Indoor visible light positioning system based on memristive convolutional neural network

To improve the performance of indoor visible light positioning systems and address the von Neumann bottleneck problem, this paper proposes an indoor visible light positioning system based on memristive convolutional neural networks. By integrating ResNeXt, ESCA mechanism, and pruning algorithm, the convolutional neural network is enhanced, significantly boosting the positioning performance of the indoor visible light positioning system. This enhancement effectively resolves issues such as excessive network parameters, gradient vanishing, and exploding gradients in traditional neural networks. Experimental results demonstrate that in a positioning environment with dimensions of 6 m x 3 m x 3.6 m, the proposed enhanced network greatly enhances positioning accuracy while successfully reducing network parameters. Based on this improved network, a low-power and highefficiency memristive convolutional neural network constructed using memristors also demonstrates excellent indoor visible light positioning capabilities.