Pipeline connector based on optimized YOLOv3 algorithm and defect accurate identification method

Abstract

The invention provides a pipeline connector based on an optimized YOLOv3 algorithm and a defect accurate identification method. The method comprises the following steps: collecting data; processing the data; obtaining a training set and a verification set; carrying out an optimization test on network model parameters, extracting picture feature values, and freezing a part of the training set to train the pipeline connector and the defect identification model; judging whether convergence occurs or not, judging whether a loss function is over-fitted or not, unfreezing data set optimization parameters, and continuing to train the model; verifying the final training model, calculating an intersection-to-union ratio of a pipeline connector and a defect area obtained by the model and an actual marking area, and drawing an average value curve of each category of precision values so as to analyze whether the model meets conditions or not; and identifying the connector and the defect by using the trained model. The detection result is suitable for internal detection of small-diameter pipelines which cannot be positioned by a GPS (Global Positioning System), such as underground deep-buried and indoor small-diameter pipelines, the pipeline connector can be effectively identified, the positioning precision is improved, meanwhile, the pipeline defect condition can be detected, the detection efficiency is improved, and the detection cost is saved.

Description

technical field [0001] The invention relates to a pipeline connector based on optimized YOLOv3 algorithm and defect image recognition technology to assist pipeline robot positioning and pipeline daily maintenance method, in particular to a pipeline connector based on optimized YOLOv3 algorithm and accurate defect identification method, which belongs to small-diameter pipelines In-house testing technology field. Background technique [0002] There are more than 3 million kilometers of pipelines buried underground and under the seabed in the world (more than 600,000 in China). Some of them are small-diameter pipelines. Explosion accidents occur frequently, which not only pollutes transported materials and causes waste of resources, but also damages the environment and causes ecological disasters. In addition, it will seriously endanger personal safety in densely populated areas and cause a great negative impact on society. Therefore, regular inspection and maintenance of the ...

AI Technical Summary

Problems solved by technology

[0004] A variety of sensor devices commonly used in the prior art are loaded on pipeline robots to detect pipelines. Due to the limited size and power consumption of sensor equipment in small-diameter pipelines, there is a lack of pipeline robots for small-diameter pipeline detection on the market. The present invention uses Vision-assisted positioning detection is suitable for small-diameter pipeline robots. Without increasing the number of sensors and changing the overall volume of the small robot, it can improve the accuracy of pipeline detection and solve the problems of low internal detection accuracy, high cost and cumbersome detection work in small-diameter pipelines.

[0005] At present, no introduction of a method similar to this invention has been found in core journals and patent inquiries

Images