국가기본도와 항공정사영상을 활용한 객체 탐지 데이터셋 구축 연구

With the development of the artificial intelligence industry, many deep learning models have been released as open sources, creating an environment where anyone can easily utilize deep learning. However, in order to build enough training data to use a deep learning model, a lot of time and money are consumed, so there is a limit to its utilization. In particular, most studies on object detection in aerial orthographic images are constructing training data manually. This way is not only time consuming and costly, but also has disadvantages in that data is constructed differently depending on operator's error, mistakes, and subjective judgments. In order to solve this problem, this study proposes a method for producing training data using the pre-constructed data. This study utilizes the National Base Map and the 25cm pixel size aerial orthographic images, provided by the National Geographic Information Institute, for producing object detection training data. By training object detection model using training data that constructed according to the presented methodology, we verify the effectiveness of the training data.