Indoor Localization via Deep Learning with Self-Attention and Stacked Denoising Autoencoders

Although GPS excels in outdoor navigation, its indoor localization capabilities are inadequate. This study presents an indoor positioning model designed for complex, multi-level environments, utilizing Stacked Denoising Autoencoders (SDAE), attention mechanisms, and Convolutional Neural Networks (CNN). SDAE enhances data quality and feature extraction, attention mechanisms focus on key localization features, and CNN processes spatial aspects of WiFi fingerprint data. Evaluated with the UJIIndoorLoc dataset, our model achieved an accuracy of above 99% in detecting buildings and floors, with a mean localization error of 6.41 meters, proving its effectiveness in indoor contexts.