• Korean Journal of Remote Sensing
• /
• v.38 no.6_1
• /
• pp.1505-1514
• /
• 2022

Image matching is a crucial preprocessing step for effective utilization of multi-temporal and multi-sensor very high resolution (VHR) satellite images. Deep learning (DL) method which is attracting widespread interest has proven to be an efficient approach to measure the similarity between image pairs in quick and accurate manner by extracting complex and detailed features from satellite images. However, Image matching of VHR satellite images remains challenging due to limitations of DL models in which the results are depending on the quantity and quality of training dataset, as well as the difficulty of creating training dataset with VHR satellite images. Therefore, this study examines the feasibility of DL-based method in matching pair extraction which is the most time-consuming process during image registration. This paper also aims to analyze factors that affect the accuracy based on the configuration of training dataset, when developing training dataset from existing multi-sensor VHR image database with bias for DL-based image matching. For this purpose, the generated training dataset were composed of correct matching pairs and incorrect matching pairs by assigning true and false labels to image pairs extracted using a grid-based Scale Invariant Feature Transform (SIFT) algorithm for a total of 12 multi-temporal and multi-sensor VHR images. The Siamese convolutional neural network (SCNN), proposed for matching pair extraction on constructed training dataset, proceeds with model learning and measures similarities by passing two images in parallel to the two identical convolutional neural network structures. The results from this study confirm that data acquired from VHR satellite image database can be used as DL training dataset and indicate the potential to improve efficiency of the matching process by appropriate configuration of multi-sensor images. DL-based image matching techniques using multi-sensor VHR satellite images are expected to replace existing manual-based feature extraction methods based on its stable performance, thus further develop into an integrated DL-based image registration framework.

• Journal of Korea Water Resources Association
• /
• v.55 no.6
• /
• pp.393-404
• /
• 2022

Reliable terrestrial rainfall observations from satellites at finer spatial resolution are essential for urban hydrological and microscale agricultural demands. Although various traditional "top-down" approach-based satellite rainfall products were widely used, they are limited in spatial resolution. This study aims to assess the potential of a novel "bottom-up" approach for rainfall estimation, the parameterized SM2RAIN model, applied to the C-band SAR Sentinel-1 satellite data (SM2RAIN-S1), to generate high-spatial-resolution terrestrial rainfall estimates (0.01° grid/6-day) over Central South Korea. Its performance was evaluated for both spatial and temporal variability using the respective rainfall data from a conventional reanalysis product and rain gauge network for a 1-year period over two different sub-regions in Central South Korea-the mixed forest-dominated, middle sub-region and cropland-dominated, west coast sub-region. Evaluation results indicated that the SM2RAIN-S1 product can capture general rainfall patterns in Central South Korea, and hold potential for high-spatial-resolution rainfall measurement over the local scale with different land covers, while less biased rainfall estimates against rain gauge observations were provided. Moreover, the SM2RAIN-S1 rainfall product was better in mixed forests considering the Pearson's correlation coefficient (R = 0.69), implying the suitability of 6-day SM2RAIN-S1 data in capturing the temporal dynamics of soil moisture and rainfall in mixed forests. However, in terms of RMSE and Bias, better performance was obtained with the SM2RAIN-S1 rainfall product over croplands rather than mixed forests, indicating that larger errors induced by high evapotranspiration losses (especially in mixed forests) need to be included in further improvement of the SM2RAIN.

