Flooded and vegetation areas detection from UAV images using multiple descriptors

Abstract

This paper presents a method to detect small flooded areas from images which contain also vegetation zones. So, two classes are considered: flood class and vegetation. For the learning phase a supervised technique based on small patches is used. Based on efficiency analysis, the Histograms of Oriented Gradients on H colour channel and mean intensity on gray level are selected as discriminated features. The classification/ segmentation phase considers two separate classifiers: one for flood class and another for vegetation class. Because there are mixed patches (with water and also vegetation) a new class (common parts) is created as logical OR between the binary decisions of the classifiers. For each test image, the percentage of flooded, vegetation or common parts is calculated.