Automatic classification of defects on the product surface in grinding and polishing

Grinding and polishing are standard operations in material processing. It is important to inspect and classify the potential defects existing on the product surfaces after grinding and polishing in order to obtain high quality in both functionality and aesthetics. Post-processing handling can be carried out after the defects have been correctly classified. A vision system already exists to detect and classify defects based on captured grayscale images automatically. The system is able to find and locate the defects precisely, but is incapable of placing those defects into the right predefined classes. The old system classifies the defects 30% of the time into the 15 predefined classes based on shape features. In this paper, a new classification strategy has been introduced using diverse extracted features. In addition to shape features, some other feature extraction methods were tried, e.g. Laws filter bank, DCT (Discrete Cosine Transformation) filter bank, Gabor filter bank, and statistical features based on co-occurrence matrix. The Support Vector Machine (SVM) was used as a multi-class classifier with input of the extracted features. By combining the Gabor filter features and the statistical features, the classification rate of the system can reach 82% overall right rate, which is applicable practically.