Image Quality Indicator Localization Using Mask R-CNN

Ensuring image quality is integral to the inspection of components in industrial radiography. It is important, to the art of characterizing unknown defects in radiography, that the quality of the radiograph meets the sensitivity level required for successful detection of defects that exceed a predetermined critical type and size. This research work focuses on applying an existing object detection and instance segmentation framework called Mask R-CNN to the recognition of image quality indicators (IQIs) typically used in industrial radiography. Mask R-CNN is shown to be capable of properly segmenting IQIs from radiographs when the standard practices for IQI placement are followed. The results presented are obtained with less than 103 training examples. This study explored the difference in Mask R-CNN performance when the datasets used for training and inference are both small and contain IQIs at various orientations. Presented herein are the results for comparing models trained and inferred with only parallel examples to models trained and inferred with parallel, transverse, and askew examples. The results of this study have important implications for the application of existing computer vision and narrow artificial intelligence systems toward the detection of quality assurance objects, and by extension defects, within industrial radiography.