It has become common for AI techniques to generate visual symbols to provide inspiration for designers. Therefore, in this paper, two attention mechanism modules, SE and CA, are integrated into the FPN feature network to propose an improved visual symbol recognition algorithm. Structural improvements are made in the original style migration network, and style migration is utilized for visual symbol generation. Combining the recognition and style migration algorithms, an intelligent design system for streaming media posters is established. It can meet the requirements of easy and fast poster generation and real-time presentation. The results of simulation experiments show that the improved algorithm converges faster than the original algorithm loss function, and the number of training times is less. The recognition rate of both the improved visual symbol detection and style migration algorithms can reach more than 86%, with good recognition effect. The mean values of the streaming posters designed by applying the method of this paper are all above 3.6 points in the four aspects of symbol location appropriateness, overall color harmony, text readability and design creativity, respectively. Therefore, the design results have location diversity, color diversity, aesthetic properties and creativity.
