Ground object classification based on UAV multispectral images

The classification of ground objects has very important practical significance. However, the current classification mainly relies on field investigation or RGB images, which is either costly or inaccurate. This paper proposed a new fast and effective method for the classification of ground objects based on UAV multispectral images. A four-rotor Unmanned Aerial Vehicle (UAV) equipped with a multispectral camera was used to collect multispectral images for classification. Firstly, the acquired multispectral images were corrected and denoised. Then, a series of characteristic parameters were extracted, including gray mean values, vegetation indices, textural features. Finally, three different classification models considering the existence of multi-collinearity in independent variables were established, based on stepwise regression, partial least squares, principal component analysis - extreme learning machine respectively. The classification accuracy of the above models were all more than 87 percents for trees, grass, lanes and sidewalks.