LBPriu2 Features for Classification of Radiographic Weld Images

Welding defects arises in welding. The welding material needs appropriate examination for its smooth operation and design. Non – Destructive Inspection is one of the significant methodology for proper recognition of the flaw defect. In the present work, an effort has been made to correctly identify and classify the weld defects. A dataset of 79 images with 08 defects is collected from Mechanical and Industrial Engineering Department of Indian Institute of Technology Roorkee. The image dataset has been pre-processed and the features have been extracted by LBPriu2 and processed by artificial neural network for further classification. The 10 level features have been extracted by LBPriu2 and fed to neural network after Image Segmentation. The features have been analyzed by Feed Forward neural network for classification. A detailed analysis of the different image segmentation methods with LBPriu2 features is analyzed. Irrespective of the poor quality of image dataset, classification accuracy of 89.9% is obtained.