Research on Generative Design of Pocket Park Layout Based on Machine Learning

With the rapid pace of urbanization, the demand for urban public green space is becoming increasingly evident, and pocket parks are emerging as an ideal solution in high-density urban environments due to their notable advantages, including small footprint, diverse forms, and flexible functions. However, traditional pocket park design processes face challenges such as short design cycles and time-consuming multi-program conceptualization. In response, this paper proposes an automatic generation method for pocket park layout plans based on generative adversarial networks (GANs). By constructing a training dataset consisting of 119 pairs of finely labeled images, a model capable of generating pocket park layout plans is successfully trained. The performance of the model is comprehensively and meticulously evaluated from both quantitative and qualitative perspectives to validate the feasibility and efficiency of the method. Experimental results show that the proposed method rapidly generates a large number of pocket park layouts with high design quality and innovation, significantly improving design efficiency.