DETECTION OF WELD DEFECTS IN RADIOGRAPHIC IMAGES BASED ON REGION COUNTING AND CLASSIFICATION USING FEED FORWARD NEURAL NETWORK WITH BACK PROPAGATION

摘要

An application of image processing and neural network applied to the analysis of radiographic images to detect weld defects is illustrated. The motivation of the work is to identify different types of weld defects created, thereby reducing the costs of any successive repairs. The application developed has been tested through a radiographic image of carbon steel submerged arc welded test pieces. The implanted defects from the burn-through to porosity were detected and classified in terms of the total area of defect size, shape, and position within the weld region. For developing application Visual C++ is used.