• The Journal of the Acoustical Society of Korea
• /
• v.40 no.5
• /
• pp.479-487
• /
• 2021

In this paper, we propose an Sound Event Detection (SED) model using self-training based on a noisy student model. The proposed SED model consists of two stages. In the first stage, a mean-teacher model based on an Residual Convolutional Recurrent Neural Network (RCRNN) is constructed to provide target labels regarding weakly labeled or unlabeled data. In the second stage, a self-training-based noisy student model is constructed by applying different noise types. That is, feature noises, such as time-frequency shift, mixup, SpecAugment, and dropout-based model noise are used here. In addition, a semi-supervised loss function is applied to train the noisy student model, which acts as label noise injection. The performance of the proposed SED model is evaluated on the validation set of the Detection and Classification of Acoustic Scenes and Events (DCASE) 2020 Challenge Task 4. The experiments show that the single model and ensemble model of the proposed SED based on the noisy student model improve F1-score by 4.6 % and 3.4 % compared to the top-ranked model in DCASE 2020 challenge Task 4, respectively.

• Journal of Korea Water Resources Association
• /
• v.46 no.12
• /
• pp.1235-1247
• /
• 2013

This study is to evaluate the climate change impact on future storage behavior of Chungju dam($2,750{\times}10^6m^3$) and the regulation dam($30{\times}10^6m^3$) using SWAT(Soil Water Assessment Tool) model. Using 9 years data (2002~2010), the SWAT was calibrated and validated for streamflow at three locations with 0.73 average Nash-Sutcliffe model Efficiency (NSE) and for two reservoir water levels with 0.86 NSE respectively. For future evaluation, the HadCM3 of GCMs (General Circulation Models) data by scenarios of SRES (Special Report on Emission Scenarios) A2 and B1 of the IPCC (Intergovernmental Panel on Climate Change) were adopted. The monthly temperature and precipitation data (2007~2099) were spatially corrected using 30 years (1977~2006, baseline period) of ground measured data through bias-correction, and temporally downscaled by Change Factor (CF) statistical method. For two periods; 2040s (2031~2050), 2080s (2071~2099), the future annual temperature were predicted to change $+0.9^{\circ}C$ in 2040s and $+4.0^{\circ}C$ in 2080s, and annual precipitation increased 9.6% in 2040s and 20.7% in 2080s respectively. The future watershed evapotranspiration increased up to 15.3% and the soil moisture decreased maximum 2.8% compared to baseline (2002~2010) condition. Under the future dam release condition of 9 years average (2002~2010) for each dam, the yearly dam inflow increased maximum 21.1% for most period except autumn. By the decrease of dam inflow in future autumn, the future dam storage could not recover to the full water level at the end of the year by the present dam release pattern. For the future flood and drought years, the temporal variation of dam storage became more unstable as it needs careful downward and upward management of dam storage respectively. Thus it is necessary to adjust the dam release pattern for climate change adaptation.

• Journal of Korea Water Resources Association
• /
• v.46 no.8
• /
• pp.833-842
• /
• 2013

The quantile mapping is utilized to reproduce reliable GCM(Global Climate Model) data by correct systematic biases included in the original data set. This scheme, in general, projects the Cumulative Distribution Function (CDF) of the underlying data set into the target CDF assuming that parameters of target distribution function is stationary. Therefore, the application of stationary quantile mapping for nonstationary long-term time series data of future precipitation scenario computed by GCM can show biased projection. In this research the Nonstationary Quantile Mapping (NSQM) scheme was suggested for bias correction of nonstationary long-term time series data. The proposed scheme uses the statistical parameters with nonstationary long-term trends. The Gamma distribution was assumed for the object and target probability distribution. As the climate change scenario, the 20C3M(baseline scenario) and SRES A2 scenario (projection scenario) of CGCM3.1/T63 model from CCCma (Canadian Centre for Climate modeling and analysis) were utilized. The precipitation data were collected from 10 rain gauge stations in the Han-river basin. In order to consider seasonal characteristics, the study was performed separately for the flood (June~October) and nonflood (November~May) seasons. The periods for baseline and projection scenario were set as 1973~2000 and 2011~2100, respectively. This study evaluated the performance of NSQM by experimenting various ways of setting parameters of target distribution. The projection scenarios were shown for 3 different periods of FF scenario (Foreseeable Future Scenario, 2011~2040 yr), MF scenario (Mid-term Future Scenario, 2041~2070 yr), LF scenario (Long-term Future Scenario, 2071~2100 yr). The trend test for the annual precipitation projection using NSQM shows 330.1 mm (25.2%), 564.5 mm (43.1%), and 634.3 mm (48.5%) increase for FF, MF, and LF scenarios, respectively. The application of stationary scheme shows overestimated projection for FF scenario and underestimated projection for LF scenario. This problem could be improved by applying nonstationary quantile mapping.

