User data selection using CNN-feature extractor for fingerprint localization

Abstract

This paper examines a method for fingerprint indoor localization that employs CNN. CNN is trained using AP information. The estimation accuracy of CNN improves as the number of AP information increases. However, gathering AP information is expensive. The problem can be solved using UD (User Data). The UD is unlabeled data because the measuring method does not know the exact location of the user. As a result, we can perform semi-supervised learning with the estimation result as the correct label. In this paper, we propose a method for selecting UD using a CNN-feature extractor.