• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.40 no.5
• /
• pp.381-391
• /
• 2022

This paper is intended to find one of the prevailing deep learning models that are a type of AI (Artificial Intelligence) that helps rapidly detect damaged buildings where disasters occur. The models selected are SSD-512, RetinaNet, and YOLOv3 which are widely used in object detection in recent years. These models are based on one-stage detector networks that are suitable for rapid object detection. These are often used for object detection due to their advantages in structure and high speed but not for damaged building detection in disaster management. In this study, we first trained each of the algorithms on xBD dataset that provides the post-disaster imagery with damage classification labels. Next, the three models are quantitatively evaluated with the mAP(mean Average Precision) and the FPS (Frames Per Second). The mAP of YOLOv3 is recorded at 34.39%, and the FPS reached 46. The mAP of RetinaNet recorded 36.06%, which is 1.67% higher than YOLOv3, but the FPS is one-third of YOLOv3. SSD-512 received significantly lower values than the results of YOLOv3 on two quantitative indicators. In a disaster situation, a rapid and precise investigation of damaged buildings is essential for effective disaster response. Accordingly, it is expected that the results obtained through this study can be effectively used for the rapid response in disaster management.

• Journal of Korean Society of Forest Science
• /
• v.111 no.4
• /
• pp.603-612
• /
• 2022

The study aims were to evaluate a machine-learning, algorithm-based, forest biomass-estimation model to estimate subnational forest biomass and to comparatively analyze REDD+ forest reference emission levels. Time-series Landsat satellite imagery and ESA Biomass Climate Change Initiative information were used to build a machine-learning-based biomass estimation model. The k-nearest neighbors algorithm (kNN), which is a non-parametric learning model, and the tree-based random forest (RF) model were applied to the machine-learning algorithm, and the estimated biomasses were compared with the forest reference emission levels (FREL) data, which was provided by the Paraguayan government. The root mean square error (RMSE), which was the optimum parameter of the kNN model, was 35.9, and the RMSE of the RF model was lower at 34.41, showing that the RF model was superior. As a result of separately using the FREL, kNN, and RF methods to set the reference emission levels, the gradient was set to approximately -33,000 tons, -253,000 tons, and -92,000 tons, respectively. These results showed that the machine learning-based estimation model was more suitable than the existing methods for setting reference emission levels.

• Korean Journal of Remote Sensing
• /
• v.39 no.3
• /
• pp.269-282
• /
• 2023

The purpose of this study was to construct land cover maps using a deep learning model and to select the optimal deep learning model for land cover classification by adjusting the dataset such as input image size and Stride application. Two types of deep learning models, the U-net model and the DeeplabV3+ model with an Encoder-Decoder network, were utilized. Also, the combination of the two deep learning models, which is an Ensemble model, was used in this study. The dataset utilized RapidEye satellite images as input images and the label images used Raster images based on the six categories of the land use of Intergovernmental Panel on Climate Change as true value. This study focused on the problem of the quality improvement of the dataset to enhance the accuracy of deep learning model and constructed twelve land cover maps using the combination of three deep learning models (U-net, DeeplabV3+, and Ensemble), two input image sizes (64 × 64 pixel and 256 × 256 pixel), and two Stride application rates (50% and 100%). The evaluation of the accuracy of the label images and the deep learning-based land cover maps showed that the U-net and DeeplabV3+ models had high accuracy, with overall accuracy values of approximately 87.9% and 89.8%, and kappa coefficients of over 72%. In addition, applying the Ensemble and Stride to the deep learning models resulted in a maximum increase of approximately 3% in accuracy and an improvement in the issue of boundary inconsistency, which is a problem associated with Semantic Segmentation based deep learning models.