• Journal of the Korea Society of Computer and Information
• /
• v.29 no.10
• /
• pp.35-45
• /
• 2024

In this paper, we propose Multi-Scale Feature Knowledge Distillation for 3D Object Detection (M3KD), which extracting knowledge from the teacher model, and transfer to the student model consider with multi-scale feature map. To achieve this, we minimize L2 loss between feature maps at each pyramid level of the student model with the correspond teacher model so student model can mimic the teacher model backbone information which improves the overall accuracy of the student model. We apply the class logits knowledge distillation used in the image classification task, by allowing student model mimic the classification logits of the teacher model, to guide the student model to improve the detection accuracy. In KITTI (Karlsruhe Institute of Technology and Toyota Technological Institute) dataset, our M3KD (Multi-Scale Feature Knowledge Distillation for 3D Object Detection) student model achieves 30% inference speed improvement compared to the teacher model. Additionally, our method achieved an average improvement of 1.08% in 3D mean Average Precision (mAP) across all classes and difficulty levels compared to the baseline student model. Furthermore, when integrated with the latest knowledge distillation methods such as PKD and SemCKD, our approach achieved an additional 0.42% and 0.52% improvement in 3D mAP, respectively, further enhancing performance.

• Journal of Korea Water Resources Association
• /
• v.42 no.6
• /
• pp.445-455
• /
• 2009

In this paper proportional hazards models for the first through seventh break of 150 mm cast iron pipes in a case study area are established. During the modeling process the assumption of the proportional hazards for covariates on the hazards is examined to include the time-dependent covariate terms in the models. As a result, the pipe material/joint type and the number of customers are modeled as time-dependent for the first failure, and for the second failure only the number of customers is modeled as time-dependent. From the analysis on the baseline hazard functions the failure hazards are found to be generally increasing for the first and second failure, while the hazards of the third break and beyond showed a form of a bath-tub. Furthermore, the changes in the baseline hazard rates according to the time and number of break reflect that the general condition of the pipes is deteriorating. The factors causing pipe break and their effects are analyzed based on the estimated regression coefficients and their hazard ratios, and the constructed models are verified using the deviance residuals of the models.

• Korean Journal of Agricultural and Forest Meteorology
• /
• v.15 no.1
• /
• pp.50-58
• /
• 2013

This study was carried out to evaluate the geospatial characteristics of blooming date migration in three major spring flowers across North and South Korea as influenced by climate change. A thermal time-based phenology model driven by daily maximum and minimum temperature was adjusted for the key parameters (i.e., reference temperature, chilling requirement, heating requirement) used for predicting blooming of forsythia, azaleas, and Japanese cherry. The model was run by the RCP 8.5 projected temperature outlook over the Korean Peninsula and produced the mean booming dates for the three climatological normal years in the future (2011-2040, 2041-2070, and 2071-2100) at a 12.5 km grid spacing. Comparison against the observed blooming date patterns in the baseline climate (1971-2000) showed that there will be a substantial acceleration in blooming dates of the three species, resulting in cherry booming in February and flowers of azaleas and forsythia found at the top of mountain Baikdu by the 2071-2100 period. Flowering dates of the three species in the near future (2011-2040) may be accelerated by 3-5 days at minimum and 10-11 days at maximum compared with that in the baseline period (1971-2000). Those values corresponding to the middle future (2041-2070) can be from a minimum of 9-11 days to a maximum of 23-24 days. Blooming date of Japanese cherry can be accelerated by 26 days on average for the far future (2071-2100). The acceleration seems more prominent at islands and coastal plain areas than over inland mountainous areas.

• Journal of Radiation Protection and Research
• /
• v.24 no.3
• /
• pp.131-142
• /
• 1999

By using the Korean demographic data and the modified relative risk projection model given in the Committee on the Biological Effect of Ionizing Radiation (BEIR) report-V under the U.S. National Academy of Science, the radiogenic excess risk in Korean population has been evaluated. On the basis of this risk, a safety goal for the safe operation of domestic nuclear power plants has been further derived in terms of personal dose. The baseline risk of death due to all causes in Korea and the trivial risk level, which the society considers safe, were estimated to be $5.2{\times}10^{-3}$ and $5.2{\times}10^{-6}$, respectively. The radiogenic excess cancer risk in Korea has been estimated to be $5.2{\times}10^{-3}$ for tie case of acute exposure to 0.1 Gy and $3.7{\times}10^{-3}$ for the case of chronic lifetime exposure to 1.0 mGy/y. On the basis of these risks estimate, the resulting safety goal for one year opeation of a reactor was 0.05 mSv, which is quite identical with the ALARA guideline prescribed by the USNRC in the Appendix I, 10CFR50.

