Research On Pipeline Circle Image Positioning Algorithm Of VVER Vortex Thermal Conductivity Pipe System

With the popularity of digital cameras and the continuous advancement of computer technology,the study of digital image processing is increasingly applied in industrial scenes to solve some practical problems in industrial field.The practical application background of this thesis is that the nuclear power operation system needs to perform regular flaws detection maintenance on the pipe wall of the VVER(water-water energy reactor)vortex thermal conductivity pipe.This requires the steering gear to control the flaws detector to protrude into the thermal conductivity pipe to perform flaws detection,but because of the accuracy of the controlling algorithm for steering gear operation,the detector’s position may be offset from the target pipe port.Therefore,digital image processing technology is needed to obtain the horizontal and vertical coordinate offset of the detector and the target pipe port,thereby realizing the pipeline circle positioning of the VVER vortex thermal conductivity pipe,so that the detector can be completely aligned with the target pipe port and protrude into the pipe for pipe wall’s flaws detection.Threshold segmentation,pipeline circles edge extraction,reference key circle recognition and screening of the image positioning algorithm are deeply studied in this thesis as follows:Firstly,in order to improve the segmentation ability of the segmentation algorithm for images with small gray scale differences between the segmentation regions,an improved minimum cross entropy pipeline circles segmentation algorithm according to edge information is proposed.The basic principle of this improved algorithm is to increase the frequency of the edge grayscale range to change the grayscale distribution of the image,and then the minimum cross entropy algorithm is applied to determine the grayscale threshold for the binary segmentation of image.The improved algorithm improves the average inter-region gray scale contrast(GC)of the segmented image of 100 sample images by 4.85% compared with the traditional minimum cross entropy segmentation method;At the same time,for the segmentation of the Lena image with more Gaussian noise,the grayscale contrast(GC)reduction of the improved algorithm caused by the noise is 2.88% less than the traditional minimum cross entropy segmentation method,so the anti-noise ability is more strong;And the reference key circle recognition rate of the improved method applied to the VVER pipeline circles positioning task is 1.41% higher than the minimum cross entropy method.Then,in order to solve the problem that the determination of the hysteresis threshold of the traditional Canny edge extraction is not self-adaptive,an algorithm is proposed in this thesis to realize the self-adaptability of hysteresis threshold of the pipeline circles edge extraction based on gradient amplitude distribution.Cross entropy is applied to measure the information distance between the gradient amplitude distribution of the image before and after classified into two types of regions: edge and background.And then this algorithm aims to search for the gradient amplitude classifying threshold by making the cross entropy of the gradient amplitude distribution before and after the image is classified into the edge and background regions minimum,which can be used as the hysteresis threshold of Canny edge extraction.The average image structure similarity(SSIM)of 100 sample binary images before and after edge extraction based on this improved algorithm is increased by 0.64% compared with the traditional Canny operator,so this improved edge extraction algorithm has similar performance in image structure similarity to traditional Canny operator,and at the same time,the self-adaptability of the hysteresis threshold determination is achieved in this improved algorithm.Finally,the reference key circle recognition and screening is deeply studied,and a key circle recognition and screening algorithm based on the constraint region with the reference point is proposed.The algorithm firstly performs 2-1 Hough circle detection on the obtained edge image,and then designs a pseudo arc elimination method to eliminate the pseudo circles in the detected Hough circle.The reference key circle is screened according to whether it is within the constraint circle domain firstly and then whether it is closest to the reference template circle center.At the same time,the algorithm designs the solution of the upper and lower boundary conditions of the network of vortex thermal conductivity pipe,and solves the problem of pipe circle positioning of the sample image of the upper and lower boundary conditions.Finally,the algorithm is applied in the VVER vortex thermal conductivity pipe circle positioning algorithm,and it is used to test the sample image data set.The algorithm achieves a 99.33% reference key circle recognition rate on the 1638 sample images dataset and it plays an auxiliary role in the target pipeline circle flaws detection on the nuclear power operation system experimental platform.