• Korean Journal of Remote Sensing
• /
• v.34 no.2_1
• /
• pp.267-282
• /
• 2018

Despite the continuous development of phenology detection studies using satellite imagery, verification through comparison with the field observed data is insufficient. Especially, in the case of Korean forests patching in various forms, it is difficult to estimate the start of season (SOS) by using only satellite images due to resolution difference. To improve the accuracy of vegetation phenology estimation, this study reconstructed the large scaled forest type map (1:5,000) with MODIS pixel resolution and produced time series vegetation phenology curves from Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) derived from MODIS images. Based on the field observed data, extraction methods for the vegetation indices and SOS for Korean forests were compared and evaluated. We also analyzed the correlation between the composition ratio of forest types in each pixel and phenology extraction from the vegetation indices. When we compared NDVI and EVI with the field observed SOS data from the Korea National Arboretum, EVI was more accurate for Korean forests, and the first derivative was most suitable for extracting SOS in the phenology curve from the vegetation index. When the eight pixels neighboring the pixels of 7 broadleaved trees with field SOS data (center pixel) were compared to field SOS, the forest types of the best pixels with the highest correlation with the field data were deciduous forest by 67.9%, coniferous forest by 14.3%, and mixed forest by 7.7%, and the mean coefficient of determination ($R^2$) was 0.64. The average national SOS extracted from MODIS EVI were DOY 112.9 in 2014 at the earliest and DOY 129.1 in 2010 at the latest, which is about 0.16 days faster since 2003. In future research, it is necessary to expand the analysis of deciduous and mixed forests' SOS into the extraction of coniferous forest's SOS in order to understand the various climate and geomorphic factors. As such, comprehensive study should be carried out considering the diversity of forest ecosystems in Korea.

• Journal of the Korean Institute of Landscape Architecture
• /
• v.37 no.3
• /
• pp.46-53
• /
• 2009

Urbanization brings several changes to the natural environment. Its consequences can have a direct effect on climatic features, as in the Urban Heat Island Effect. One factor that directly affects the urban climate is the green area. In urban areas, vegetation is suppressed in order to accommodate manmade buildings and streets. In this paper we analyze the effect of green areas on the urban temperature in Seoul. The period selected for analysis was July 30th, 2007. The ground temperature was measured using Landsat TM satellite imagery. Land cover was calculated in terms of city area, water, bare soil, wet lands, grass lands, forest, and farmland. We extracted the surface temperature using the Linear Regression Model. Then, we did a regression analysis between air temperature at the Automatic Weather Station and surface temperature. Finally, we calculated the temperature decrease area and the population benefits from the green areas. Consequently, we determined that a green area with a radius of 500m will have a temperature reduction area of $67.33km^2$, in terms of urban area. This is 11.12% of Seoul's metropolitan area and 18.09% of the Seoul urban area. We can assume that about 1,892,000 people would be affected by this green area's temperature reduction. Also, we randomly chose 50 places to analysis a cross section of temperature reduction area. Temperature differences between the boundaries of green and urban areas are an average of $0.78^{\circ}C$. The highest temperature difference is $1.7^{\circ}C$, and the lowest temperature difference is $0.3^{\circ}C$. This study has demonstrated that we can understand how green areas truly affect air temperature.

• Korean Journal of Remote Sensing
• /
• v.37 no.5_1
• /
• pp.989-998
• /
• 2021

In the utilization of optical satellite imagery, which is greatly affected by clouds, periodic composite technique is a useful method to minimize the influence of clouds. Recently, a technique for selecting the optimal pixel that is least affected by the cloud and shadow during a certain period by directly inputting cloud and cloud shadow information during period compositing has been proposed. Accurate extraction of clouds and cloud shadowsis essential in order to derive optimal composite results. Also, in the case of an surface targets where spectral information is important, such as crops, the loss of spectral information should be minimized during cloud-free compositing. In thisstudy, clouds using two spectral indicators (Haze Optimized Tranformation and MeanVis) were used to derive a detection technique with low loss ofspectral information while maintaining high detection accuracy of clouds and cloud shadowsfor cabbage fieldsin the highlands of Gangwon-do. These detection results were compared and analyzed with cloud and cloud shadow information provided by Sentinel-2A/B. As a result of analyzing data from 2019 to 2021, cloud information from Sentinel-2A/B satellites showed detection accuracy with an F1 value of 0.91, but bright artifacts were falsely detected as clouds. On the other hand, the cloud detection result obtained by applying the threshold (=0.05) to the HOT showed relatively low detection accuracy (F1=0.72), but the loss ofspectral information was minimized due to the small number of false positives. In the case of cloud shadows, only minimal shadows were detected in the Sentinel-2A/B additional layer, but when a threshold (= 0.015) was applied to MeanVis, cloud shadowsthat could be distinguished from the topographically generated shadows could be detected. By inputting spectral indicators-based cloud and shadow information,stable monthly cloud-free composited vegetation index results were obtained, and in the future, high-accuracy cloud information of Sentinel-2A/B will be input to periodic cloud-free composite for comparison.

