Detecção automática de nuvem e sombra de nuvem em imagens de sensoriamento remoto

Abstract

This paper presents an automatic method to detect cloud and cloud shadow in remote sensing images that works in visible and near infrared spectrum. The aim of the paper is the detection in AWFI images of the AMAZONIA-1 satellite. The cloud detection is based on applying three filters. The reflectance of spectral bands is used to compute the NDVI, WI and HOT values used by classifier. The cloud shadow detection is done by integration of dark pixels and water masks with the image resulting from difference between the scene to be classified and a reference of the same region, ideally free cloud and perfectly registered. In the absence of AWFI images, since the AMAZONIA-1 satellite is still in construction phase, the tests have been done with LANDSAT-5 TM images. The mean accuracy of cloud detection is 88.70% and of cloud shadow detection is 75.03%. The global accuracy of classification has shown that over 90% of the pixels have been classified correctly as cloud, no cloud, no shade and shadow.