With the rapid progress of modern information technology and the explosive growth in the number of IoT devices, there is an urgent need to propose new communication paradigms in order to cope with the growing communication demands. Semantic communication systems show great potential in improving end-to-end transmission performance and have a broad research prospect in future communication. However, deep learning-based semantic communication has certain security concerns, and tiny attacks can cause deep learning models to make wrong judgments, which may hinder the practical deployment of semantic communication in the future. In this paper, we propose a deep learning-based multilevel semantic communication system with joint transmission to accomplish downstream tasks. Considering the real scenarios that semantic communication systems face in practical applications, we incorporate channel effects and counter attacks. Through experiments, this system is proved to be resistant to attacks, which provides insights for designing more robust semantic communication systems.