Design Research On Pocket Park Plan Layout Generation Based On Deep Learning

Pocket park design is a fundamental and creative work in the process of park construction,and its quality directly determines the effectiveness of park construction.The traditional design method requires designers to participate in the analysis,creation,and expression throughout the design process,with computers only serving as auxiliary tools.However,with the advent of the era of artificial intelligence,machines can extract useful information and patterns from data through algorithms,forming "data knowledge" for machine research on problems.By constructing a knowledge graph and combining machine learning related algorithms to train models,this research aims to enable machines to autonomously learn and master the layout experience of pocket park design,and to achieve automated generation of initial sketches or concept maps of pocket park design to save design time and costs,and provide more design ideas and inspiration for landscape designers.At the same time,this research will also provide reference and inspiration for landscape generation design based on deep learning technology,to change the traditional pattern of landscape analysis and design,enabling designers to play a more important role in creativity and communication,and promote the development of landscape design towards intelligence and automation.To achieve the goal of this research,the following research work has been completed:firstly,by reviewing the research and literature on artificial intelligence technology in the field of landscape layout generation,the theoretical basis of the experiment has been established.Then,through the analysis of four pocket park cases at home and abroad and research on traditional design methods,the elements and features of pocket park design under different types and spatial organization modes have been analyzed and discussed,to lay a theoretical and data foundation for using deep learning technology to automatically generate pocket park design layouts.Subsequently,experiments were conducted,proposing a topology-based data augmentation strategy to address the small sample problem unique to landscape design,effectively improving sample diversity,and recording the experiment process in detail.Based on the experimental results,the experimental steps were continuously optimized to enable machines to better learn the intrinsic rules and representation hierarchy of the samples.Finally,the feasibility of the application of the deep learning-based generation design method was demonstrated with three pocket park examples as comparison objects.The experimental results show that:(1)Machines can generate relatively reasonable landscape layout plans in a short period of time based on existing original site information and surrounding environmental constraints.(2)By employing appropriate data processing and data augmentation techniques,machines are capable of deriving universally applicable design principles from limited samples and applying them to address novel scenarios in landscape architecture.With the rapid development and continuous iteration of deep learning technology,landscape architecture has encountered new opportunities and challenges,while also bringing more possibilities to the design field.In the future,the accuracy and reliability of models can be further improved and applied to a wider range of scenarios,exploring diversified and innovative design ideas.In this process,it is necessary to value humanmachine collaborative design,leverage the advantages of machine-assisted design,and combine human creative thinking with machine computing capabilities to achieve personalized design,thereby enhancing the applicability and user experience of the design.