Examination of a UAV image classification method by using machine learning and wavelet transform

This paper examined a machine learning technique with the wavelet transform for classifying land cover conditions in Unmanned Aerial Vehicle (UAV) images of a riverine landscape. The UAV images were taken in a river course of Kurobe River, Japan. Each UAV image analyzed was composed of RGB, Normalized Difference Vegetation Index (NDVI), and a Digital Surface Model (DSM) of the river geomorphology made from a Structure from Motion (SfM) image processing of the UAV images. In a pre-processing of the machine learning, the DSM was decomposed into low/high wavenumber components through wavelet transform, and its edges were further extracted to effectively utilize the height difference information in DSM. The result of the machine learning showed that the F-measure had high enough above 0.91 in the dataset including all characteristic values from RGB, DMS, and NDVI into the machine learning algorithm.