuzzy images.