Partially-supervised updating of land-cover maps: A P2S2VM technique and a circular validation strategy

In this paper, we address automatic updating of land- cover maps by using multitemporal images without a complete knowledge of training data. In particular, two main novel contributions are proposed: a progressive partially-supervised support vector machine (P²S²VM) technique that extends the SVM method to the partially-supervised classification framework; ii) a circular accuracy assessment strategy for the validation of the learning of the classifier when no labeled test samples are available. Experimental results obtained on a multitemporal and multispectral data set confirmed the effectiveness and the reliability of both the proposed P²S²VM technique and the related circular validation strategy.