Answers are grounded in this paper and cite their sources.
Experimental analysis shows improved positioning accuracy using a memristive convolutional neural network, suggesting innovative potential for VLP systems in coal mines.
To enhance the performance of visible light positioning (VLP) systems for underground coal mine personnel and mitigate the von Neumann bottleneck, this study proposes a novel positioning system based on a memristive convolutional neural network (MCNN). First, a comprehensive channel model integrating irregular wall reflections and multi-scale coal dust particle extinction characteristics is established to model the complex mine environment. Subsequently, a novel hybrid model, ResNeXt-ESCA-DG-Net (REDG-Net), is proposed, and its hyperparameters are optimized using the walrus optimizer (WO), significantly improving positioning accuracy. Furthermore, to address the discrete conductance characteristics of memristive devices, a joint progressive pruning and quantization-aware training (JPPQAT) method is introduced, which substantially reduces the number of memristive units and conductance states required by the MCNN while maintaining accuracy. Experimental results demonstrate that the low-power, high-efficiency MCNN constructed with memristive devices exhibits outstanding performance in underground coal mine VLP tasks.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: