DNBとBTデータを用いたANNによる漁火検出モデルについて

抄録

Day and Night Band (DNB) of Visible Infrared Imaging Radiometer Suite (VIIRS) on the Suomi National Polar-Orbiting Partnership (S-NPP) has exhibited a capability to detect the fishery lights in the night time. The distribution of fishery lights, which are corresponding to the distribution of fishery resources, has a significant information for fishery industries and resource managements. Unfortunately, it is difficult to distinguish the fishery lights on DNB images because the lunar lights reflected by clouds are observed simultaneously. In this study, the artificial neural network (ANN) was developed to detect the fishery lights apart from the lunar lights reflected by clouds. The ANN was trained to simulate the DNB lights with the brightness temperature at 3.7μm (BT_3.7) and the fraction of Moon illumination in date and location. The fishery lights were given as the absolute error between the predicted and the observed DNB. The errors were compared between the pixel based ANN and the convolutional NN (CNN), and the pixel based ANN was superior to the CNN due to convolution of the cloud pixels.