Optimization of Image Transmission in UAV-Enabled Semantic Communication Networks

Unmanned Aerial Vehicles (UAVs) are widely used in emergency communication systems. However, the lack of adequate communication power and bandwidth in UAV-enabled communication networks hinders transmitting large files like images and videos. The semantic communication is introduced to tackle this challenge by transmitting the semantic features instead of raw data. In this paper, we propose a UAV semantic communication architecture by jointly optimizing the UAV flight trajectories, semantic information selection, and resource allocation. Accordingly, an optimization problem is formulated to maximize the total similarity score of semantic information, and then, a soft actor-critic based entropy-maximization (SAC-EM) algorithm is proposed. Simulations demonstrate that the proposed algorithm increases the total semantic similarity score by 54.2% and improves convergence speed by 80% compared to traditional reinforcement learning algorithms.