SemComInf: Semantic Communication System for Infrared Image Transmission

Infrared images, valued for their ability to capture thermal characteristics, are widely employed in applications such as military operations, forest fire prevention, and beyond. However, these scenarios often encounter challenging channel conditions. With the rapid advancement of deep learning (DL), semantic communication (SemCom) has emerged as a novel paradigm. By leveraging joint source-channel coding (JSCC), SemCom effectively mitigates channel noise while avoiding the cliff effect associated with traditional communication systems, positioning it as a potential framework for infrared image transmission. This paper proposes a feature extraction framework for infrared images, leveraging the Masked Autoencoder (MAE) architecture. We introduce a tailored masking strategy that prioritizes texture information and employs a multiscale encoder to extract infrared image features effectively. The proposed framework achieves image reconstruction at the receiver even when $75 \%$ of image patches are masked at the transmitter, resulting in a relatively low channel bandwidth ratio (CBR). Simulation results demonstrate that our system delivers superior performance in infrared image transmission, surpassing baseline methods, particularly in low signal-to-noise ratio (SNR) channels.