As one of the representatives of Chinese traditional culture, Jiangnan garden is known for its unique landscape design and deep cultural connotation. Jiangnan gardens are not only the embodiment of garden art, but also incorporate elements such as natural landscape and humanistic history, forming a unique aesthetic mood. This study proposes a visual recognition model based on the improved YOLOv4 algorithm for recognizing key elements in the design of Jiangnan gardens and exploring the correlation between these elements and the aesthetic context of the gardens. First, the study constructs a rich dataset of Jiangnan garden elements through image acquisition and data enhancement techniques, which provides sufficient samples for training the target detection model. Subsequently, the model was optimized for the YOLOv4 model with a lightweight YOLOv4-tiny network, and the recognition accuracy and computational speed of the model were improved by introducing mosaic data enhancement, CSPNet structure, and cosine annealing learning rate. The experimental results show that the improved YOLOv4 model improves 5.19% in accuracy and 19.82% in detection speed compared to YOLOv4. Further Pearson correlation analysis and logistic regression modeling show that the sky and water elements have a significant impact on the aesthetic mood of Jiangnan gardens, with correlation coefficients exceeding 0.5, suggesting that the expansive skyline and water quality environment play a central role in enhancing the aesthetic perception of the gardens. This study quantifies the correlation between the elements of Jiangnan gardens and the aesthetic mood through the visual recognition technique, which provides new ideas for garden design.
