Deep learning landscape evaluation system integrating poetic emotion and visual features

This study aims to quantitatively measure emotions in landscape designs using ancient poetry while developing a corresponding evaluation system. A high-quality poetry dataset covering various emotional types and dynasties was selected. Chinese-CLIP models were used to extract the visual features of the poetry, and a neural network model was trained for emotional analysis of landscape designs. The evaluation system used a self-trained neural network and the CLIP model based on the emotions conveyed in poetry. Heatmaps were produced using an attention mechanism to show the regions the model focuses on, enabling focused design optimization. The experimental findings show that this evaluation system provides quantitative emotional evaluations of landscape designs and may successfully direct the optimization process.