Research On Welding Seam Detection And Identification Under Industrial Dust Condition

The vigorous development of computer technology promotes research to the application stage,and more production links realize automation under the support of high technology.In the actual grinding and polishing workshop,the image quality of the welding seam is easily affected by the adverse industrial conditions,such as dust in the air and uneven light,which will lead to the blurred image of the welding seam and the serious loss of the contour information of the welding seam.The boundary between the weld and base material zone is not obvious,which causes strong interference to the identification and detection of weld targets,resulting in the failure of detection accuracy and speed to meet industrial requirements.Therefore,it has become a hot research topic to promote the welding seam grinding and polishing industry to realize intelligent and more efficient welding seam target identification and detection technology by automation.In this paper,based on atmospheric scattering theory and dark channel prior theory,the formation principle of dust-bearing weld image was analyzed,and the image data set with higher definition was obtained by optimizing image restoration algorithm.Then,by analyzing the mainstream target detection algorithm,the YOLOv3 target detection framework,which is fast in detection but needs to be improved in accuracy,is selected to preliminarily realize weld target detection through data enhancement and optimization clustering analysis and position loss function.Then,feature extraction module and attention mechanism of enhanced receptive field are added to further improve the accuracy and integrity of weld target detection.Finally,the effectiveness and superiority of the algorithm in this subject are verified through a number of experiments,so as to build a high-precision welding seam identification and detection algorithm system.The main content and innovation points of this paper are as follow.First,based on the atmospheric scattering model and the dark channel prior theory,the hypothesis of the bright channel is deduced,and an algorithm of weld image restoration based on the fusion of the bright channel and the weighted guided dark channel is proposed,which can effectively improve the weld image restoration quality.Second,the robustness of the target detection model and its detection and recognition accuracy are effectively improved through data enhancement,K-means++ clustering analysis and optimization of position regression loss function.The last,by embedding the feature extraction module ASPP based on cavity convolution and the attention mechanism based on mixed domain CBAM into YOLOv3,the contradiction between high-resolution input and receptive field was alleviated.Finally,the detection and recognition accuracy of the optimized model reached 90%,and the average accuracy was improved by 20.35%compared with the original network.