A High-Accuracy Semantic-Driven Resource Allocation Scheme for UAV Networks

With the growing computer vision applications, lots of videos are transmitted for AI tasks in Unmanned Aerial Vehicles network (UAVs). The allocation of resources can significantly affect the performance of video semantic analysis. Traditional resource allocation schemes for UAV video transmission, such as quality-of-service (QoS) based and quality-of-experience (QoE) based methods, are deemed suboptimal, as they fail to satisfy emerging requirements (e.g., dynamic channel adaptation and semantic-aware efficiency). In this paper, we propose an efficient semantic-driven resource allocation (SDRA) scheme for video transmission in UAV networks under bandwidth and power constraints in order to improve the video semantic analysis performance, in terms of object detection accuracy. A resource allocation model is first proposed. Then the semantic-driven resource allocation problem is formulated and converted to a solvable convex optimization problem. Finally, simulation results show our proposed scheme outperforms the benchmarks in terms of the object detection accuracy.