The unique style of early works of fine art depends on the infinite derivation of the image factor symbols metaphorically behind them, so the symbolic interpretation of early works of fine art is a search for the missing beauty of contemporary art. In this paper, we start from multi-scale association rules, use Gaussian pyramid and cubic convolution methods to extract image metaphor features in fine art works, and weight the features. Based on Putschke’s emotion classification, a multimodal metaphor dataset is constructed. The symbolization method is used to represent the symbolic emotions in the art works, and then the bidirectional Bi-LSTM model is used for recognition. On this basis, using the model constructed in this paper, the symbolic and imagery features of art works are analyzed, and the symbolic techniques of art works are interpreted from three perspectives: brushstroke, color, and spatial metaphor. Structural metaphors in fine art works contain 40% neutral emotions, followed by 36.4% positive emotions. Stroke A has a higher mean score at 3 and 4 line sample points, respectively, of -1.1646 and -1.1106, and the emotions triggered by these two line samples are more significant. Interpreting early art works can help enhance the aesthetics of contemporary art for modern audiences.