Value-of-Information Optimization for Object Detection-Driven Joint Video Transmission and Processing in UAV-Enabled Wireless Networks

In the wake of disasters, rapid and efficient search and rescue operations are essential. Unmanned aerial vehicles (UAVs) have become instrumental in such scenarios, providing real-time video streaming that can be used for object detection to locate survivors. This technology, however, faces significant challenges due to the limited communication and onboard computational resources, which are critical for processing and transmitting high-quality video data. To address these issues, this paper proposes a novel approach that leverages the concept of the value of information (VoI) to optimize the trade-off between the accuracy of object detection and the associated communication costs. By dynamically adjusting the video stream’s quality, the proposed system aims to ensure that the most valuable information is transmitted within the constraints of bandwidth and computational power. To operationalize this concept, we introduce a Deep reinforcement learning (DRL) algorithm that employs the soft actor-critic (SAC) method. The algorithm benefits from the integration of object features and contextual information extracted by ResNet50, which is then processed through a cross-attention structure within the critic network. Our simulation results indicate that our approach significantly enhances the VoI, achieving higher accuracy in object detection with better resource management compared to traditional strategies.