In this paper, we are interested in how computers can be used to better serve us humans, such as helping humans control their nutrient intake, with higher level shortcuts. Specifically, the neural network model was used to help humans identify and analyze the content and proportion of nutrients in daily food intake, so as to help humans autonomously choose and reasonably match diets. In this study, we formed the program we wanted to obtain by establishing four modules, in which the imagination module sampled the environment, then relied on the encoder to extract the implicit features of the image, and finally relied on the decoder to obtain the required feature vector from the implicit features, and converted it into the battalion formation table information through the semantic output module. Finally, the model achieved extremely high accuracy on recipe1M+ and food2K datasets.