Kawaii is a Japanese cultural uniqueness that attracts attention around the world. It has been considered as an important value that increases impressions on various products such as Hello Kitty. Since 2017, we have conducted our research on fashion and investigate kawaii fashion trends in Japan. Our recent study introduces deep learning as a new approach to classify images of kawaii fashion styles. In this study, we propose to construct a classification model to classify five kawaii fashion styles in our dataset consisting of Classic, Harajuku-type Kawaii, Lolita, Orthodox, and Street, each of which has approximately 100 images. We constructed and compared the classification performance between two deep learning models: color-image model and grayscale-image model. As the results, we clarified that the model trained by color images has high performance especially in classify between the HK and the other four kawaii fashion styles. Our proposed model contributes to future application in the fashion industry as a quantitative method in positioning the style for new kawaii fashion designs.