Neural network painting defect calssification using Karhunen-Loeve transformation

摘要

This paper deals with the problem of painting defect detection on reflecting surface objects. The problem has been approached with an optical inspecting method. A laser beam hits the object surface. The light scattered from the rough surface generates a digital speckle. The speckle is affected by the painting defect. Using the Karhunen-Loeve transformation, the speckle pattern is transformed into a feature vector. This information is used to train the neural-networks in recovering the defect. The reliability and effectiveness of a prototype is validated by experimental results. At the end, the proposed method is compared with another optical inspection method.