Automatic inspection system for billet

In billet production, the quality of billet is an important issue to assure. In this study, we proposed an automatic inspection system to locate and classify defects for billet. The proposed system is consisted of three modules: (1) image processing and defect location, (2) feature extraction and selection, (3) incremental learning classifier. In the first module, the region of interest is extracted and normalized to reduce the effects of uneven illumination. We then develop two methods to detect different types of defects based on their characteristics. In the second module, k-nearest neighbor classifier and tabu search are employed to select the best set of features for classification. In the last module, a classifier with incremental learning capability called Learn is used to classify the detected defects. Experiments show that the proposed system provides defect detection with good accuracy and speed. Comparing with the conventional BPN, the Learn++ classifier is much more efficient in training and obtains better classification rates.