Focusing on the low accuracy of weld size measurement algorithm in the radiographic image of small diameter tube placed outside the supercritical CO_2 boiler

a size measurement algorithm based on maximum entropy threshold segmentation is proposed. The proposed method mainly includes three parts: inner wall extraction

welding area segmentation and weld detection. Firstly

the vertical line of pipe wall is detected by adaptive thresholding and probabilistic Hough transform(PHT)

and then the image is cropped to get the inner weld wall region. UNet semantic segmentation algorithm is used to obtain the through-hole region segmentation mask

and the inner weld wall ellipse is fitted according to the through-hole edge and prior constraints. Furthermore

the weld area is precisely segmented by the maximum entropy threshold to calculate the weld size in the weld detection module. The experimental results show that

the accuracy of welding beads' number reaches 96.79%

and the accuracy of welding beads' dimensions reaches 95.07%

indicating the size of weld beads can be measured accurately.