Abstract: The algorithm of forest training data automation(TDA)has been successfully applied to Landsat images. Taking Guangping Town, Cangwu County, Guangxi Province as the study area, we selected the ALOS image of 2007 and the RapidEye image of 2011 to explore the algorithm’s application in high resolution remote sensing images. The pure forest training samples were automatically identifed at first, and the change detection result was then obtained by the forest/non-forest classification which extracted by the normalized integrated forest index image involved in the anlaysis.The accurate evaluation results showed that the total area error was -2.6% and the spatial location accuracy was 87.7%. It was shown that this algorithm could be effectively applied to high resolution remote sensing images to extract pure forest training samples for the forest/non-forest classification and change detection as the original data.