The spread of Chinese martial arts is crucial for the world to understand Chinese culture. If only relying on one transmission method, it will lead to the difference of transmission and its lack of certain real time. This will lead to differences in the understanding of Chinese martial arts, which is also not conducive to the spread of Chinese glorious culture. Cross-media communication technology can solve this communication difference problem very well. The deep neural network method was used to fuse relevant features of Chinese martial arts, and it also analyzes the feasibility of neural network technology in cross-media communication. At the same time, this study uses deep neural network to study the timeliness of Chinese martial arts in the process of cross-media communication. The research results show that the convolutional neural network can effectively extract the characteristics of Chinese martial arts and carry out effective dissemination. However, the hybrid convolutional neural network with temporal features has higher accuracy in extracting Chinese martial arts features. This hybrid convolutional neural network is more conducive to the dissemination of Chinese martial arts through cross-media technology, which can ensure its timeliness. The maximum error of deep neural network technology in predicting Chinese martial arts culture is only 2.67%. This part of the error comes from the action characteristics of Chinese martial arts culture, which shows that neural network technology has good feasibility.