• Korean Journal of Remote Sensing
• /
• v.39 no.6_1
• /
• pp.1211-1224
• /
• 2023

Clouds cause many difficult problems in observing land surface phenomena using optical satellites, such as national land observation, disaster response, and change detection. In addition, the presence of clouds affects not only the image processing stage but also the final data quality, so it is necessary to identify and remove them. Therefore, in this study, we developed a new cloud detection technique that automatically performs a series of processes to search and extract the pixels closest to the spectral pattern of clouds in satellite images, select the optimal threshold, and produce a cloud mask based on the threshold. The cloud detection technique largely consists of three steps. In the first step, the process of converting the Digital Number (DN) unit image into top-of-atmosphere reflectance units was performed. In the second step, preprocessing such as Hue-Value-Saturation (HSV) transformation, triangle thresholding, and maximum likelihood classification was applied using the top of the atmosphere reflectance image, and the threshold for generating the initial cloud mask was determined for each image. In the third post-processing step, the noise included in the initial cloud mask created was removed and the cloud boundaries and interior were improved. As experimental data for cloud detection, CAS500-1 L2G images acquired in the Korean Peninsula from April to November, which show the diversity of spatial and seasonal distribution of clouds, were used. To verify the performance of the proposed method, the results generated by a simple thresholding method were compared. As a result of the experiment, compared to the existing method, the proposed method was able to detect clouds more accurately by considering the radiometric characteristics of each image through the preprocessing process. In addition, the results showed that the influence of bright objects (panel roofs, concrete roads, sand, etc.) other than cloud objects was minimized. The proposed method showed more than 30% improved results(F1-score) compared to the existing method but showed limitations in certain images containing snow.

• Korean Journal of Remote Sensing
• /
• v.39 no.5_3
• /
• pp.1009-1029
• /
• 2023

Urban trees play a vital role in urban ecosystems,significantly reducing impervious surfaces and impacting carbon cycling within the city. Although previous research has demonstrated the efficacy of employing artificial intelligence in conjunction with airborne light detection and ranging (LiDAR) data to generate urban tree information, the availability and cost constraints associated with LiDAR data pose limitations. Consequently, this study employed freely accessible, high-resolution multispectral satellite imagery (i.e., Sentinel-2 data) to estimate fractional tree canopy cover (FTC) within the urban confines of Suwon, South Korea, employing machine learning techniques. This study leveraged a median composite image derived from a time series of Sentinel-2 images. In order to account for the diverse land cover found in urban areas, the model incorporated three types of input variables: average (mean) and standard deviation (std) values within a 30-meter grid from 10 m resolution of optical indices from Sentinel-2, and fractional coverage for distinct land cover classes within 30 m grids from the existing level 3 land cover map. Four schemes with different combinations of input variables were compared. Notably, when all three factors (i.e., mean, std, and fractional cover) were used to consider the variation of landcover in urban areas(Scheme 4, S4), the machine learning model exhibited improved performance compared to using only the mean of optical indices (Scheme 1). Of the various models proposed, the random forest (RF) model with S4 demonstrated the most remarkable performance, achieving R² of 0.8196, and mean absolute error (MAE) of 0.0749, and a root mean squared error (RMSE) of 0.1022. The std variable exhibited the highest impact on model outputs within the heterogeneous land covers based on the variable importance analysis. This trained RF model with S4 was then applied to the entire Suwon region, consistently delivering robust results with an R² of 0.8702, MAE of 0.0873, and RMSE of 0.1335. The FTC estimation method developed in this study is expected to offer advantages for application in various regions, providing fundamental data for a better understanding of carbon dynamics in urban ecosystems in the future.

