CONSTRUCTION OF THE HIGHLY PRECISE PLANT IDENTIFICATION SYSTEM BY THE DEEP LEANING METHOD AND APPLICATION TO SATOYAMA CONSERVATION

Abstract

 Conservation of the forest environment in satoyama is important from the viewpoint of biodiversity and environmental education. In this study, in order to obtain the vegetation distribution of satoyama, plant images taken by reconnaissance and drone ware trained and the species of tree canopy images extracted from panoramic images taken by drone aerial photography using superpixel division ware identified by deep learning method. We achieved a high accuracy of 90% or more in 6 species of trees from the learning model created from the original image. In canopy identification, in addition to unifying the aspect ratio and removing similar images, it was found that the accuracy can be improved by using images taken at the same altitude as panoramic images for learning. Although it was possible to create a tree distribution map using this method, the number of tree species and captured images is small, and the applicable locations are limited. In the future, we will further increase the number of plant species, verify the effect of imaging conditions such as angle and time on identification accuracy, and expand the application area for satoyama conservation.