BLINK: An Effective Indoor Localization Method Based on CSI by Broad Learning and Kolmogorov-Arnold Network

With the widespread use of Wi-Fi in indoor environments, fingerprint-based indoor localization methods that leverage channel state information (CSI) have garnered growing attention. These methods offer high-precision navigation and task execution support for indoor robots. However, similarity-based methods are not well-suited for large-scale scenarios, while neural network-based methods are hindered by long offline training times and limited localization accuracy. In this paper, we propose an effective CSI indoor fingerprint localization method using the Broad Learning System (BLS) and Kolmogorov-Arnold Network (KAN). Firstly, we use the BLS to train the model weights, significantly reducing the training time. Next, we utilize the B-spline transformation component of the KAN to effectively capture and represent the complex nonlinear relationships in the features, enhancing localization accuracy through confidence coefficients. Finally, we implement the proposed algorithm in two real-world environments and perform extensive experiments to evaluate its performance. Compared with several existing methods, our approach not only significantly shortens training time but also achieves a 39.05% improvement in localization accuracy.