Aesthetic Evaluation Framework for Plant Color Schemes for Sustainable Landscape Architecture Design

Plant color exerts a profound influence in both ecological function and human perception. However, current aesthetic evaluations in landscape design rely heavily on subjective judgment, lacking quantitative and standardized evaluation metrics. To address this gap, we propose a data-driven framework for aesthetic evaluation that leverages Contrastive Learning (CL) to automatically identify aesthetically pleasing color schemes from landscape images, overcoming the limitations of traditional methods over-reliance on expert knowledge and rule-based summaries. Unlike previous methods, our end-to-end approach directly processes the entire image, capturing all visual elements for a comprehensive aesthetic evaluation, leveraging a pre-trained Convolutional Neural Network (CNN) based on ResNet. On the test dataset, our model reached an Area Under the ROC Curve (AUC) value of 0.8377, indicating a high consistent with the human-based evaluation. This approach offers an empirical model that integrates into design workflows, providing real-time feedback and improving efficiency.