• Korean Journal of Remote Sensing
• /
• v.34 no.5
• /
• pp.811-827
• /
• 2018

The purpose of this study is to compare machine learning algorithm and deep learning algorithm in crop classification using multi-temporal remote sensing data. For this, impacts of machine learning and deep learning algorithms on (a) hyper-parameter and (2) training sample size were compared and analyzed for Haenam-gun, Korea and Illinois State, USA. In the comparison experiment, support vector machine (SVM) was applied as machine learning algorithm and convolutional neural network (CNN) was applied as deep learning algorithm. In particular, 2D-CNN considering 2-dimensional spatial information and 3D-CNN with extended time dimension from 2D-CNN were applied as CNN. As a result of the experiment, it was found that the hyper-parameter values of CNN, considering various hyper-parameter, defined in the two study areas were similar compared with SVM. Based on this result, although it takes much time to optimize the model in CNN, it is considered that it is possible to apply transfer learning that can extend optimized CNN model to other regions. Then, in the experiment results with various training sample size, the impact of that on CNN was larger than SVM. In particular, this impact was exaggerated in Illinois State with heterogeneous spatial patterns. In addition, the lowest classification performance of 3D-CNN was presented in Illinois State, which is considered to be due to over-fitting as complexity of the model. That is, the classification performance was relatively degraded due to heterogeneous patterns and noise effect of input data, although the training accuracy of 3D-CNN model was high. This result simply that a proper classification algorithms should be selected considering spatial characteristics of study areas. Also, a large amount of training samples is necessary to guarantee higher classification performance in CNN, particularly in 3D-CNN.

• Korean Journal of Agricultural and Forest Meteorology
• /
• v.8 no.4
• /
• pp.229-241
• /
• 2006

Global air temperature has risen by $0.6^{\circ}C$ over the last one hundred years due to increased atmospheric greenhouse gases. Moreover, this global warming trend is projected to continue in the future. This study was carried out to evaluate spatial variations in rice production areas by simulating rice-growth and development with projected high resolution climate data in Korea far 2011-2100, which was geospatially interpolated from the 25 km gridded data based on the IPCC SRES A2 emission scenario. Satellite remote sensing data were used to pinpoint the rice-growing areas, and corresponding climate data were aggregated to represent the official 'crop reporting county'. For the simulation experiment, we used a CERES-Rice model modified by introducing two equations to calculate the leaf appearance rate based on the effective temperature and existing leaf number and the final number of leaves based on day-length in the photoperiod sensitive phase of rice. We tested the performance of this model using data-sets obtained from transplanting dates and nitrogen fertilization rates experiments over three years (2002 to 2004). The simulation results showed a good performance of this model in heading date prediction [$R^2$=0.9586 for early (Odaebyeo), $R^2$=0.9681 for medium (Hwasungbyeo), and $R^2$=0.9477 for late (Dongjinbyeo) maturity cultivars]. A modified version of CERES-Rice was used to simulate the growth and development of three Japonica varieties, representing early, medium, and late maturity classes, to project crop status for climatological normal years between 2011 and 2100. In order to compare the temporal changes, three sets of data representing 3 climatological years (2011-2040, 2041-2070, and 2071-2100) were successively used to run the model. Simulated growth and yield data of the three Japonica cultivars under the observed climate for 1971-2000 was set as a reference. Compared with the current normal, heading date was accelerated by 7 days for 2011-2040 and 20 days for 2071-2100. Physiological maturity was accelerated by 15 days for 2011-2040 and 30 days for 2071-2100. Rice yield was in general reduced by 6-25%, 3-26%, and 3-25% per 10a in early, medium, and late maturity classes, respectively. However, mid to late maturing varieties showed an increased yield in northern Gyeonggi Province and in most of Kwangwon Province in 2071-2100.

• Journal of the Korean Society for Marine Environment & Energy
• /
• v.10 no.2
• /
• pp.67-85
• /
• 2007

A three-dimensional hydrodynamic model with the fine grid is applied to simulate the barotropic tides, tidal currents, residual currents and salinity dispersions in the Yellow Sea and the East China Sea. Data inputs include seasonal hydrography, mean wind and river input, and oceanic tides. Computed tidal distributions of four major tides($M_2,\;S_2,\;K_1$ and $O_1$) are presented and results are in good agreement with the observations in the domain. The model reproduces well the tidal charts. The tidal residual current is relatively strong around west coast of Korea including the Cheju Island and southern coast of China. The current by $M_2$ has a maximum speed of 10 cm/s in the vicinity of Cheju Island with a anti-clockwise circulation in the Yellow Sea. General tendency of the current, however, is to flow eastward in the South Sea. Surface residual current simulated with $M_2$ and with $M_2+S_2+K_1+O_1$ tidal forcing shows slightly different patterns in the East China Sea. The model shows that the southerly wind reduces the southward current created by freshwater discharge. In summer during high runoff(mean discharge about $50,000\;m^3/s$ of Yangtze), low salinity plume-like structure(with S < 30.0 psu) extending some 160 km toward the northeast and Changjiang Diluted Water(CDW), below salinity 26 psu, was found within about 95 km. The offshore dispersion of the Changjiang outflow water is enhanced by the prevailing southerly wind. It is estimated that the inertia of the river discharge cannot exclusively reach the around sea of Cheju Island. It is noted that spatial and temporal distribution of salinity and the other materials are controlled by mixture of Changjiang discharge, prevailing wind, advection by flowing warm current and tidal current.

