Cognitive semantic communications with codebook-based adaptive correction for image classification

Abstract

Conventional bit-oriented communication paradigms face several critical challenges in emerging sixth-generation (6G) wireless scenarios. One of the key challenges is low-efficient symbol transmission without considering task-oriented semantic importance, thus leading to high bandwidth consumption. To overcome this challenge, we propose a novel cognitive semantic communication (CSC) framework for the image classification, where the remote facial expression recognition for drivers is considered. Unlike conventional semantic communication methods that passively handle predefined semantic features, the proposed CSC framework actively identifies and adapts the most critical emotional features for transmission, substantially reducing bandwidth and signal quality requirements while preserving essential expression details. Moreover, for the first time, we design a cognitive-inspired semantic inference model equipped with a codebook-based correction mechanism to mitigate semantic distortion caused by physical channel impairments, enhancing reliability and interpretability under adverse conditions. Extensive simulation results demonstrate that proposed CSC can significantly outperform existing methods in recognition accuracy under constrained communication scenarios. By integrating cognitive-inspired semantic extraction, intelligent encoding–decoding processes, and context-aware inference, our proposed approach advances the state of the art in semantic communications for the image classification, offering a more robust and resource-efficient solution.