Understanding the Distribution of Flowering Individuals of Rhododendron reticulatum Using UAV and Image Recognition by Machine Learning

Abstract

Wild azaleas are declining in the secondary forests in the Kansai region, and their conservation has become an important issue. In order to strategically conserve wild azaleas, we believe it is necessary to develop a method for obtaining quantitative data on flowering individuals over an area of several hundred hectares. A total of 109.4 ha of forest in Takaragaike Park (Sakyo-ku, Kyoto City) was the subject of this study. An orthomosaic image was created from data taken by a UAV on April 8, 2020. Using a machine learning, we automatically detected the flowers of Rhododendron reticulatum from orthomosaic images. At the same time, flowering individuals of R. reticulatum were surveyed at 46 sites. The accuracy of detecting the flowers of R. reticulatum using a machine learning was high (Overall Accuracy＝97.9％). There was a strong correlation between the number of flowering individuals at the field study sites and the area of flowers calculated from image detection results (r＝0.75). A new method for understanding the distribution of flowering individuals of R. reticulatum was demonstrated.