- 著者
-
杉本 隼斗
濱川 礼
- 雑誌
- 研究報告エンタテインメントコンピューティング(EC) (ISSN:21888914)
- 巻号頁・発行日
- vol.2021-EC-60, no.5, pp.1-4, 2021-05-25
It is said that the eating time of an avocado can be determined using the color, texture, and firmness of the rind as indicators, but it is difficult to determine the eating time of an avocado with high accuracy only for experienced users, which is a problem faced by end users of avocados. In this study, we developed a deep learning model that classifies avocados into three classes (unripe, ripe, and overripe) based on image input, and implemented a mobile application that can be run on a smartphone equipped with the model. For the deep learning model, we investigated a classification method using deep metric learning. Deep metric learning has achieved many successes in face recognition tasks. It is considered useful when applying deep learning to datasets with small differences in image features between classes and a small amount of data for each class, and the dataset we collected in this study has the same characteristics. The model was able to classify the evaluation data with an accuracy of 89.77%.