A Semantic Communication Method for Transmitting 2D Human Body Image Information

In this paper, a 2D human target semantic communication method based on deep learning and Transformer model is proposed. First, convolutional neural network and Region Proposal Network are used for human target detection and segmentation, and the segmented human targets and their unique segmentation masks are obtained. Then, a Transformer-based semantic coder is designed to encode the human target pixel data while preserving the key information of the human target. The encoding process includes image cutting into patches, linear mapping and adding position encoding, which is then encoded by Transformer model. At the receiving end, the semantic information is decoded by designing a semantic decoder, followed by patch mapping and image reconstruction to achieve decoding the semantic data into human target images. The simulation results show that this method achieves a certain effect in realizing the semantic encoding transmission and decoding reconstruction of human target.