• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.14 no.5
• /
• pp.430-437
• /
• 2021

In this paper, we propose a robust face recognition method against soft errors with a deep convolutional generative adversarial network(DCGAN) based compensation method by a cross-layer approach. When soft-errors occur in block data of JPEG files, these blocks can be decoded inappropriately. In previous results, these blocks have been replaced using a mean face, thereby improving recognition ratio to a certain degree. This paper uses a DCGAN-based compensation approach to extend the previous results. When soft errors are detected in an embedded system layer using parity bit checkers, they are compensated in the application layer using compensated block data by a DCGAN-based compensation method. Regarding soft errors and block data loss in facial images, a DCGAN architecture is redesigned to compensate for the block data loss. Simulation results show that the proposed method effectively compensates for performance degradation due to soft errors.

• The Bulletin of The Korean Astronomical Society
• /
• v.46 no.2
• /
• pp.46.3-47
• /
• 2021

Recently a multilayer spectral inversion (MLSI) model has been proposed to infer the physical parameters of plasmas in the solar chromosphere. The inversion solves a three-layer radiative transfer model using the strong absorption line profiles, H alpha and Ca II 8542 Å, taken by the Fast Imaging Solar Spectrograph (FISS). The model successfully provides the physical plasma parameters, such as source functions, Doppler velocities, and Doppler widths in the layers of the photosphere to the chromosphere. However, it is quite expensive to apply the MLSI to a huge number of line profiles. For example, the calculating time is an hour to several hours depending on the size of the scan raster. We apply deep neural network (DNN) to the inversion code to reduce the cost of calculating the physical parameters. We train the models using pairs of absorption line profiles from FISS and their 13 physical parameters (source functions, Doppler velocities, Doppler widths in the chromosphere, and the pre-determined parameters for the photosphere) calculated from the spectral inversion code for 49 scan rasters (~2,000,000 dataset) including quiet and active regions. We use fully connected dense layers for training the model. In addition, we utilize a skip connection to avoid a problem of vanishing gradients. We evaluate the model by comparing the pairs of absorption line profiles and their inverted physical parameters from other quiet and active regions. Our result shows that the deep learning model successfully reproduces physical parameter maps of a scan raster observation per second within 15% of mean absolute percentage error and the mean squared error of 0.3 to 0.003 depending on the parameters. Taking this advantage of high performance of the deep learning model, we plan to provide the physical parameter maps from the FISS observations to understand the chromospheric plasma conditions in various solar features.

• Ocean and Polar Research
• /
• v.28 no.4
• /
• pp.475-484
• /
• 2006

In order to reveal the vertical variation of physical properties in deep-sea sediments, deep-sea sediment cores were collected at 78 stations using a multiple corer in the KR5 area, one of the Korea contract areas for manganese nodule exploration, located in the northeast equatorial Pacific. Based on the color of sediments, sampled sediment cores were characterized into three lithologic units (unit 1,2, and 3). In all sediment cores, three units appear systematically; unit 1 lies at the top of cores and unit 2 and/or unit 3 appear to underlie unit 1 or alternate with unit 3. Unit 1 layer from the top of cores shows dark grayish brown to dark brown with mean thickness of 10.2cm. Unit 2 and 3 layers show very dark brown to black color and yellowish brown to brown color, respectively. According to the physical properties of the deep-sea sediment cores, sediment column can be divided into three sections. Section A $(0{\sim}15cm)$ in subbottom depth consists mostly of unit 1. Mean values of physical properties of section B $(15{\sim}30cm)$ in subbottom depth are similar to those of section C (>30 cm) in subbottom depth. However, the physical properties of section B were more variable than those of section C because of the high activity of bioturbation in section B. These results will provide valuable information for selecting suitable sites for mining manganese nodules in the Korea contract areas.

