[4F3-GS-10n] AI応用：建設・不動産

[4F3-GS-10n-02] Automatic building structure classification of single-family houses using exterior images

〇Masashi Inoue¹, Yuki Moriya¹ (1. Tohoku Institute of Technology)

Keywords:Building structure, Pre-trained model, LIFULL HOME'S dataset, City, Exterior

There are various types of building structures, such as wooden and steel structures, with different characteristics. However, it is not easy to classify these building structures visually. In this research, we use a pre-trained general-purpose image recognition model based on deep learning to automatically classify building structures from exterior images. By automating the process, it is possible to add building structure information to a large amount of building data. We constructed a classification model by fine-tuning pre-trained models (VGG-16 and VGG-19) for detached rental houses in a real estate information service. When the number of layers of the neural network used for classification is large, the training process takes a long time. If classification performance similar to that of a pre-trained model can be obtained with a small number of layers, the time required for training can be reduced. Therefore, we used our original CNN with a reduced number of layers to search for conditions that would allow us to achieve a classification rate similar to that of a pre-trained model.