• Atmosphere
• /
• v.33 no.5
• /
• pp.457-475
• /
• 2023

Accurately predicting localized heavy rainfall is challenging without high-resolution mesoscale cloud information in the numerical model's initial field, as precipitation intensity and amount vary significantly across regions. In the Korean Peninsula, the radar observation network covers the entire country, providing high-resolution data on hydrometeors which is suitable for data assimilation (DA). During the pre-processing stage, radar reflectivity is classified into hydrometeors (e.g., rain, snow, graupel) using the background temperature field. The mixing ratio of each hydrometeor is converted and inputted into a numerical model. Moreover, assimilating saturated water vapor mixing ratio and decomposing radar radial velocity into a three-dimensional wind vector improves the atmospheric dynamic field. This study presents radar DA experiments using a numerical prediction model to enhance the wind, water vapor, and hydrometeor mixing ratio information. The impact of radar DA on precipitation prediction is analyzed separately for each radar component. Assimilating radial velocity improves the dynamic field, while assimilating hydrometeor mixing ratio reduces the spin-up period in cloud microphysical processes, simulating initial precipitation growth. Assimilating water vapor mixing ratio further captures a moist atmospheric environment, maintaining continuous growth of hydrometeors, resulting in concentrated heavy rainfall. Overall, the radar DA experiment showed a 32.78% improvement in precipitation forecast accuracy compared to experiments without DA across four cases. Further research in related fields is necessary to improve predictions of mesoscale heavy rainfall in South Korea, mitigating its impact on human life and property.

• The Journal of Korean Institute of Electromagnetic Engineering and Science
• /
• v.25 no.4
• /
• pp.467-475
• /
• 2014

In this paper, we introduce a sparse recovery algorithm applied to a radar signal model, based on the compressive sensing(CS), for the formulation of the radar signatures, such as high-resolution range profile(HRRP) and ISAR(Inverse Synthetic Aperture Radar) image. When there exits missing data in observed RCS data samples, we cannot obtain correct high-resolution radar signatures with the traditional IDFT(Inverse Discrete Fourier Transform) method. However, high-resolution radar signatures using the sparse recovery algorithm can be successfully recovered in the presence of data missing and qualities of the recovered radar signatures are nearly comparable to those of radar signatures using a complete RCS data without missing data. Therefore, the results show that the sparse recovery algorithm rather than the DFT method can be suitably applied for the reconstruction of high-resolution radar signatures, although we collect incomplete RCS data due to unwanted interferences or jamming signals.

• Proceedings of the Korean Society of Surveying, Geodesy, Photogrammetry, and Cartography Conference
• /
• 2004.11a
• /
• pp.179-184
• /
• 2004

High-resolution satellite imagery are used in various application field such as generation of DEM, orthophto, and three dimensional city model. To define the relation between image and object space, sensor modelling and generation of the epipolar image is essential processes. As the header information or physical sensor model becomes unavailable for the end users due to the national security or commercial purpose, generation of epipolar images without these information becomes one of important processes. In this study, epipolar geometry is generated and analysed by applying two generalized sensor models; parallel and parallel-perspective model Epipolar equation of the parallel model has linear property which is relatively simple; Epipolar geometry of the parallel-perspective model is non-linear. This linear property enable us to generate epipolar image efficiently.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.05a
• /
• pp.479-481
• /
• 2023

Generative Adversarial Networks (GANs) have facilitated substantial improvement in single-image super-resolution (SR) by enabling the generation of photo-realistic images. However, the high memory requirements of GAN-based SRs (mainly generators) lead to reduced performance and increased energy consumption, making it difficult to implement them onto resource-constricted devices. In this study, we propose an efficient and compressed architecture for the SR-GAN (generator) model using the model compression technique Knowledge Distillation. Our approach involves the transmission of knowledge from a heavy network to a lightweight one, which reduces the storage requirement of the model by 58% with also an increase in their performance. Experimental results on various benchmarks indicate that our proposed compressed model enhances performance with an increase in PSNR, SSIM, and image quality respectively for x4 super-resolution tasks.

• Journal of Korean Society for Geospatial Information Science
• /
• v.13 no.1 s.31
• /
• pp.55-61
• /
• 2005

Digital Elevation Model (DEM) generation from remotely sensed imagery is crucial for a variety of mapping applications such as ortho-photo generation, city modeling. High resolution imaging satellites such as SPOT-5, IKONOS, QUICK-BIRD, ORBVIEW constitute an excellent source for efficient and economic generation of DEM data. However, prerequisite knowledge in the areas of sensor modeling, epipolar resampling, and image matching is required to generate DEM from these high resolution satellite imagery. From the above requirements, epipolar resampling emerges as the most important factors. Research attempts in this area are still in high demand and short supply. Another cause that adds to the complication of the problem is that most studies of DEM generation from IKONOS scenes have been based on rational function model. In this paper, we proposed a new methodology for DEM generation from satellite scenes using parallel projection model which is sensor independent, makes it possible for sensor modeling and epipolar resampling by only few control points. The performance and feasibility of the developed methodology is evaluated through real dataset captured by IKONOS.