• Journal of Ecology and Environment
• /
• v.43 no.4
• /
• pp.400-420
• /
• 2019

The aims of the study were to analyze (1) the effects of Prosopis juliflora (Prosopis) on the spatial distribution and soil seed banks (SSB) diversity and density, (2) the effects of environmental factors on SSB diversity and density (number of seeds in the soil per unit area), and (3) the effects of animal fecal droppings on SSB diversity, density, and dispersal. Aboveground vegetation data were collected from different Prosopis-infested habitats from quadrats (20 × 20 m) in Prosopis thickets, Prosopis + native species stand, non-invaded woodlands, and open grazing lands. In each Prosopis-infested habitats, soil samples were collected from the litter layer and three successive soil layer, i.e., 0-3 cm, 3-6 cm, and 6-9 cm. Seeds from soil samples and animal fecal matter were separated in the green house using the seedling emergence technique. Invasion of Prosopis had significant effects on the soil seed bank diversity. Results revealed that the mean value of the Shannon diversity of non-invaded woodlands was being higher by 19.2%, 18.5%, and 11.0% than Prosopis thickets; Prosopis + native species stand and open grazing lands, respectively. The seed diversity and richness, recovered from 6-9-cm-deep layer were the highest. On the other hand, the density of Prosopis seeds was the highest in the litter layer. About 156 of seeds/kg (92.9%) of seeds were germinated from cattle fecal matter. However, in a small proportion of seedlings, 12 of seeds/kg (7.1%) were germinated from shot fecal matter. Thus, as the seeds in the soil were low in the study areas, in situ and ex situ conservation of original plants and reseeding of persistent grass species such as Cynodon dactylon, Cenchrus ciliaris, Chrysopogon plumulosus, and Brachiaria ramosa are recommended.

• Journal of the Korean Society for Marine Environment & Energy
• /
• v.18 no.3
• /
• pp.143-156
• /
• 2015

In order to understand the dynamic characteristics of water column environments in the Western Pacific seamount area (approximately $150.2^{\circ}E$, $20^{\circ}N$), we investigated the water mass and the behavior of water column parameters such as dissolved oxygen, inorganic nutrients (N, P), and chlorophyll-a. Physico-chemical properties of water column were obtained by CTD system at the nine stations which were selected along the east-west and south-north direction around the seamount (OSM14-2) in October 2014. From the temperature-salinity diagram, the main water masses were separated into North Pacific Tropical Water and Thermocline Water in the surface layer, North Pacific Intermediate Water in the intermediate layer, and North Pacific Deep Water in the bottom layer, respectively. Oxygen minimum zone (OMZ, mean $O_2$ $73.26{\mu}M$), known as dysoxic condition ($O_2<90{\mu}M$), was distributed in the depth range of 700~1,200 m throughout the study area. Inorganic nutrients typified by nitrite + nitrate and phosphate showed the lowest concentration in the surface mixed layer and then gradually increased downward with representing the maximum concentration in the OMZ, with lower N:P ratio (13.7), indicating that the nitrogen is regarded as limiting factor for primary production. Vertical distribution of water column parameters along the east-west and south-north station line around the seamount showed the effect of bottom water inflowing at around 500 m deep in the western and southern region, and concentrations of water column parameters in the bottom layer (below 2,500 m deep) of the western and southern region were differently distributed comparing to those of the other side regions (eastern and northern). The value of Excess N calculated from Redfield ratio (N:P=16:1) represented the negative value throughout the study area, which indicated the nitrogen sink dominant environments, and relative higher value of Excess N observed in the bottom layer of western and southern region. These observations suggest that the topographic features of a seamount influence the circulation of bottom current and its effects play a significant role in determining the behavior of water column environmental parameters.

• Ocean and Polar Research
• /
• v.28 no.3
• /
• pp.245-258
• /
• 2006

We investigated seasonal variations of the upper ocean temperature and the mixed layer depth (MLD) in an eddy-permitting global ocean general circulation model (OGCM) to assess the OGCM perfermance. The OGCM is based on the GFDL MOM3 which has a horizontal resolution of 0.5 degree and 30 vertical levels. The OGCM was integrated for 68 years using a monthly-mean climatological wind stress forcing. The model sea surface temperature (SST) and sea surface salinity were restored to the Levitus climatology with a time scale of 30 days. Annual-mean model SST shows a cold bias $(<\;-2^{\circ}C)$ in the summer hemisphere and a warm bias $(>\;1^{\circ}C)$ in the winter hemisphere mainly due to the restoring boundary condition of temperature. The model MLD captures well the observed features in most areas, with a slightly deep bias. However, in the Ross Sea and Weddell Sea, the model shows significantly deeper MLD than the climatology-mainly due to weak salinity stratifications in the model. For amplitude of seasonal variation, the model SST is smaller $(1{\sim}3^{\circ}C)$ than the observation largely due to the restoring surface boundary condition while the model MLD has larger seasonal variation $({\sim}50m)$. It is suggested that for more realistic simulation of the upper ocean structure in the present eddy-permitting ocean model, more refinements in the surface boundary condition for the thermohaline forcing and parameterization for vertical mixing are required, together with the incorporation of a sea-ice model.

