Contrastive Learning based Semantic Communication for Wireless Image Transmission

Shunpu Tang,Qianqian Yang,Lisheng Fan,Xianfu Lei,Yansha Deng,Arumugam Nallanathan
2023-04-19
Abstract:Recently, semantic communication has been widely applied in wireless image transmission systems as it can prioritize the preservation of meaningful semantic information in images over the accuracy of transmitted symbols, leading to improved communication efficiency. However, existing semantic communication approaches still face limitations in achieving considerable inference performance in downstream AI tasks like image recognition, or balancing the inference performance with the quality of the reconstructed image at the receiver. Therefore, this paper proposes a contrastive learning (CL)-based semantic communication approach to overcome these limitations. Specifically, we regard the image corruption during transmission as a form of data augmentation in CL and leverage CL to reduce the semantic distance between the original and the corrupted reconstruction while maintaining the semantic distance among irrelevant images for better discrimination in downstream tasks. Moreover, we design a two-stage training procedure and the corresponding loss functions for jointly optimizing the semantic encoder and decoder to achieve a good trade-off between the performance of image recognition in the downstream task and reconstructed quality. Simulations are finally conducted to demonstrate the superiority of the proposed method over the competitive approaches. In particular, the proposed method can achieve up to 56\% accuracy gain on the CIFAR10 dataset when the bandwidth compression ratio is 1/48.
Information Theory
What problem does this paper attempt to address?
The paper aims to address the issue of semantic communication in wireless image transmission, specifically how to effectively extract the semantic information of images under limited communication conditions and accurately reconstruct this information at the receiving end. Existing semantic communication methods have limitations in the inference performance of downstream AI tasks (such as image recognition) or struggle to achieve a good balance between inference performance and reconstructed image quality. To this end, this paper proposes a semantic communication method based on Contrastive Learning (CL). By treating image degradation during the transmission process as a data augmentation operation, contrastive learning is used to reduce the semantic distance between the original image and the reconstructed image while maintaining the semantic distance between unrelated images, thereby improving the discriminability of downstream tasks. Additionally, a two-stage training procedure and corresponding loss functions are designed to jointly optimize the semantic encoder and decoder, achieving a good balance between image recognition performance and reconstruction quality. Experimental results show that the proposed method can achieve up to a 56% accuracy improvement on the CIFAR10 dataset when the bandwidth compression ratio is 1/48.