The analysis of landscape design and plant selection under deep learning

This paper explores the application of deep learning (DL) techniques in landscape design and plant selection, aiming to enhance design efficiency and quality through automated plant leaf image recognition (PLIR). A novel framework based on Convolutional Neural Network (CNN) and Fully Convolutional Network (FCN) is proposed. This framework integrates multi-scale feature fusion, attention mechanisms, and object detection technologies to improve the recognition of landscape elements and the selection of plant leaves. Experimental results demonstrate that the proposed DL framework significantly improves performance in landscape element classification tasks. Specifically, the enhanced FCN model achieves a 4.5% improvement in classification accuracy on the Sift Flow dataset, while fine-grained PLIR accuracy increases by 4.8%. Furthermore, the strategy combining object detection and FCN-based image segmentation further boosts accuracy, reaching 90.4% and 88.7%, respectively. These results validate the model's effectiveness in practical simulations, highlighting its innovative contribution to the digitalization and intelligent advancement of landscape design. The key innovation of this paper lies in the first application of multi-scale feature fusion and attention mechanisms within the FCN model, effectively improving the segmentation capability of complex landscape images. Moreover, background noise interference is reduced by using object detection techniques. Additionally, a domain-adaptive transfer learning strategy and region-weighted loss function are designed, further enhancing the model's accuracy and robustness in plant classification tasks. Through the application of these technologies, this paper not only advances the field of landscape design but also provides technical support for biodiversity conservation and sustainable urban planning.