Fast Welding Defect Inspection on Digital Radiography Images Using Saliency Detection

Digital radiography (DR) is one of the most reliable technique for welding inspections. This industrial application of DR is mostly used for evaluation of the quality of metal welding lines and joints. Due to the effects of low resolution and contrast, high noise and uneven background brightness, the inspection and understanding of DR images are always a challenging task. Therefore, in this this work, we proposed a novel framework using saliency detection method for fast locating welding defects in the Digital frames automatically. Comparative experiments suggest that our method yields quality inference on defect detection in the welding area with the overall accuracy of 93.3% at the speed of 3 fps which makes real-time inspection possible.