As an important part of China’s national intangible cultural heritage, Geng Lu Bu contains the voyaging spirit and the historical inheritance of quantitative cultivation of the sea. This paper proposes a multimodal machine translation technique that fuses visual information based on the visual representation of the perceptual text, and introduces a multimodal network to select and fuse the visual information. The generated image features and text features are deeply interconnected at multiple levels and granularities, thus enhancing the practical application effect of the whole multimodal machine translation system. Firstly, this paper conducts translation comparison experiments on two datasets, VATEX and MSVD-Turkish, respectively, and in the BLEU-4 metrics, the model proposed in this paper obtains BLEU scores of 37.43 and 37.72 on the two datasets, respectively. Finally, three different translation techniques: multimodal machine translation, human translation, and traditional translation, are used to collect the translated text data of Geng Lu Bu, and the effectiveness of the application of the translation model is analysed at the vocabulary and sentence length levels, respectively. The results show that the multimodal machine translation model proposed in this paper has a high degree of text sentence complexity, a high degree of text formality, and a well-proportioned and beautiful sentence style, which provides considerable reference value for the translation of the literature of Geng Lu Bu.