Research On Pipeline Infrared Image Temperature Calibration And Defect Detection Technology Based On Machine Vision

According to statistics,China has built the world’s largest heating pipe network.The pipeline may appear aging and rust after being idle,causing water seepage at the mouth of the pipe.The increase in heating area brings challenges to pipeline maintenance.Relying on the detection method of the pipeline robot,it can only work during non-heating periods and cannot operate during the heating season.Infrared thermal imaging non-destructive testing technology is an emerging testing technology that has been widely used in recent years.Infrared non-destructive testing technology uses infrared temperature measurement,does not touch the object to be tested,and does not damage the temperature field.By intuitively reflecting the two-dimensional temperature distribution of the object to be measured in the form of a heat map,the physical properties under the surface of the material are displayed through changes in the surface temperature distribution.This paper proposes a defect detection method based on Efficient Net-Yolo V3 to complete the precise location and recognition of pipeline defects on infrared images.And proposes to use fully convolutional networks(FCN)to estimate the temperature of pipeline infrared images.The angle of the defect area is large,and the temperature analysis is easy to interfere.An accelerated Gaussian mixture model(GMM)is proposed for pixel-level segmentation of the defect area.This paper initially solved the problem of detecting defects in the heating pipe network,saving a lot of labor costs and improving the efficiency of maintenance.The research work and innovations of this article are as follows:(1)The infrared image temperature calibration of heating pipes stores the temperature point by point,with large storage space and high transmission cost.A method for predicting the temperature of infrared images based on FCN is proposed.First,divide the temperature range of the infrared image to generate a temperature label.The original FCN has low accuracy in infrared temperature estimation.The FCN is modified and adjusted on the backbone network architecture,activation function,and loss function,and comparative experiments are used to verify the necessity of network adjustment.At the same time,a polynomial regression was used to fit the relationship between the infrared image heat value and the actual temperature,and a comparative experiment was carried out.The experimental results show that the infrared image temperature estimation method based on FCN has 98.4% of the infrared image temperature calibration error less than ±2°C,and effectively saves 60.5% of the storage space of the infrared image.(2)The detection efficiency of defects in heating pipelines is low and the accuracy is poor.By combining UAV technology and infrared non-destructive testing technology,a method of pipeline defect detection and classification based on Efficient Net-Yolo V3 is proposed.Pipeline infrared image has high signal-to-noise ratio and few detailed features.It is proposed to use Efficient-Net to replace Yolo V3’s original backbone network Dark Net-53.The Efficient-Net multi-scale network adaptive mechanism is used to realize the depth feature recognition of infrared images of various sizes.Through the Bi-FPN feature pyramid fusion of multi-scale features,high-precision detection of infrared image defects is achieved.Experiments show that Efficient Net-Yolo V3 achieves 88.53% m AP on the test set,which is better than many classic target detection networks.It proves the superiority of the network structure and improves the accuracy of pipeline defect detection.(3)The angle of the defect area is large,which causes the background area to interfere with the defect analysis.A pixel-level infrared image defect segmentation method based on GMM is proposed.The processing speed of GMM is improved by adopting infrared image whole-line modeling technology.First,it analyzes the shortcomings of GMM in image segmentation.The original GMM is optimized in model initialization and update mode,which improves the image segmentation speed of the original GMM;by filtering and denoising the processing results of the GMM,the accuracy of defect region segmentation is further improved.Experiments show that the processing speed of our method is better than that of the original GMM,which improves the segmentation speed of pipeline defects.The defect detection m AP of the heating pipeline proposed in this paper is as high as 88.53% after testing;the estimated error of the infrared image temperature is less than ±2°C as high as 98.4%,and 60.5% of storage space is saved,which meets the needs of the industry.