Semantic Communication Enabled Lightweight Deep Learning Framework for Adaptive Image Transmission in Resource Constrained Iot Networks

Abstract

The rapid development of Internet of Things (IoT) applications puts forward demands on efficient image transmission under the constraints of limited bandwidth, energy, and computation. Existing deep joint source-channel coding schemes and traditional sourcechannel coding schemes usually suffer from high complexity and poor adaptability to dynamic wireless environments. To address these issues, a lightweight semantic communication framework is proposed that combines semantic feature extraction, feature compression, and channel-aware adaptive transmission. A lightweight encoder based on depthwise separable convolution is proposed to extract compact semantic representations. An adaptive transmission module is designed to dynamically adjust the amount of transmitted semantic information according to the channel conditions. At the receiver, a semantic decoder reconstructs images preserving important visual semantics. The framework is evaluated over AWGN channels with SNR ranging from -5 dB to 20 dB on CIFAR-10, STL-10, Kodak and ImageNet datasets. The experimental results show that the proposed method has a PSNR of 34.2 dB, SSIM of 0.941, LPIPS of 0.110 and semantic similarity of 0.958. Furthermore, the proposed framework can be applied for resourceconstrained intelligent IoT and future 6G communication networks, since it significantly reduces model complexity, memory consumption and energy usage compared with existing DeepJSCC-based approaches.