Against the backdrop of the increasingly prominent issue of mental health among college students worldwide, exploring environmental intervention pathways to alleviate mental fatigue is of significant importance. This study constructs data using Street View Images (SVI) and the Perceived Restorative Scale-11 (PRS-11), and employs Trueskill and CNN-BiLSTM models to extract restorative perception data for analyzing the attention restoration status of teachers and students. By applying XGBoost regression and SHAP explanation methods, the study explores the nonlinear effects of seven landscape indicators, such as openness, water view, and paving degree, on attention restoration, and reveals the marginal effects and interactions of key indicators. The findings indicate that: (1) Various environmental elements play a role in attention restoration, with the ranking of contribution degrees showing that pavement coverage rate, color complexity, and openness are important indicators affecting restoration quality; (2) There are interaction effects among environmental elements in terms of attention restoration, collectively influencing restorative outcomes, suggesting that landscape design should emphasize the synergistic optimization of these elements; (3) The effects of various environmental factors on attention restoration generally exhibit nonlinear marginal effects and fall within certain optimal ranges, indicating that changes in the input of these factors and improvements in restoration outcomes are not in a simple linear correspondence. Based on the study of the mechanism by which campus landscape elements influence attention restoration, it is concluded that coordinated design among various environmental elements is essential, providing a scientific basis for creating healthy and safe campus environments.