Detection of Convallaria Keiskei Flower by Combining Convolutional Neural Network and Fuzzy C-means Using Unmanned Aerial Vehicle-based Data

Abstract

Services and operations have been systematized to realize Society 5.0, and the use of processing technology for images acquired by unmanned aerial vehicles (UAVs) is highly effective in improving productivity. Conversely, the Convallaria keiskei, a plant native to Japan, was once on the verge of extinction. Since then, the management and species identification of Convallaria keiskei have been performed manually. Therefore, automatic species identification and population estimation of Convallaria keiskei using image processing and machine learning can reduce the management costs. In a previous study, we proposed flower number estimation by combining image processing, convolutional neural network, and fuzzy c-means (method A). In this study, we evaluate the method A using UAV data obtained in 2022. To validate the method A, it was compared with the CNNbased detection method (method B). The results show that the method A improved the F-measure by up to six of the seven UAV datasets compared with the method B.