<p>The integration of Artificial Intelligence (AI) into landscape architecture has enabled advanced approaches for evaluating ecological and aesthetic qualities in design. This research presents a Deep Reinforcement Learning (DRL) approach to determine the aesthetic value and environmental performance of landscape-based designs for sustainable landscape planning. An ecological landscape image dataset contains 10,200 images, such as urban parks, water bodies, and mixed landscapes. Preprocessing employs image resizing and normalizes them to have consistency and model robustness. To extract visual and ecological features, the Visual Geometry Group (VGG16) was used to extract the following features: green coverage ratio, sky openness, vegetation diversity, spatial composition, and edge density. The described features were further incorporated into an Enriched Fish School Search-driven Double Deep Q-Network (EFSS-Double DQN), which uses an intelligent agent to respond to landscape feature states and subsequently optimize aesthetic assessment scores using rewards. The association between landscape features and aesthetic score is more accurately learned by Double DQN. EFSS enhances the optimal combination of features, which has a strong impact on aesthetic quality. To enhance transparency, SHAP-based XAI is integrated to interpret model decisions by quantifying feature contributions. According to the experimental findings applied to Python, the EFSS-Double DQN model is predicting the aesthetic quality scores and identifying the most important factors that influence the perception of a landscape, especially with an MAE of 0.12 and an accuracy of 0.93. Overall, the research concludes that the proposed method provides a robust and interpretable approach for optimizing landscape design aesthetics and ecological performance.</p> Graphical abstract <p></p>