To enhance the robustness and positioning accuracy of the Elman indoor visible light position sensing model, a bio-inspired multi-feature fusion learning method for indoor visible light position sensing is proposed. This method first preprocesses the acquired visible light images to ensure the accuracy of feature extraction. Then, by fusing features from different levels of a pre-trained neural network model, it constructs a position-sensing feature library, thereby enhancing feature representation capa bility and richness, which improves the model's position sensing precision. Finally, the dung beetle optimization(DBO) algorithm is employed to optimize the topology and weight parameters of the Elman neural network, addressing issues where traditional Elman neural networks easily fall into local optima in indoor position sensing, accelerating convergence speed, and enhancing generalization performance. The experimental results show that within a 3D space of 4 m×3.5 m×3 m, the proposed algorithm achieves an average positioning error of 0.21 m, with 91.3% probability of average positioning error is less than 0.4 m, improving positioning accuracy by 22.3% compared to the Elman algorithm.