• KIPS Transactions on Software and Data Engineering
• /
• v.11 no.1
• /
• pp.29-34
• /
• 2022

We propose a novel network architecture and build dataset for recognizing clickable objects on mobile device screens. The data was collected based on clickable objects on the mobile device screen that have numerous resolution, and a total of 24,937 annotation data were subdivided into seven categories: text, edit text, image, button, region, status bar, and navigation bar. We use the Deconvolution Single Shot Detector as a baseline, the backbone network with Squeeze-and-Excitation blocks, the Single Shot Detector layer structure to derive inference results and the Feature pyramid networks structure. Also we efficiently extract features by changing the input resolution of the existing 1:1 ratio of the network to a 1:2 ratio similar to the mobile device screen. As a result of experimenting with the dataset we have built, the mean average precision was improved by up to 101% compared to baseline.

• Nuclear Engineering and Technology
• /
• v.52 no.2
• /
• pp.287-295
• /
• 2020

Group method of data handling (GMDH) is considered one of the earliest deep learning methods. Deep learning gained additional interest in today's applications due to its capability to handle complex and high dimensional problems. In this study, multi-layer GMDH networks are used to perform uncertainty quantification (UQ) and sensitivity analysis (SA) of nuclear reactor simulations. GMDH is utilized as a surrogate/metamodel to replace high fidelity computer models with cheap-to-evaluate surrogate models, which facilitate UQ and SA tasks (e.g. variance decomposition, uncertainty propagation, etc.). GMDH performance is validated through two UQ applications in reactor simulations: (1) low dimensional input space (two-phase flow in a reactor channel), and (2) high dimensional space (8-group homogenized cross-sections). In both applications, GMDH networks show very good performance with small mean absolute and squared errors as well as high accuracy in capturing the target variance. GMDH is utilized afterward to perform UQ tasks such as variance decomposition through Sobol indices, and GMDH-based uncertainty propagation with large number of samples. GMDH performance is also compared to other surrogates including Gaussian processes and polynomial chaos expansions. The comparison shows that GMDH has competitive performance with the other methods for the low dimensional problem, and reliable performance for the high dimensional problem.

• Journal of the Korean Society of Fisheries and Ocean Technology
• /
• v.35 no.4
• /
• pp.404-409
• /
• 1999

The in situ target strengths of, maurolicus muelleri were measured by the split beam echo sounder system at the frequency of 38kHz.Target strengths were measured during the night time in order to obtain the pure separated echoes from the scattered individual. And also it was to establish reasonable threshold due to taking the signals like as the planktons and etc.Since Maurolicus muelleri is a typical micronektonic fish, they mainly consisted of deep scattering layers(DSLs), and had a vertical migration perrodically during daytime and at night.We found that the Maurolicus muelleri occupied about 99% of total catch. Total length ranged from 4.5 to 5.7cm with a mean of 5.2cm and a standard deviation of 0.22cm.The target strengths of Maurolicus muelleir ranged from -60.4 to - 52.7dB and -59.2 to - 52.5 dB in the water layer of 10~30m and 30~50m depth, perspectively. Mean target strength was -57.1dB/fish and -28.5dB/kg. The target strength had the relation with the total length of the fish as, TS=20logL-71.4.

• Journal of the Korean Society of Manufacturing Process Engineers
• /
• v.18 no.11
• /
• pp.47-56
• /
• 2019

A deep neural network (DNN) model was proposed to predict the upset in the inertia friction welding process using a database comprising results from a series of FEM analyses. For the database, the upset length, upset beginning time, and upset completion time were extracted from the results of the FEM analyses obtained with various of axial pressure and initial rotational speed. A total of 35 training sets were constructed to train the proposed DNN with 4 hidden layers and 512 neurons in each layer, which can relate the input parameters to the welding results. The mean of the summation of squared error between the predicted results and the true results can be constrained to within 1.0e-4 after the training. Further, the network model was tested with another 10 sets of welding input parameters and results for comparison with FEM. The test showed that the relative error of DNN was within 2.8% for the prediction of upset. The results of DNN application revealed that the model could effectively provide welding results with respect to the exactness and cost for each combination of the welding input parameters.