Goal-Oriented Communication With Semantic Reconstruction in Vehicular Networks

In recent years, semantic communication has received a lot of attention due to its ability to solve the challenges faced by traditional communication systems. However, little attention has been paid to the fact that during data compression and transmission, the lost data can be reconstructed by neural networks to improve transmission efficiency. In order to solve the impact of the loss of semantic information on the transmission performance in vehicular networks, this paper proposes a goal-oriented communication based on semantic reconstruction (GOCSR). By designing a semantic reconstruction network at the receiver, the lost semantic information is predicted and reconstructed, and then the complete semantic information is used to perform downstream tasks. To evaluate the efficiency of GOCSR, extensive simulation experiments are conducted using the Cityscapes dataset. Simulation results show that GOCSR can achieve higher target execution performance than the existing semantic communication schemes.