Multi-task Semantic Communication System for UAV Image Collection

초록

This paper proposes MSR-DeepSC, a novel multi-task semantic communication system for efficient UAV image collection. Traditional transmission methods often face limited bandwidth and high latency. MSR-DeepSC addresses these challenges by transmitting task-relevant semantic features—rather than raw pixels—for both image super-resolution and scene classification. A lightweight Gated CNN encoder on the UAV side and a Vision Transformer-based decoder at the ground station are jointly optimized. Experimental results show that MSR-DeepSC surpasses existing baselines in reconstruction quality, classification accuracy, model size, and processing latency, making it particularly suitable for resource-constrained UAV applications.

키워드