Welding Image Edge Detection and Identification Research Based on Canny Operator

Abstract

Edge is the most important feature of welding image. The classical algorithm of edge extraction is that the gray change of each pixel in neighborhood is considered and the change rule of first or second derivative in edge adjacent is used. In several edge detection operators commonly used, Lap lace operator often produces double-margin and others such as Sobel operator often form non-closed area. In this paper, Canny edge detection operator based on optimization algorithm is firstly discussed, which has advantages of large signal and noise ratio and high accuracy. Secondly, the welding image processing with Canny operator is described and for every step, detailed description and processing results are given. At last, Canny operator edge detection results are projected to 2-d histogram. The histograms of projection results are compared with the qualified image histogram. When the similarity is higher than some threshold, we think that the welding is qualified. Otherwise, it is considered as defective welding. Through the edge extraction and detection of lead-acid battery electrode welding that are done by welding robot, it can be seen that the recognition results meet the requirements of production line.