Automatic detection of clouds from aerial photographs of snowy volcanoes
- Authors
- Chang, Carolina
- Format
- Article
- Status
- publishedVersion
- Description
We propose a method for cloud detection from RGB aerial photographs of snow-capped volcanoes of Ecuador. For cartography purposes, clouds are undesired objects that occlude the terrain, while snow-covered areas are valid regions of a map. The traditional approach of image thresholding does not suffice when snowy areas cannot be dismissed from the image in advanced. We combine image thresholding with region growing and neural networks classification to detect clouds at the object level. We show that there is overlap at the pixel level of clouds and snow. At the classification task a fuzzy ARTMAP neural net achieves 91.4 % of success in fast learning mode and 95.5 % of success in slow learning mode at the same vigilance level, for 32?32 pixel images. Incremental learning is achieved at a loss of 0.4 % of the network performance.
Instituto Geogr?fico Militar
https://link.springer.com/chapter/10.1007%2F978-3-319-19264-2_15
- Publication Year
- 2015
- Language
- eng
- Topic
- CLOUD DETECTION OBJECT
BASED IMAGE ANALYSIS
FUZZY ARTMAP NEURAL NETWORK
- Repository
- Repositorio SENESCYT
- Rights
- openAccess
- License
- openAccess