CBERS-4/MUX automatic detection of clouds and cloud shadows using decision trees

Favorite this paper

Cloud contamination can compromise surface observation on satellite images and impossibility land cover and land use mapping, due to their high reflectance. Similarly, cloud shadows candarkentheimageorbeconfusedwithwater,makingithardertodifferentiatetargets. Thispaperaims at evaluating an automatic cloud and cloud shadow detection method using decision tree classifier for CBERS-4 (China Brazil Earth Resources Satellite) MUX (Multispectral Camera) camera. In relation to the features used in the classification process, 3 methods were tested to classify 10 images of CBERS-4 MUXcamera. Thefirst oneusedspectral informationandspectralindices, suchasNDVI,WIandHOT; the second one added shape attributes in the feature set, and the third one added texture attributes. The classification process considered 3 classes: cloud, cloud shadow and cloud-free, which were validated using visually interpreted images. The results presented an overall accuracy of about 92.98%. The accuracy for the cloud detection was 0.91, while for the cloud shadow the classification accuracy was 0.67. These results point out that for sensors that has only visible and near infrared spectral bands, like CBERS-4/MUX, the NDVI, WI and HOT spectral indices are relevant for cloud detection. On the other hand, for cloud shadow detection it is necessary to explore other features capable to discriminate it from dark objects in the images.