自動融雪装置の開発に向けた深層学習を利用した積雪画像の識別

抄録

Snow on roads is removed by snowplow and they piled up aside of roads. Residents along the road have to remove the snow in front of their houses. It is heavy burden especially for elderly people. In our laboratory, we aim to develop to put into practical use an automatic snow melting device. In this research, we develop a system to recognize snow scenery from the information of photo images taken by a camera. We use a model learned by deep learning for yes-no classification of the images data. These complicated images are inflated by 30 times to distinguish with higher accuracy. We made hypotheses about features with common tendencies for miss-classified images. In particular, we examined improvement methods for checking the hypotheses by reprocessing the images with similar tendencies.