Feature Extraction Method Research Of Pipeline Defect Based On Image Analysis

As the most important sources of energy, oil and gas resources are transported by road, rail, water, air and pipeline. In these five ways, pipeline transportation is the main method. With the development of the oilfield in our country, the position of oil and gas pipeline transportation in the national economic development of China is becoming more and more significant. With the increasing of oil delivery time, oil and gas pipelines are prone to be damaged and have defects, and oil spill occurs. This not only affects the normal production of the oil fields, but also can cause environmental pollution of the ground, and cause ecological disaster. So the safety problem of the pipeline transportation has aroused widely attention in the society. At the same time, feature extraction of pipeline defect has been put on the agenda gradually.MFL detect technology is one of the most reliable method of detecting pipeline for the moment; it is an important safeguard for pipeline transportation safety. At the same time, image processing has been a hot issue by research scholars in recent years. With the technological advance on MFL detect and image processing, at present, feature extraction method research on pipeline defect contains some aspects as follows:the graphical conversion method on the MFL curve, filtering pretreatment on image matrix, image segmentation, feature information extraction, and so on. According to this situation, to determine the security status of the pipeline which has important practical significance on reducing the state economic losses、protecting the natural environment and safety of personal property.The paper contains four main aspects:firstly, analyze and summarize the development status of MFL detect technology and image processing technology; analyze the feature of pipeline defect and MFL Signal. Secondly, design MFL detect system. The first step, design the underlying acquisition circuit and the top-level control circuit. The second step, use verilog language to fulfill the data acquisition system. Thirdly, transform the MFL data to image. Make use of the improved image preprocessing algorithm and the improved image segmentation algorithm to extract the defect boundary. Make use of the edge detection to get area, perimeter, gradient, major and minor axes of the gray image defect. Fourthly, regard the feature information and the length, width, deep of the corresponding defect as the training samples. Build neural networks on defect size to train the samples. At last, take advantage of the defects feature in formation on gray image to predict the 1 ength, width, dee p of the unknown defect.