Classification of Vernacular Landscape Elements and Design Applications Based on Deep Learning Fully Convolutional Neural Networks

The integration of vernacular landscape elements is a crucial aspect of contemporary garden landscape planning and design, serving as a significant means of expressing traditional culture and evoking folk nostalgia. The external regional, ecological, and symbolic, as well as the internal cultural aspects of vernacular landscape elements, directly influence the overall appearance of residential planning. However, landscape image processing is still largely confined to the designer’s professional cognition and experience. Traditional methods for classifying landscape elements primarily rely on manual processing. By integrating an advanced convolutional neural network (CNN) architecture, this study demonstrates its practical utility through rigorous experiments. This paper presents squeeze‐and‐excitation (SE)‐DenseNet, a deep‐learning model that uses SE blocks to classify vernacular landscape elements. It achieves 97.3% accuracy, outperforming traditional methods. This reduces manual work and helps integrate cultural elements into modern design.