• Journal of the Korean Association of Geographic Information Studies
• /
• v.5 no.2
• /
• pp.1-15
• /
• 2002

Atmospheric aerosols interact with sunlight and affect the global radiation balance that can cause climate change through direct and indirect radiative forcing. Because of the spatial and temporal uncertainty of aerosols in atmosphere, aerosol characteristics are not considered through GCMs (General Circulation Model). Therefor it is important physical and optical characteristics should be evaluated to assess climate change and radiative effect by atmospheric aerosols. In this study GMS-5 satellite data and surface measurement data were analyzed using a radiative transfer model for the Yellow Sand event of April 7~8, 2000 in order to investigate the atmospheric radiative effects of Yellow Sand aerosols, MODTRAN3 simulation results enable to inform the relation between satellite channel albedo and aerosol optical thickness(AOT). From this relation AOT was retreived from GMS-5 visible channel. The variance observations of satellite images enable remote sensing of the Yellow Sand particles. Back trajectory analysis was performed to track the air mass from the Gobi desert passing through Korean peninsular with high AOT value measured by ground based measurement. The comparison GMS-5 AOT to ground measured RSR aerosol optical depth(AOD) show that for Yellow Sand aerosols, the albedo measured over ocean surfaces can be used to obtain the aerosol optical thickness using appropriate aerosol model within an error of about 10%. In addition, LIDAR network measurements and backward trajectory model showed characteristics and appearance of Yellow Sand during Yellow Sand events. These data will be good supporting for monitoring of Yellow Sand aerosols.

• Spatial Information Research
• /
• v.22 no.1
• /
• pp.55-63
• /
• 2014

Near surface air temperature data which are one of the essential factors in hydrology, meteorology and climatology, have drawn a substantial amount of attention from various academic domains and societies. Meteorological observations, however, have high spatio-temporal constraints with the limits in the number and distribution over the earth surface. To overcome such limits, many studies have sought to estimate the near surface air temperature from satellite image data at a regional or continental scale with simple regression methods. Alternatively, we applied various Kriging methods such as ordinary Kriging, universal Kriging, Cokriging, Regression Kriging in search of an optimal estimation method based on near surface air temperature data observed from automatic weather stations (AWS) in South Korea throughout 2010 (365 days) and MODIS land surface temperature (LST) data (MOD11A1, 365 images). Due to high spatial heterogeneity, auxiliary data have been also analyzed such as land cover, DEM (digital elevation model) to consider factors that can affect near surface air temperature. Prior to the main estimation, we calculated root mean square error (RMSE) of temperature differences from the 365-days LST and AWS data by season and landcover. The results show that the coefficient of variation (CV) of RMSE by season is 0.86, but the equivalent value of CV by landcover is 0.00746. Seasonal differences between LST and AWS data were greater than that those by landcover. Seasonal RMSE was the lowest in winter (3.72). The results from a linear regression analysis for examining the relationship among AWS, LST, and auxiliary data show that the coefficient of determination was the highest in winter (0.818) but the lowest in summer (0.078), thereby indicating a significant level of seasonal variation. Based on these results, we utilized a variety of Kriging techniques to estimate the surface temperature. The results of cross-validation in each Kriging model show that the measure of model accuracy was 1.71, 1.71, 1.848, and 1.630 for universal Kriging, ordinary Kriging, cokriging, and regression Kriging, respectively. The estimates from regression Kriging thus proved to be the most accurate among the Kriging methods compared.

