Research On Image Recognition Method Of In-service Pipeline Corrosion Fault

Steam injection pipeline used by oilfield is one of principal equipments of steam injection system concerning thick oil steam process technology, Because the media of internal transport is the steam under high temperature and high pressure, certain hole volume defects in internal and external surface of steam injection pipeline can be easily caused by oxidation and corrosion, which will lead to some major accidents.This paper, based on the practical demands of in-service pipeline detection, applied Matlab to compiling analytical and recognizable system of X-ray digital corrosion defective image, including image preprocessing,edge extraction,pattern recognition. The categories of and Weld pipeline defects, quantitative recognition was achieved through the building of interface by Visual Basic.In the light of some drawbacks such as low image SNR and poor contrast when processing image, mathematical morphology wavelet denoising innovative ways, on the basis of separate suspicious points, was accomplished; smoothing and sharpening of image was improved; meanwhile, the suspicious points identified as defects were combined and output defects by searching those regions; edge extraction of defective aspects was carried out using B bar wavelet multi-measure local modulus maximum, which was verified to be superior to other common operators; some key parameters in favor of defects identification like fine length,moment,gray features and acute angle were selected to recognize patterns by means of single-output BP neural network. As demonstrated, the correct rate of unknown defects recognition was higher than three-layer network of multi-output node. In addition, the BP network was also utilized to gain a precise quantitative identification of defects by making use of these advantages shown during the process of processing multi-parameter and nonlinear issues. We established identification network so as to increase the accuracy of quantitative identification for each defect respectively and choose features rigidly.Because engineering physically medium of piping corrosion under the influence of various factor, shape complications anomaly, precision fixed amount identify very difficult, so give anomaly decay blemish the biggest depth and average circle as to it's diameter and depth.On a PPR pipeline with a specification:75 mm in diameter and 11 mm in width, stimulative hole defects and bar,crack defects were produced to be used in training samples and non-training samples detective trials. After analysis, Correct identification rate of 90%, it was found that the absolute deviation was under 0.01 mm when BP neural network identifies training samples, however, as for non-training samples, the absolute deviation was under 0.02 mm regardless of the impact of processing errors, with a high accuracy. We can better replace artificial intelligence with computer intelligence and automatically judge the detective outcome through the detection and recognition of X-ray digital defective image by automatically analytical and recognizable system.