• Korean Journal of Remote Sensing
• /
• v.35 no.6_3
• /
• pp.1235-1249
• /
• 2019

Kilauea Volcano is one of the most active volcano in the world. In this study, we used the ALOS-2 PALSAR-2 satellite imagery to measure the surface deformation occurring near the summit of the Kilauea volcano from 2015 to 2017. In order to measure two-dimensional surface deformation, interferometric synthetic aperture radar (InSAR) and multiple aperture SAR interferometry (MAI) methods were performed using two interferometric pairs. To improve the precision of 2D measurement, we compared root-mean-squared deviation (RMSD) of the difference of measurement value as we change the effective antenna length and normalized squint value, which are factors that can affect the measurement performance of the MAI method. Through the compare, the values of the factors, which can measure deformation most precisely, were selected. After select optimal values of the factors, the RMSD values of the difference of the MAI measurement were decreased from 4.07 cm to 2.05 cm. In each interferograms, the maximum deformation in line-of-sight direction is -28.6 cm and -27.3 cm, respectively, and the maximum deformation in the along-track direction is 20.2 cm and 20.8 cm, in the opposite direction is -24.9 cm and -24.3 cm, respectively. After stacking the two interferograms, two-dimensional surface deformation mapping was performed, and a maximum surface deformation of approximately 30.4 cm was measured in the northwest direction. In addition, large deformation of more than 20 cm were measured in all directions. The measurement results show that the risk of eruption activity is increasing in Kilauea Volcano. The measurements of the surface deformation of Kilauea volcano from 2015 to 2017 are expected to be helpful for the study of the eruption activity of Kilauea volcano in the future.

• Journal of the Korean earth science society
• /
• v.29 no.1
• /
• pp.1-12
• /
• 2008

A geophysical mapping was performed for Hwasan caldera which is located in Euisung Sub-basin of the southeastern part of the Korean Peninsula. In order to overcome the limitation of the previous studies, remote sensing technic was used and dense potential data were obtained and analyzed. First, we analyzed geological lineament for target area using geological map, digital elevation model (DEM) data and satellite imagery. The results were greatly consistent with the previous studies, and showed that N-S and NW-SE direction are the most dominant one in target area. Second, based on the lineament analysis, highly dense gravity data were acquired in Euisung Sub-basin and an integrated interpretation considering air-born magnetic data was made to investigate the regional structure of the target area. The results of power spectrum analysis for the acquired potential data revealed that the subsurface of Euisung Sub-basin have two density discontinuities at about 1 km and 3-5 km depth. A 1 km depth discontinuity is thought as the depth of pyroclastic sedimentary rocks or igneous rocks which were intruded at the ring vent of Hwasan caldera, while a 3-5 km depth discontinuity seems to be associated with the depth of the basin basement. In addition, three-dimensional gravity inversion for the total area of Euisung Sub-basin was carried out, and the inversion results indicated two followings; 1) Cretaceous Palgongsan granite and Bulguksa intrusion rocks, which are located in southeastern part and northeastern part of Euisung Sub-basin, show two major low density anomalies, 2) pyroclastic rocks around Hwasan caldera also have lower density when compared with those of neighborhood regions and are extended to 1.5 km depth. However, a poor vertical resolution of potential survey makes it difficult to accurately delineate the detailed structure caldera which has a vertically developed characteristic in general. To overcome this limitation, integrated analysis was carried out using the magnetotelluric data on the corresponding area with potential data and we could obtain more reasonable geologic structure.