CLASSIFICATION OF POTENTIAL DEFECTS IN THE AUTOMATIC INSPECTION OF ALUMINIUM CASTINGS USING STATISTICAL PATTERN RECOGNITION

In this paper we report the results obtained recently by classifying potential defects in the automated x-ray inspection of aluminium castings using statistical pattern recognition. In our classification, 71 features (e.g. area, perimeter, roundness, invariant moments, Fourier descriptors, mean grey level, several contrasts, texture features, etc.) were analysed to characterise the potential flaws. The extracted features were measured from more than 10.000 regions segmented in 56 radioscopic images (without frame averaging) of cast aluminium wheels. In addition, five statistical classifiers were tested.