Summary
International Conference on Emerging Technologies for Communications
2021
Session Number:D1
Session:
Number:D1-1
User Data Selection using CNN Feature Extractor for Fingerprint Localization
Yohei Konishi, Satoru Aikawa, Shinichiro Yamamoto, Yuta Sakai ,
pp.-
Publication Date:2021/12/1
Online ISSN:2188-5079
Summary:
This paper scopes a method that applies CNN to Fingerprint indoor localization. AP information are used to train the CNN. As the number of AP information with correct labels increases, the estimation accuracy by the CNN improves. However, it costs a lot to collect AP information with correct labels. UD (User Data) can be used to solve the problem. The UD is unlabeled data because the measuring method of the UD does not know the user’s place exactly. We perform semi-supervise assuming the estimation result of the UD as the correct label. However, the estimation result of the UD may be incorrect. Therefore, it is needed to select UD which is correctly estimated and use it for training the CNN. In this study, we propose a way to select UD by feature value using CNN feature extractor.