Using Filtered Synthetic Data and Cluster K-Nearest Neighbor to Enhance the Accuracy of Fingerprint-Based Indoor Localization

Accurate location information is essential for location-based services in indoor environments. In recent years, Deep Neural Networks (DNN) based on Received Signal Strength Indicator (RSSI) fingerprints have shown significant potential in localization performance and cost-effectiveness. However, these methods necessitate a substantial volume of training data to effectively train DNN models, consequently escalating the expenses associated with data collection. This paper proposes a method to enhance RSSI data by generating synthetic RSSI data using Filtered Generative Adversarial Networks (FGANs) and mixing it with partially labeled real data to form a comprehensive dataset to address this issue. Additionally, a Cluster K-Nearest Neighbor (CKNN) algorithm is introduced to classify of the mixed dataset. This data augmentation and classification scheme significantly improves the overall accuracy of indoor localization systems, as validated through simulation experiments. The results demonstrate that the proposed method enhances the localization accuracy of simulated data by 12.66%, while saving data collection costs, thus achieving acceptable localization accuracy.