구름은 대기 중에 떠 있는 작은 물방울이나 얼음 알갱이들 또는 혼합물 등으로 구성되며 지구 표면의 약 2/3를 덮고 있다. 위성영상내에서의 구름은 일부 다른 지상 물체 또는 지표면과 유사한 반사도 특성으로 인해 구름과 구름이 아닌 영역을 분리하는 구름탐지는 매우 어려운 작업이다. 특히 뚜렷한 특징을 가지는 두꺼운 구름과 달리 얇은 반투명 구름은 위성영상내에서 구름과 배경의 대비가 약하고 지표면과 혼합되어져 나타나 기 때문에 대부분 구름탐지에서 쉽게 놓쳐지고 많은 어려움을 주는 대상으로 작용한다. 이러한 구름탐지의 반 투명 구름의 한계점을 극복하기 위해, 본 연구에서는 머신러닝 기법(Random Forest [RF], Convolutional Neural Networks [CNN])을 활용하여 반투명 구름을 중점으로 한 구름탐지 연구를 수행하였다. Reference자료로는 MOderate Resolution Imaging Spectroradiometer (MODIS)에서 제공하는 MOD35자료에서 Cloud Mask와 Cirrus Mask를 활용하였으며 반투명 구름 픽셀을 고려한 모델 훈련을 위해 훈련 데이터의 픽셀 비율을 구름, 반투명 구름, 청천이 약 1:1:1이 되도록 구성하였다. 연구의 정성적 비교 결과, RF와 CNN 모두 반투명 구름을 포함한 다양한 형태의 구름 등을 잘 탐지하였고, RF 모델 결과와 CNN 모델 결과를 혼합한 RF+CNN경우에는 개별 모 델의 한계점을 개선시키며 구름탐지가 잘 수행되어진 것을 확인하였다. 연구의 정량적 결과 RF의 전체 정확도 (OA) 값은 92%, CNN은 94.11%를 보였고, RF+CNN은 94.29%의 정확도를 보였다.

This study was conducted to evaluate the potential of a multimodal model for land cover classification. The performance of the Clipseg multimodal model was compared with two unimodal models including Convolutional Neural Network (CNN)-based Unet and Transformer-based Segformer for land cover classification. Using orthophotos of two areas (Area1 and Area2) in Wonju City, Gangwon Province, classification was performed for seven land cover categories (Forest, Cropland, Grassland, Wetland, Settlement, Bare Land, and Forestry-managed Land). The results showed that the Clipseg model demonstrated the highest generalization performance in new environments, achieving the highest accuracy among the three models with an Overall Accuracy of 83.9% and Kappa of 0.72 in the test area (Area2). It performed particularly well in classifying Forest (F1-Score 94.7%), Cropland (78.0%), and Settlement (78.4%). While Unet and Segformer models showed high accuracy in the training area (Area1), they exhibited limitations in generalization ability with accuracy decreases of 29% and 20% respectively in the test area. The Clipseg model required the most parameters (approximately 150 million) and the longest training time (10 hours 48 minutes) but showed stable performance in new environments. In contrast, Segformer achieved considerable accuracy with the least parameters (about 16 million) and the shortest training time (3 hours 21 minutes), demonstrating its potential for use in resource-limited environments. This study shows that image-text-based multimodal models have a high potential for land cover classification. Their superior generalization ability in new environments suggests they can be effectively applied to land cover classification in various regions. Future research could further improve classification accuracy through model structure improvements, addressing data imbalances, and additional validation in diverse environments.

As the use of drone-based hyperspectral data increases, atmospheric correction methods that are suitable for the characteristics of drone data are required. In this letter, we evaluated the performance of the drone atmospheric correction (DROACOR) model for atmospheric correction of drone hyperspectral data by comparing it with the results of empirical line correction. The results using the normalized difference vegetation index and Fe3+ index showed a high correlation (R2=0.99) between the two correction models in the visible/near-infrared (VNIR). However, the DROACOR model estimated the overall reflectance of kaolin in the short-wave infrared (SWIR) relatively low and the characteristic absorption depth occurring around 2200 nm weakly. These results show the high performance of the DROACOR model in VNIR and its limitations in SWIR.