• Journal of Korean Society of Coastal and Ocean Engineers
• /
• v.29 no.4
• /
• pp.206-216
• /
• 2017

The temporal distribution of the turbulence kinetic energy (TKE) and the vertical component of Reynolds stresses ($-{\bar{u^{\prime}w^{\prime}}}$) was measured during one wave period under high wave energy conditions. The wave data were obtained at Hujeong Beach in the east coast of Korea at January 14~18 of 2017 when an extratropical cyclone was developed in the East Sea. Among the whole thousands of waves measured during the period, hundreds of regular waves that had with similar pattern were selected for the analysis in order to give three representing mean wave patterns using the ensemble average technique. The turbulence properties were then estimated based on the selected wave data. It is interesting to find out that $-{\bar{u^{\prime}w^{\prime}}}$ has one clear peak near the time of flow reversal while TKE has two peaks at the corresponding times of maximum cross-shore velocity magnitudes. The distinguished pattern of Reynolds stress indicates that vertical fluxes of such properties as suspended sediments may be enhanced at the time when the horizontal flow direction is reversed to disturb the flows, supporting the turbulence convection process proposed by Nielsen (1992). The characteristic patterns of turbulence properties are examined using the CADMAS-SURF Reynolds-Averaged Navier-Stokes (RANS) model. Although the model can reasonably simulate the distribution of TKE pattern, it fails to produce the $-{\bar{u^{\prime}w^{\prime}}}$ peak at the time of flow reversal, which indicates that the application of RANS model is limited in the prediction of some turbulence properties such as Reynolds stresses.

• Korean Journal of Remote Sensing
• /
• v.35 no.6_1
• /
• pp.919-932
• /
• 2019

Fine particulate matter (PM2.5) is not only affected by anthropogenic emissions, but also intensifies, migrates, decreases by hydrometeorological factors. Therefore, it is essential to understand relationships between the hydrometeorological factors and PM2.5 concentration. In Korea, PM2.5 concentration is measured at the ground observatories and estimated data are given to locations where observatories are not present. In this way, the data is not suitable to represent an area, hence it is impossible to know accurate concentration at such locations. In addition, it is hard to trace migration, intensification, reduction of PM2.5. In this study, we analyzed the relationships between hydrometeorological factors, acquired from Global Land Data Assimilation System (GLDAS), and PM2.5 by means of Bayesian Model Averaging (BMA). By BMA, we also selected factors that have meaningful relationship with the variation of PM2.5 concentration. 4 PM2.5 concentration models for different seasons were developed using those selected factors, with Aerosol Optical Depth (AOD) from MODerate resolution Imaging Spectroradiometer (MODIS). Finally, we mapped the result of the model, to show spatial distribution of PM2.5. The model correlated well with the observed PM2.5 concentration (R ~0.7; IOA ~0.78; RMSE ~7.66 ㎍/㎥). When the models were compared with the observed PM2.5 concentrations at different locations, the correlation coefficients differed (R: 0.32-0.82), although there were similarities in data distribution. The developed concentration map using the models showed its capability in representing temporal, spatial variation of PM2.5 concentration. The result of this study is expected to be able to facilitate researches that aim to analyze sources and movements of PM2.5, if the study area is extended to East Asia.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.42 no.6
• /
• pp.1-8
• /
• 2005

In this paper, we propose a two-stage head tracking algorithm adequate for real-time active camera system having pan-tilt-zoom functions. In the color convergence stage, we first assume that the shape of a head is an ellipse and its model color histogram is acquired in advance. Then, the min-shift method is applied to roughly estimate a target position by examining the histogram similarity of the model and a candidate ellipse. To reflect the temporal change of object color and enhance the reliability of mean-shift based tracking, the target histogram obtained in the previous frame is considered to update the model histogram. In the updating process, to alleviate error-accumulation due to outliers in the target ellipse of the previous frame, the target histogram in the previous frame is obtained within an ellipse adaptively shrunken on the basis of the model histogram. In addition, to enhance tracking reliability further, we set the initial position closer to the true position by compensating the global motion, which is rapidly estimated on the basis of two 1-D projection datasets. In the subsequent stage, we refine the position and size of the ellipse obtained in the first stage by using shape information. Here, we define a robust shape-similarity function based on the gradient direction. Extensive experimental results proved that the proposed algorithm performs head hacking well, even when a person moves fast, the head size changes drastically, or the background has many clusters and distracting colors. Also, the propose algorithm can perform tracking with the processing speed of about 30 fps on a standard PC.