Few-Shot Meta-Learning for Dynamic Indoor Fingerprinting Localization Based on CSI Images

While fingerprinting localization is favored for its effectiveness, it is hindered by high data acquisition costs and the inaccuracy of static database-based estimates. Addressing these issues, this work presents an innovative dynamic indoor localization method using a few-shot meta-learning algorithm to optimize performance with limited CSI data. By leveraging the “Learning to Learn” paradigm, this method utilizes past localization tasks and requires only a few updates, enhancing adaptability and learning efficiency in new environments. Furthermore, we introduce a task-weighted loss to improve knowledge transfer within this framework. Experiments confirm the method's robustness and superiority, achieving a notable 25.08% average gain in Mean Euclidean Distance (MED), proving especially effective in scenarios with limited CSI data.