인공위성 자료를 이용한 Faster R-CNN 기반 태풍 식별 및 강도․중심위치 산출

Abstract

Locating the center and predicting the intensity of typhoons are critical issues to prevent damages caused bythem. In this paper, we propose two models based on Faster R-CNN for predicting the intensity and the centerlocation of typhoons via satellite images without pre-processing; one of the models utilizes an only single-channeland the other uses whole four channels. The biggest advantage of proposed models is that it can produce both thecenter coordinates and the class of the intensity regardless of the number of typhoons occurred, which never havebeen studies yet. Compared to the single-channel model, multi-channel model achieves higher typhoon detection ratethan single-channel model since it can extract various aspects of features from multiple channels through CNNbackbone network.