Robust Model-Based Learning to Discover New Wheat Varieties and Discriminate Adulterated Kernels in X-Ray Images

In semi-supervised classification, class-memberships are learnt from a trustworthy set of units. Despite careful data collection, some labels in the learning set could be unreliable (label noise). Further, a proportion of observations might depart from the main structure of the data (outliers) and new groups may appear in the test set, that were not encountered earlier in the training phase (unobserved classes). Therefore, we present here a robust and adaptive version of the Discriminant Analysis rule, capable of handling situations in which one or more of the afore-mentioned problems occur. The proposed approach is successfully employed in performing anomaly and novelty detection on geometric features recorded from X-ray photograms of grain kernels from different varieties.