With the rapid rise of the Internet of Things (IoT) and the widespread application of related binary network protocols in device-to-device communication, potential security and privacy concerns have gradually attracted attention. To address these issues, tools have been developed to identify and understand vulnerabilities in IoT systems. Most existing protocol security analysis techniques rely on a deep understanding of underlying communication protocols. This paper systematically proposes a Transformer-based neural network framework for the semantic reverse engineering of IoT binary protocols. This framework effectively utilizes the hidden correlations between the binary protocol implementations and communication functions in the IoT domain, enabling the extraction of complex patterns hidden within vast amounts of communication data. It establishes a mapping relationship between these patterns and semantic tokens, realizing the semantic reverse engineering capability for unknown IoT binary protocols. This improves the understanding of unknown IoT binary protocols across different communication scenarios and aids in discovering hidden protocol vulnerabilities and backdoor control commands, thereby enhancing the security of smart IoT systems.