• Journal of the Korean Academy of Child and Adolescent Psychiatry
• /
• v.26 no.4
• /
• pp.273-278
• /
• 2015

Objectives : We retrospectively investigated the efficacy and tolerability of risperidone monotherapy in subjects with autism spectrum disorder (ASD). In addition, we did mixed effect model analysis of the effects of risperidone in patients with ASDs naturalistically treated in a routine clinical setting to determine whether the clinical effects were maintained and the side effects were tolerable. Methods : This retrospective study assessed children and adolescents with ASD, who were on risperidone monotherapy from July 2010 to July 2011 at the Child and Adolescent ASD Clinic at Seoul National Hospital. Outcome measures included the Clinical Global Impression-Severity of Illness (CGI-S) and the CGI-Improvement (CGI-I) scales along with other clinical indices: dosage, target symptoms, and side effects. Results : The mean dose of risperidone in 47 children and adolescents with ASD (40 males, 7 females; age range 5-19 years) who were on risperidone monotherapy was $1.6{\pm}0.8mg/day$, and the mean duration of the treatment period was $20.2{\pm}17.3months$. Aggressive behavior, stereotypic behavior, irritability, and self-injurious behavior were the most frequent target symptoms of risperidone. The most common side effects were weight gain followed by somnolence and extrapyramidal symptoms. In a mixed effects model analysis of CGI-I scores, the mean CGI-I score at the 1 month follow-up was significantly different from the mean CGI-I score of the 3-month follow-up (p=.046), and the CGI-I scores were equally maintained over 3 to 48 months [F(6, 28.9)=4.393, p=.003]. Of the 47 patients, 33 patients (70.2%) were identified as the response group, showing an end point CGI-I rating of 3 or under and having continued risperidone treatment for at least 6 months. The baseline CGI-S score showed significant association with clinical response to risperidone (p=.005), the mean baseline CGI-S was higher in the response group compared to the non-response group. Conclusion : In this study, clinical improvement of risperidone stabilized around 3 months and was equally maintained up to 48 months with tolerable side effects, supporting maintenance of risperidone treatment in children and adolescents with ASDs.

• Journal of Korea Water Resources Association
• /
• v.44 no.3
• /
• pp.169-177
• /
• 2011

The impacts of climate change on paddy irrigation water demands in Korea have been analyzed. High-resolution ($27{\times}27\;km$) climate data for the SRES A2 scenario produced by the Korean Meteorological Research Institute (METRI) and the observed baseline climatology dataset were used. The outputs from the ECHO-G GCM model were dynamically downscaled using the MM5 regional model by the METRI. The Geographic information system (GIS) was used to produce maps showing the spatial changes in irrigation water requirements for rice paddies. The results showed that the growing season mean temperature for future scenarios was projected to increase by $1.5^{\circ}C$ (2020s), $3.3^{\circ}C$ (2050s) and $5.3^{\circ}C$ (2080s) as compared with the baseline value (1971~2000). The growing season rainfall for future scenarios was projected to increase by 0.1% (2020s), 4.9% (2050s) and 19.3% (2080s). Assuming cropping area and farming practices remain unchanged, the total volumetric irrigation demand was projected to increase by 2.8% (2020s), 4.9% (2050s) and 4.5% (2080s). These projections are contrary to the previous study that used HadCM3 outputs and projected decreasing irrigation demand. The main reason for this discrepancy is the difference with the projected climate of the GCMs used. The temporal and spatial variations were large and should be considered in the irrigation water resource planning and management in the future.

• Journal of The Korean Society of Agricultural Engineers
• /
• v.59 no.1
• /
• pp.45-55
• /
• 2017

Climate change causes changes in rainfall patterns, temperature and drought frequency. Climate change impact influences on water management and crop production. It is critical issue in agricultural industry. Rice is a staple cereal crop in South Korea and Korea uses a ponding system for its paddy fields which requires a significant amount of water. In addition, water supply has inter-relationship with crop production which indicates water productivity. Therefore, it is important to assess overall impacts of climate change on water resource and crop production. A water footprint concept is an indicator which shows relationship between water use and crop yield. In addition, it generally composed of three components depending on water resources: green, blue, grey water. This study analyzed the change trend of water footprint of paddy rice under the climate change. The downscaled climate data from HadGEM3-RA based on RCP 8.5 scenario was applied as future periods (2020s, 2050s, 2080s), and historical climate data was set to base line (1990s). Depending on agro-climatic zones, Suwon and Jeonju were selected for study area. A yield of paddy rice was simulated by using FAO-AquaCrop 5.0, which is a water-driven crop model. Model was calibrated by adjusting parameters and was validated by Mann-Whitney U test statistically. The means of water footprint were projected increase by 55 % (2020s), 51 % (2050s) and 48 % (2080s), respectively, from the baseline value of $767m^2/ton$ in Suwon. In case of Jeonju, total water footprint was projected to increase by 46 % (2020s), 45 % (2050s), 12 % (2080s), respectively, from the baseline value of $765m^2/ton$. The results are expected to be useful for paddy water management and operation of water supply system and apply in establishing long-term policies for agricultural water resources.