Classification of surface defects on wood boards
Abstract
Most of manufacturing processes of a furniture manufacturing plant are labor oriented. To improve the productivity and quality of products, automation of the different stages in the production is needed. Classification of surface defects on wood boards is one of the important steps towards a completely automated wood processing plant. Two different approaches are applied to locate and classify the nine classes of surface defects: the approach based on texture information and the approach based on a priori knowledge and texture information. In the approach based on texture information, features characterizing the images of wood boards are extracted by two methods: the co-occurence matrix and image modeling methods. In the co-occurence matrix method, the gray level mean, variance, and features calculated from the co-occurence matrix form a feature set. In the image modeling method, the feature set consists of the gray level mean and the CAR-model parameters calculated by the robust estimation algorithms. Feature selection method is then applied to the calculated features to increase the separability between classes. Bayesian and tree classifiers are constructed and tested. In the approach based on a priori knowledge of the surface defects and their texture properties, a hierarchical classification procedure is proposed to reduce the computation time and improve the resolution of the defect detection. The background and wane are identified using their locations as a priori knowledge. The rest of the surface defects and clear wood are divided into three subsets by identifying their shapes. The surface defects in each subset are recognized based on the gray level mean and the CAR-model parameters. In order to investigate the feasibility of using the proposed classification procedure in the dusty environment of a wood processing plant, three sets of image data which mimic the dusty board surfaces are generated and utilized in the classification experiments.
Degree
Ph.D.
Advisors
Koivo, Purdue University.
Subject Area
Electrical engineering|Wood|Technology
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.