2022 年 42 巻 1 号 p. 51-62
Due to the strong demand for renewable energy resources, the number of newly built photovoltaic cells has increased dramatically. However, these photovoltaic cells are sometimes subject to disasters such as floods and mudflow. To keep such facilities safe from disasters, a method that can monitor the locations and extent of photovoltaic cells in hazardous zones in cost-effective and less time-consuming ways is necessary. In this study, we developed a multi-temporal and multi-source machine -learning-based method for photovoltaic cell detection . The Sentinel-1 and Sentinel-2 datasets were used as data sources, and the random -forest classification method was applied to classify the land use and land cover (LULC) of the study area. Various combinations of inputs to the classifier were compared based on their performance of the LULC classification. After this process, the combination of optical -data, coherence -data, and the average of the coherence -data was selected as the best classification method. The photovoltaic cell detection process was carried out by combining the multi-temporal classification results to improve detection accuracy. This photovoltaic cell detection method achieved high -overall accuracy, high user accuracy, and a high-kappa coefficient.