Indoor visible light positioning system based on local and non-local feature fusion network

With the advancement of indoor visible light positioning (VLP) systems, existing neural network learning algorithms often rely on stacking deep layers to obtain larger receptive fields, which are prone to issues like vanishing or exploding gradients; to address these challenges, this paper proposes a VLP system based on a Local and Non-local Feature Fusion Network (LNCA-Net), which adopts a parallel dual-branch architecture integrating a Local Feature Enhancement Module (LFEM) and a Non-local Feature Extraction Module (NL-FEM) to extract multi-scale local features and model global dependencies, respectively, and further introduces a Cross-Attention Fusion Mechanism (CAFM) to effectively integrate local details with global contextual information, where experimental evaluations in a 2 m × 2 m × 2.7 m test environment demonstrate that the proposed method achieves a root mean square error (RMSE) of 0.59 cm and a maximum error of 3.61 cm in simulation, and attains an RMSE of 3.28 cm with a maximum error of 15.14 cm in practical experiments, and comparative analysis with other positioning algorithms confirms that this approach offers high positioning accuracy, providing a reliable and practical solution for high-precision indoor visible light positioning.