【摘要】 针对现有环境语义辅助的毫米波波束预测方法在工业雾霾环境下的精度下降问题，提出了一种多模态语义辅助的波束预测方法。首先设计Lite-DHNet模块对雾霾图像进行端到端去雾重建，然后将去雾图像与环境语义中的定位数据进行融合，最后设计掩码图像特征提取网络和波束索引推理网络，实现毫米波波束预测。实验结果表明，所提方法能够以较低的模型开销实现高效的波束预测，Top-3预测准确率最高可达99.5%，有效减轻了雾霾环境对波束预测精度的影响。更多还原

【Abstract】 To address the issue of decreased accuracy of existing environment semantics aided millimeter wave(mmWave) beam prediction methods in industrial haze environments, a multi-modal semantics aided beam prediction method was proposed. Firstly, a Lite-DHNet module was designed to perform end-to-end dehazing reconstruction on haze images. Then, the dehazed images were fused with localization data within the environment semantics. Finally, a masked image feature extraction network and a beam index inference network were designed to achieve mmWave beam prediction. Experimental results demonstrate that the proposed approach achieves high beam prediction accuracy with low computational overhead, and its Top-3 prediction accuracy reaches up to 99.5%, effectively mitigating the impact of haze environments on beam prediction accuracy.更多还原