Research On Underwater Weld Image Recognition Technology

With the reduction of land resources,the development of marine resources has received extensive attention at home and abroad,and the research of marine engineering technology has developed vigorously.As a key technology of marine engineering,underwater welding is used in welding methods,welding materials and welding automation.Important results have also been achieved.The underwater welding seam image recognition technology is the core of the realization of underwater welding automation,and its research is of great significance.This article takes the underwater weld image taken by an industrial camera in the laboratory as the research object,analyzes the characteristics of the collected weld image,and conducts a more systematic research on the enhancement,denoising,segmentation and center extraction of the underwater weld image.Based on this,a set of underwater weld image recognition application system is designed using Visual Studio and Open CV.The main contents of the paper are as follows:(1)Aiming at image quality problems such as low contrast of underwater weld images,and insignificant differences between weld edge features and background.By improving the traditional gray-scale transformation algorithm,design an S-curve gray-scale transformation function to stretch the gray-scale range of the underwater weld image,so as to achieve the purpose of improving the contrast of the underwater weld image.(2)Aiming at the interference problem of underwater weld image noise,through analyzing the characteristics of impurity noise,a denoising algorithm combining closed operation and median filtering is designed to finally achieve the purpose of removing most of the impurity noise,and in the process of denoising Will not blur the edges of the weld.(3)Aiming at the problem that the edge features of the four types of underwater weld images are difficult to segment,this paper proposes an underwater weld image segmentation technique that is classified and then segmented.First,use convolutional neural networks to classify various types of underwater welds;then,study the characteristics of each type of weld,design a specific image segmentation algorithm,and perform rough segmentation on underwater weld images;finally,pass Hough line transformation removes the characteristic pixels that are not the edges of the weld.The segmentation algorithm uses Tsallis entropy to extract the segmentation threshold.The algorithm only needs to adjust the gray-level correlation parameters to achieve accurate segmentation of different underwater weld types,thereby reducing the complexity of the image segmentation algorithm.In addition,particle swarm algorithm is added to optimize the image segmentation algorithm and improve the operating efficiency of the algorithm.(4)Aiming at the problem that the edge features of the welded seam obtained by segmentation are not single pixels and incomplete,they cannot guide the welding manipulator to weld.First,compare and analyze different thinning algorithms,find a suitable underwater weld image thinning algorithm,and refine the weld edge features obtained by segmentation into single-pixel feature points.Second,the refined single-pixel feature points are fitted into a complete weld centerline through least squares.And the comparative analysis adopts the multi-segment straight line fitting and polynomial fitting curve of the least square fitting idea,the result shows that the effect of multi-segment straight line fitting and polynomial fitting curve is almost the same,and the running time of multi-segment straight line fitting curve program is shorter,and welding The centerline of the seam also has a straight line fitting,so this paper selects multiple straight lines to fit the curve of the centerline of the weld.(5)Design the human-computer interaction interface for underwater weld image recognition,and modularly package the underwater weld image recognition algorithms with different functions,which is convenient for the maintenance of the later code and the change of the algorithm.At the same time,different function interfaces can be called according to different image processing procedures to realize corresponding processing functions.The test results show that the underwater weld recognition system has an accuracy of more than 99% in the classification of weld types,and can accurately identify and extract the weld center lines of the four underwater weld images,and the mean square error of the fitting curve is less than 0.5.