Gully erosion detection by hierarchical classification and tree decision

This research aims at verifying the possibility to recognize gully erosion using Object-Based Image Analysis (OBIA). The studied area is located in Uberlandia, Minas Gerais. In according to research purpose, a multiscale hierarchical semantic net was defined to represent the expertise knowledge. Spectral data from IKONOS imagery, and intensity and altimetric data from ALS (Airborne Laser Scanner) were used. The objects were generated by multirresolution segmentation (FNEA-Fractal Net Evolution Approach) applied to spectral and altimetric data. The feature recognition was performed by hierarchical classification and tree decision algorithm (CART -Classification and Regression Trees). By means of proposed methodology, it was possible to identify the relevant input data, segmentation parameters and attributes to classify gully erosion. The results were similar; either by hierarchical classification or by CART, in which the possibility to use semiautomated methods from a set of previously identified and analyzed parameters was shown.