GEO DATA

Data Article

KOMPSAT-3/3A Image-text Dataset for Training Large Multimodal Models: Han Oh, Dong-Bin Shin, Dae-Won Chung; GEO DATA. 2025;7(1):27-35. Published online March 19, 2025; DOI: https://doi.org/10.22761/GD.2025.0003

516 View
37 Download

Abstract PDF: This study aims to improve the accuracy and interpretability of large multimodal models (LMMs) specialized in satellite image analysis by constructing an image-text dataset based on KOMPSAT-3/3A imagery and presenting the results of training using this dataset. Conventional LMMs are primarily trained on general images, limiting their ability to effectively interpret the specific characteristics of satellite imagery, such as spectral bands, spatial resolution, and viewing angles. To address this limitation, we developed an image-text dataset, divided into pretraining and finetuning stages, based on the existing KOMPSAT object detection dataset. The pretraining dataset consists of captions summarizing the overall theme and key information of each image. The fine-tuning dataset integrates metadata -including acquisition time, sensor type, and coordinates- with detailed object detection labels to generate six types of question-answer pairs: detailed descriptions, conversations with varying answer lengths, bounding box identification, multiple choice questions, and complex reasoning. This structured dataset enables the model to learn not only the general context of satellite images but also fine-grained details such as object quantity, location, and geographic attributes. Training with the new KOMPSAT-based dataset significantly improved the model’s accuracy in recognizing regional information and object characteristics in satellite imagery. Finetuned models achieved substantially higher accuracy than previous models, surpassing even the GPT-4o model and demonstrating the effectiveness of a domain-specific dataset. The findings of this study are expected to contribute to various remote sensing applications, including automated satellite image analysis, change detection, and object detection.

Articles

AI Dataset for Road Detection using KOMPSAT Images: Hoonhee Lee, Han Oh; GEO DATA. 2022;4(1):43-48. Published online March 31, 2022; DOI: https://doi.org/10.22761/DJ2022.4.1.005

1,846 View
55 Download

Abstract PDF: Information on shape and type of road present in an optical image of satellite is useful for digital mapping and monitoring of road changes. Processing and structuring optical image data collected from payloads mounted on KOMPSAT 3 and 3A can accelerate the development of road detection algorithms and the extraction of road information using them. In particular, if it is built with a learning dataset for AI (Artificial Intelligence) prepared to apply deep learning technology, the latest artificial intelligence technology in the field of computer science can be spun off to the field of satellite image-based road detection to attempt a wide range of analysis. Korea Aerospace Research Institute constructed an image dataset for AI learning using satellite optical images with Korean companies, and this paper explains the type and size of datasets along with examples of the use of the dataset. The established data can be used through the website, aihub.or.kr.

Dataset for Water Body Detection Using Satellite SAR Images: SeungJae Lee, Han Oh; GEO DATA. 2021;3(2):12-19. Published online July 21, 2021; DOI: https://doi.org/10.22761/DJ2021.3.2.002

1,635 View
54 Download

Abstract PDF: Satellite synthetic aperture radar (SAR) generates valid image information in all-weather. Thus, it can be effectively used for near real-time monitoring and damage analysis of flood areas which always involve overcast skies. Water body detection (WBD) using SAR images can be implemented by various techniques which discriminate electromagnetic characteristics between water and non-water areas. Especially, semantic segmentation exploiting artificial intelligence techniques can be used to develop a high-performance WBD model. To this end, Korea Aerospace Research Institute has built an WBD dataset using KOMPSAT-5 images. The dataset is currently available through the website, aihub.or.kr.

AI Training Dataset for Cloud Detection of KOMPSAT Images: Bo-Ram Kim, Han Oh; GEO DATA. 2020;2(2):56-62. Published online December 30, 2020; DOI: https://doi.org/10.22761/DJ2020.2.2.008

1,627 View
61 Download
1 Citations

Abstract PDF

Clouds that appear inevitably when acquiring optical satellite images hinder the interpretation of surface information, so removing them is a crucial procedure to increase the utilization of satellite images. Currently, for KOMPSAT (Korea Multi-purpose Satellite) images, only the cloud amount by visual measurement is proved for the entire scene and detailed cloud masks are not provided. Since cloud detection is a time-consuming task, we built a cloud dataset for KOMPSAT images so as to develop an algorithm that expedites the task with state-of-the-art artificial intelligent techniques. In the dataset, satellite images were selected from various regions considering that clouds have different characteristics depending on the region, and masks were classified into thin clouds, thick clouds, cloud shadows, and clear sky. The size of dataset is over 4,000 image/mask pairs by an image size of 1000x1000 and one of the largest among publicly available cloud datasets, as of this writing. The dataset is built by a government AI (artificial intelligent) training dataset building program and will be available through the website, aihub.or.kr.

Citations

Citations to this article as recorded by

Cloud Detection Using a UNet3+ Model with a Hybrid Swin Transformer and EfficientNet (UNet3+STE) for Very-High-Resolution Satellite Imagery
Jaewan Choi, Doochun Seo, Jinha Jung, Youkyung Han, Jaehong Oh, Changno Lee
Remote Sensing.2024; 16(20): 3880. CrossRef

First
Prev
Page of 1
Next
Last

Search