Examination of landscape element classification method based on knowledge of previous research

Toward the creation of judgment criteria for tagging in machine learning

Abstract

In this study, as a theory of landscape management assuming machine learning, I verified the creation of judgment criteria when tagging image data for buildings. First, I reviewed the research in the field of architecture, and then tried to devise an image data organizing method for use in machine learning. In conclusion, it was suggested that image data segmentation and schematization based on existing research accumulation could be used as judgment criteria for tagging. On the other hand, even if the organized tags were directly reflected in machine learning and operated, there were concerns that the machine would not learn well due to various circumstances such as image resolution and image distortion. Regarding this point, it is a problem found in the results of previous research that the author has been pursuing. However, even if the reason for the drop in detection rate is due to subdivided classification, it would be meaningful to add traceable subclass tags in the metadata. This is because, for example, even if the classification of image data is integrated as a large classification in order to improve the judgment accuracy, it is possible to restore the classification to a small classification according to the situation. In the future, it will be necessary to further verify the validity of tagging through trial and error in the practice of machine learning.