Forest plantations in Japan cover an area of 102,000 km2 and 70% are composed of Japanese cedar and cypress trees. Half of the trees in these planted forests are overdue harvesting; therefore, it is necessary to realize the forest cycle by actively harvesting these trees and using them as domestic timber. Related to harvesting, the rate of fatalities and injuries per 1,000 in the forestry industry in 2019 was 2–10 times higher than in other industries; thus, the implementation of new technologies to improve productivity and safety is required in the industry. In this study, we focused on a labor saving technique for investigating and classifying trees in forests using tree-bark images taken by mobile devices or monocular cameras. Previously, tree classification using such images had not been considered; however, if tree species can be classified using a lightweight monocular camera, it will be possible to create a tree species distribution map at low cost, even in wide areas of mountain forest, by installing lightweight cameras on drones. Here, we were able to verify the classification of Japanese cedar and cypress trees by integrating existing technologies. Specifically, we developed a classification method for tree-bark images using two types of image data (cedar and cypress) acquired with a monocular camera. By preparing image data appropriately, this method achieved a classification rate >80% with minimum labor requirements; therefore, the developed classifier allowed us to successfully identify the trees in images. Our method may also facilitate information sharing and centralized database management, contributing to the adoption of IoT (Internet of Things) in the forestry industry.
View full abstract