Segmentation and classification of SAR imagery on flood zones in Colombia, a computing tool for disaster prevention
Share
Date
2014-09-08
Director(s)
Publisher
Universidad Antonio Nariño
Campus
Faculty
Program
Degree obtained
Document type
COAR type
http://purl.org/coar/resource_type/c_6501
Citation
Bibliographic Managers
item.page.resume
In order to prevent natural flood disasters it important to identify the flood areas. In Colombia, there is space to develop automatic tools able to detect and study flood areas. For this reason, in this work we propose a computational tool in MATLAB, able to detect and classify Colombia’s flood zones in SAR imager. In particular, we used different classifiers, and according to the performance we selected the best. The training database was generated with the results of Fuzzy Clustering, K -means and Region -Growing segmentations on flood zones in SAR imagery. We used two different classifiers: the first one is a Bayes classifier, while the second one is a Support Vector Machine (SVM). In order to evaluate the performance, we used indices such as the overall accuracy, user accuracy and Kappa index. According to the results, the SVM classifier presents better accuracy. However, the Bayes classifier had better results classifying pixels corresponding to populations even with little training data.