Improvements on Classification by Tolerating NoData Values - Application to a Hybrid Classifier to Discriminate Mediterranean Vegetation with a Detailed Legend Using Multitemporal Series of Images

Natural and crop vegetation phenologic data become indispensable when creating thematically and geographically detailed maps through satellite images classification. Several date acquisition is necessary to achieve this cartography. However, the presence of clouds, shadows, snow, etc, is usual when many different dates are used and that fact implies an important loss in classifiable surface. This work presents a hybrid classifier designed to deal with the common problems appeared in the classification of Mediterranean vegetation. Specifically, IsoMM, the first phase of the hybrid methodology, is an unsupervised classifier that allows a better use of temporal series thanks to a particular treatment of no data values (or missing values) in the images. This methodology has been applied to a Mediterranean forestry zone with a legend of eleven categories and has been compared to a Maximum Likelihood classifier. The presented improvements allow classifying more surface than a common no data treatment strategy (whether unsupervised, maximum likelihood classification or the extraction of a problematic date) and achieving high accuracy level.