Abstract
With the rapid pace of modern urbanization, indoor landscape design has become increasingly important for enhancing the quality of indoor environments and overall user experience. Traditional aesthetic evaluation methods, which often rely on subjective human judgment, lack objectivity and efficiency. To address these limitations, this study proposes a deep learning-based framework for indoor landscape aesthetic evaluation. The proposed approach integrates convolutional neural networks (CNNs) and graph neural networks (GNNs) to extract and analyze both global and local aesthetic features from indoor landscape images. Experimental results on benchmark indoor landscape datasets demonstrate that our method achieves an accuracy of 97.74%, improving by 7.54% points compared to conventional approaches. In addition, the proposed model provides a 14.21% higher aesthetic score and a 10.6-point improvement in functional evaluation metrics. These findings highlight the potential of this CNN-GNN framework as a robust, objective, and efficient tool for indoor landscape aesthetic evaluation and design optimization.