• Proceedings of the IEEK Conference
• /
• 2005.11a
• /
• pp.481-484
• /
• 2005

Inherent opportunities on research for restoring high resolution image from low resolution images are increasing in these days. Super resolution image reconstruction is the process of combining multiple low resolution images to form a higher resolution one. To achieve super resolution reconstruction, proper observation model which is based on subpixel shift information is required. In this context, the importance of the subpixel registration cannot be estimated because subpixel shift information cannot be obtained from original image. This paper presents a regularized adaptive super resolution reconstruction method based on phase correlated subpixel registration, where the Constrained Least Squares(CLS) Restoration is adopted as a post process.

• Smart Media Journal
• /
• v.8 no.4
• /
• pp.64-71
• /
• 2019

The performance of an OMR (Optical Music Recognition) system is usually determined by the characterizing features of the input music score images. Low resolution is one of the main factors leading to degraded image quality. In this paper, we handle the low-resolution problem using the super-resolution technique. We propose the use of a deep neural network with instance normalization to improve the quality of music score images. We apply instance normalization which has proven to be beneficial in single image enhancement. It works better than batch normalization, which shows the effectiveness of shifting the mean and variance of deep features at the instance level. The proposed method provides an end-to-end mapping technique between the high and low-resolution images respectively. New images are then created, in which the resolution is four times higher than the resolution of the original images. Our model has been evaluated with the dataset "DeepScores" and shows that it outperforms other existing methods.

• Journal of Environmental Science International
• /
• v.23 no.3
• /
• pp.409-431
• /
• 2014

In this study, the characteristics in the simulation of high-resolution coastal weather, i.e. sea surface wind (SSW) and significant wave height (SWH), were studied in a southeastern coastal region of Korea using the WRF and SWAN models. This analyses was performed based on the effects of various input factors in the WRF and SWAN model during M-Case (moderate days with average 1.8 m SWH and $8.4ms^{-1}$ SSW) and R-Case (rough days with average 3.4 m SWH and $13.0ms^{-1}$ SSW) according to the strength of SSW and SWH. The effects of topography (TP), land cover (LC), and sea surface temperature (SST) for the simulation of SSW with the WRF model were somewhat high on v-component winds along the coastline and the adjacent sea of a more detailed grid simulation (333 m) during R-Case. The LC effect was apparent in all grid simulations during both cases regardless of the strength of SSW, whereas the TP effect had shown a difference (decrease or increase) of wind speed according to the strength of SSW (M-Case or R-Case). In addition, the effects of monthly mean currents (CR) and deepwater design waves (DW) for the simulation of SWH with the SWAN model predicted good agreement with observed SWH during R-Case compared to the M-Case. For example, the effects of CR and DW contributed to the increase of SWH during R-Case regardless of grid resolution, whereas the differences (decrease or increase) of SWH occurred according to each effect (CR or DW) during M-Case.

• Journal of the Korean Association of Geographic Information Studies
• /
• v.23 no.3
• /
• pp.252-262
• /
• 2020

In Remote Sensing, a machine learning based SVM model is typically utilized for land cover classification. And study using neural network models is also being carried out continuously. But study using high-resolution imagery of KOMPSAT is insufficient. Therefore, the purpose of this study is to assess the accuracy of land cover classification by neural network models using high-resolution KOMPSAT-3 satellite imagery. After acquiring satellite imagery of coastal areas near Gyeongju City, training data were produced. And land cover was classified with the SVM, ANN and DNN models for the three items of water, vegetation and land. Then, the accuracy of the classification results was quantitatively assessed through error matrix: the result using DNN model showed the best with 92.0% accuracy. It is necessary to supplement the training data through future multi-temporal satellite imagery, and to carry out classifications for various items.

• Korean Journal of Remote Sensing
• /
• v.26 no.6
• /
• pp.653-664
• /
• 2010

The land surface reflectance is a key parameter influencing the climate near the surface. Therefore, it must be determined with sufficient accuracy for climate change research. In particular, the characteristics of the bidirectional reflectance distribution function (BRDF) when using earth observation system (EOS) are important for normalizing the reflected solar radiation from the earth's surface. Also, wide swath satellites like SPOT/VGT (VEGETATION) permit sufficient angular sampling, but high resolution satellites are impossible to obtain sufficient angular sampling over a pixel during short period because of their narrow swath scanning. This gives a difficulty to BRDF model based reflectance normalization of high resolution satellites. The principal objective of the study is to add BRDF modeling of high resolution satellites and to supply insufficient angular sampling through identifying BRDF components from SPOT/VGT. This study is performed as the preliminary data for apply to high-resolution satellite. The study provides surface parameters by eliminating BRD effect when calculated biophysical index of plant by BRDF model. We use semi-empirical BRDF model to identify the BRD components. This study uses SPOT/VGT satellite data acquired in the S1 (daily) data. Modeled reflectance values show a good agreement with measured reflectance values from SPOT satellite. This study analyzes BRD effect components by using the NDVI(Normalized Difference Vegetation Index) and the angle components such as solar zenith angle, satellite zenith angle and relative azimuth angle. Geometric scattering kernel mainly depends on the azimuth angle variation and volumetric scattering kernel is less dependent on the azimuth angle variation. Also, forest from land cover shows the wider distribution of value than cropland, overall tendency is similar. Forest shows relatively larger value of geometric term ($K_1{\cdot}f_1$) than cropland, When performed comparison between cropland and forest. Angle and NDVI value are closely related.