Color night vision can map natural colors to nighttime images of multiple bands (e.g., visible and long-wave infrared (LWIR)). These colors can assist the observers in better and faster understanding images, thus improving their situational awareness and shortening the reaction time. In this paper, we present an effective method combining deep learning and category colors. It utilizes the semantic segmentation for image segmentation first, and then colorize the image according to categories to avoid the same color scheme and unnatural colors. We compare our method with some others quantitatively and qualitatively, such as global colorization by single lookup table, where we show significant improvements. In addition, it can be expanded according to different environments and applications because of the fixed category colors.