Task-Oriented Image Transmission for Scene Classification in Unmanned Aerial Systems

The vigorous developments of the Internet Things make it possible to extend its computing and storage capabilities tasks in aerial system with collaboration cloud edge, especially for artificial intelligence (AI) based on deep learning (DL). Collecting a large amount image/video data, unmanned vehicles (UAVs) can only hand over intelligent analysis back-end mobile edge (MEC) server due their limited capabilities. How efficiently transmit most correlated information AI model is challenging topic. Inspired by task-oriented communication recent years, we propose new image transmission paradigm scene classification task. A lightweight developed front-end UAV semantic block perception images channel states. To achieve tradeoff between latency accuracy, reinforcement (DRL) applied explore blocks which have greatest contribution classifier under various Experimental results show that proposed method significantly improve accuracy more than 4% same conditions, compared other saliency methods.