Pattern Classification in Heterogeneous Domains Based on Evidence Theory (Poster)

In pattern classification, there exists a challenging problem when there are no training patterns. In some real applications, there may exist some labeled data in other related domain (called source domain), and such labeled data can be helpful to solve the classification problem in target domain. It is considered that the source domain and target domain are heterogeneous here. A new heterogeneous data transfer classification method based on evidence theory is proposed. Some corresponding patterns (pattern pairs) in source domain and target domain are given to build the link of these two domains. For each pattern in target domain, it is hard to determine one exact mapping value in source domain due to the distinct characteristics of these two domains. So we estimate a mapping scope in source domain using KNN technique. The target pattern is allowed to have multiple mapping values in the scope with different weights/reliabilties. These mapping values can produce different classification results. Evidence theory is good at combining the uncertain information. Therefore a new weighted DS fusion method is developed for combining these classification results, which are discounted by the corresponding weights, and the final class decision is made according to the combination result. A pair of heterogeneous remote sensing images and some UCI data sets are used in this paper to test the performance of our method with respect to several other methods, and it shows that the new method can efficiently improve the classification accuracy.