Laser Stripe Detection And System Development Of Steel Weld Seam Based On Deep Learning

Welding laser stripe detection is a key technology in automatic tracking system of steel welding,which has important research and application significance for automatic welding.In the actual welding process,there are strong dazzling light,strong arc light and splash interference,so that there is a lot of noise in the weld image collected by the vision sensor.At present,the traditional image processing algorithms(such as canny edge detection,threshold segmentation,etc.)have low accuracy and poor robustness,which is difficult to accurately extract the weld in the actual welding scene,and can not be applied to the automatic tracking of the weld track of steel parts.Therefore,aiming at the problem of high noise in the images of steel welds,this paper proposes a method based on deep learning to overcome the interference of strong glare,arc and splash and solve the problem of laser stripe detection of steel welds.The main work is as follows:(1)A laser stripe detection method for welding seams based on U-Net is proposed.The U-Net network model is improved and BatchNorm and Relu functions is added after convolution and deconvolution.The main tasks include designing network structure,building data sets,model training,and qualitative and quantitative analysis of experimental results.(2)Based on D-LinkNet network,an aggregation network model of hybrid extended convolution combined with convolutional attention module is proposed.The experimental results show that the improved D-LinkNet model has further improved the laser stripe detection accuracy of welding seams.(3)In order to better serve the application of automatic welding tracking and verify the reliability of the method proposed in this paper,a set of steel welding laser stripe detection experiment platform is designed and implemented,which mainly includes four functional modules,including data processing,welding laser stripe detection,centerline processing and manual correction.The deep learning-based method proposed in this paper can accurately detect weld laser stripe from weld images with strong noise interference,so as to meet the needs of automatic welding production of steel parts.