| Title |
A Study on the Production of Level 3 Land Cover Maps using Deep Learning |
| Authors |
서봉상(BongSang Seo) ; 서병석(ByungSuk Seo) |
| DOI |
https://doi.org/10.5573/ieie.2025.62.11.145 |
| Keywords |
Deep learning; Level 3 classification; Land cover map; Semantic segmentation; Aerial orthoimage |
| Abstract |
This study aims to automate and improve the accuracy of the production of level 3 land cover maps for establishing high-precision spatial information. The existing manual-based map production method has limitations such as inconsistent quality, high labor costs, and long time required. In particular, the 1:5,000 scale level 3 land cover maps consists of 41 items, making accurate classification difficult. To address this issue, this study proposes an automated level 3 land cover maps production method using a deep learning-based semantic segmentation technique. Land cover attributes were classified based on aerial orthoimages using the D-LinkNet and DeepLabV3+ models. Experimental results showed that the amount and diversity of training data significantly affected the classification accuracy of both models, with D-LinkNet showing strength in linear structure classification and DeepLabV3+ showing strength in segmenting objects of various sizes and shapes. This study presents an effective technical alternative for rapid updating and maintaining high quality of level 3 land cover maps. |