Currently, the number of Koreans travelers and employers to Japan has increased, and disasters caused by typhoons and earthquakes have frequently occurred in Japan. Japan uses pictogram to guide people to evacuate natural disasters such as earthquakes and tsunamis. But it is hard for Korean to understand the disaster evacuation pictogram because the pictogram is unfamiliar to Koreans. In this paper, we propose a meaning method for improving Korean understanding of Japanese disaster evacuation pictogram, which recognizes the pictogram on a portable device when a disaster such as a typhoon or an earthquake occurred in Japan. The proposed method detects a pictogram accurately by using a deep-learning based object detection algorithm and the results of detection are matched the table of pictogram interpretation in Korean. For detection of disaster evacuation pictogram, the proposed method extracts object features using MoibleNet, and uses object detection architecture of YOLOv2 to locate and classify objects using the extracted features. Experimental results show that the proposed method recognizes pictogram of disaster evacuation signboards with accuracy of 98% or more and provides processing speed of 20ms or less.