2020 年 26 巻 p. 521-526
This paper examined a new method for detecting riverine land covers by using a machine learning technique with the satellite images in different times in a Kurobe River channel. In the method, UAV (Unmanned Aerial Vehicle) images were used to train the machine learning method in several small portions where the type of riverine land covers was precise and no land cover change occurred between the different times. From these UAV images and the corresponding feature values (i.e., RGBs and Near-InfraRed) in the satellite images, accurate training data were created for the land cover detection in the whole extent of the satellite images. The results showed that the present method could detect riverine land covers at the multiple different times with high accuracy, in particular, the details of land cover change such as decreases in water surface and losses of vegetation due mainly to a flooding.