Research on Architectural Design Method Based on Machine Vision Fusion With Deep Learning

ABSTRACT

In order to enhance the intelligence level of architectural design, a design method based on machine vision and deep learning fusion is proposed. Firstly, the overall framework of the method was constructed, and the original building images were processed by machine vision to form color features, texture features, and shape features. Among them, the color features of building images are obtained by extracting color histograms, the texture features of building images are obtained by calculating the energy and entropy of the gray level co‐occurrence matrix, and the shape features of building images are obtained by calculating Hu moments. Secondly, a CNN deep learning framework is constructed by replacing VGG16 with VGG19, which has better performance. Deep convolution and pointwise convolution are used instead of conventional convolution to improve the efficiency of deep learning. The color features, texture features, and shape features of building images processed by machine vision are incorporated into deep learning based on VGG19 and fused to generate building design schemes. During the experiment, the V19‐CNN method and IV19‐CNN method were used as reference methods to design building components, building parts, and building as a whole, respectively. The experimental results show that our method has better performance compared to the V19‐CNN method and the IV19‐CNN method. This is because our method uses machine vision to obtain richer features, replacing direct input of the original building images. From the quantitative results, our method achieved an accuracy of over 90% in both local and overall design experiments, which is significantly better than the two reference methods.