Integration of VIR and thermal bands for cloud, snow/ice and thin cirrus detection in MODIS satellite images

Cloud detection and discriminating them from snow/ice pixels has been one of the concerns of the remote sensing community. In this article, we used two kinds of ensemble learning methods, boosting and random forest (RF) for fusion of visible-infrared (VIR) and thermal classifiers for a new application that has not been used yet, detection of cloud, snow/ice and thin cirrus pixels. Boosting methods applied include 4 kinds of boosting. Two approaches were used for fusion, including decision level (DL) fusion and feature level (FL) one. After fusion using boosting methods, the accuracy of cloud detection did not increase but kappa index of snow/ice pixels improved in most cases, this increase was more in FL than DL case. RF algorithm increased thin cirrus producer accuracy 3% and 2% in DL and FL approaches. Results of this article showed that FL fusion approach has similar or better performance than DL one.

Keywords:

cloud , snow/ice , thin cirrus , boosting , RF , VIR and thermal classifiers