• The Journal of the Acoustical Society of Korea
• /
• v.39 no.5
• /
• pp.414-423
• /
• 2020

In this paper, we propose a Dilated Convolution Gate Linear Unit (DCGLU) to mitigate the lack of sparsity and small receptive field problems caused by the segmentation map extraction process in sound event detection with weak labels. In the advent of deep learning framework, segmentation map extraction approaches have shown improved performance in noisy environments. However, these methods are forced to maintain the size of the feature map to extract the segmentation map as the model would be constructed without a pooling operation. As a result, the performance of these methods is deteriorated with a lack of sparsity and a small receptive field. To mitigate these problems, we utilize GLU to control the flow of information and Dilated Convolutional Neural Networks (DCNNs) to increase the receptive field without additional learning parameters. For the performance evaluation, we employ a URBAN-SED and self-organized bird sound dataset. The relevant experiments show that our proposed DCGLU model outperforms over other baselines. In particular, our method is shown to exhibit robustness against nature sound noises with three Signal to Noise Ratio (SNR) levels (20 dB, 10 dB and 0 dB).

• KIPS Transactions on Software and Data Engineering
• /
• v.3 no.9
• /
• pp.361-368
• /
• 2014

The LPR(License plate recognition) system has been developed to efficient control for complex traffic environment and currently be used in many places. However, because of light, noise, background changes, environmental changes, damaged plate, it only works limited environment, so it is difficult to use in real-time. This paper presents a heuristic segmentation algorithm for robust to noise and illumination changes and introduce a real-time license plate recognition system using it. In first step, We detect the plate utilized Haar-like feature and Adaboost. This method is possible to rapid detection used integral image and cascade structure. Second step, we determine the type of license plate with adaptive histogram equalization, bilateral filtering for denoise and segment accurate character based on adaptive threshold, pixel projection and associated with the prior knowledge. The last step is character recognition that used histogram of oriented gradients (HOG) and multi-layer perceptron(MLP) for number recognition and support vector machine(SVM) for number and Korean character classifier respectively. The experimental results show license plate detection rate of 94.29%, license plate false alarm rate of 2.94%. In character segmentation method, character hit rate is 97.23% and character false alarm rate is 1.37%. And in character recognition, the average character recognition rate is 98.38%. Total average running time in our proposed method is 140ms. It is possible to be real-time system with efficiency and robustness.

• Journal of the Korean Society for Railway
• /
• v.11 no.4
• /
• pp.349-356
• /
• 2008

Put Safety-critical systems related to the railway communications are currently undergoing changes. Mechanical and electro-mechanical devices are being replaced by programmable electronics that are often controlled remotely via communication networks. Therefore designers and operators now not only have to contend with component failures and user errors, but also with the possibility that malicious entities are seeking to disrupt the services provided by theirs systems. Recognizing the safety-critical nature of the types of communications required in rail control operations, the communications infrastructure will be required to meet a number of safety requirements such as system faults, user errors and the robustness in the presence of malicious attackers who are willing to take determined action to interfere in the correct operation of a system. This paper discusses the safety strategies employed in the railway communications and proposes a security mechanism for Korean railway communication system. We present the developed communication safety evaluation tool based on the proposed security mechanism and also evaluate its protecting capability against the threats of masquerading, eavesdropping, and unauthorized message manipulation.

• Journal of Internet Computing and Services
• /
• v.7 no.2
• /
• pp.23-35
• /
• 2006

This work presents o novel method which extract facial region und features from motion picture automatically and controls the 3D facial expression in real time. To txtract facial region and facial feature points from each color frame of motion pictures a new nonparametric skin color model is proposed rather than using parametric skin color model. Conventionally used parametric skin color models, which presents facial distribution as gaussian-type, have lack of robustness for varying lighting conditions. Thus it needs additional work to extract exact facial region from face images. To resolve the limitation of current skin color model, we exploit the Hue-Tint chrominance components and represent the skin chrominance distribution as a linear function, which can reduce error for detecting facial region. Moreover, the minimal facial feature positions detected by the proposed skin model are adjusted by using edge information of the detected facial region along with the proportions of the face. To produce the realistic facial expression, we adopt Water's linear muscle model and apply the extended version of Water's muscles to variation of the facial features of the 3D face. The experiments show that the proposed approach efficiently detects facial feature points and naturally controls the facial expression of the 3D face model.

• KIPS Transactions on Computer and Communication Systems
• /
• v.3 no.10
• /
• pp.357-364
• /
• 2014

The first mission for the IoT based hyper-connectivity communication is developing a device-to-device communication technique in infra-less low-power and lossy networks. In a low-power and lossy wireless network, IoT devices and routers cannot keep the original path toward the destination since they have the limited memory. Different from the previous light-weight routing protocols focusing on the reduction of the control messages, the proposed scheme provides the light-weight IPv6 address auto-configuration, IPv6 neighbor discovery and routing protocol in a IoT capable infra-less wireless networks with the bloom filer and enhanced rank concepts. And for the first time we evaluate our proposed scheme based on the modeling of various probability distributions in the IoT environments with the lossy wireless link. Specifically, the proposed enhanced RPL based light-weight routing protocol improves the robustness with the multi-paths locally established based on the enhanced rank concepts even though lossy wireless links are existed. We showed the improvements of the proposed scheme up to 40% than the RPL based protocol.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.47 no.2
• /
• pp.107-113
• /
• 2019

In this paper, a robust filter based wind velocity estimation algorithm without an air velocity sensor in an air vehicle is presented. The wind velocity is useful information for the air vehicle to perform precise guidance and control. In general, the wind velocity can be obtained by subtracting an air velocity which is obtained by an air velocity sensor such as a pitot-tube, and a ground velocity which is obtained by a navigation equipment. However, in order to simplify the configuration of the air vehicle, the wind estimation algorithm is necessary because the wind velocity can not be directly obtained if the air velocity measurement sensor is not used. At this time, the aerodynamic coefficient of the air vehicle changes due to the turbulence, which causes the uncertainty of the system model of the filter, and the wind estimation performance deteriorates. Therefore, in this study, we propose a wind estimation method using $H{\infty}$ filter to ensure robustness against aerodynamic coefficient uncertainty, and we confirmed through simulation that the proposed method improves the performance in the uncertainty of aerodynamic coefficient.

• Journal of Korea Water Resources Association
• /
• v.54 no.9
• /
• pp.681-692
• /
• 2021

Vegetation processes have a significant impact on rainfall runoff processes through evapotranspiration control, but are rarely considered in the conceptual lumped hydrological model. This study evaluated the model performance of the Hapcheon Dam watershed by integrating the ecological module expressing the leaf area index data sensed remotely from the satellite into the hydrological partition module. The proposed eco-hydrological model has three main features to better represent the eco-hydrological process in humid regions. 1) The growth rate of vegetation is constrained by water shortage stress in the watershed. 2) The maximum growth of vegetation is limited by the energy of the watershed climate. 3) The interaction of vegetation and aquifers is reflected. The proposed model simultaneously simulates hydrologic components and vegetation dynamics of watershed scale. The following findings were found from the validation results using the model parameters estimated by the SCEM algorithm. 1) Estimating the parameters of the eco-hydrological model using the leaf area index and streamflow data can predict the streamflow with similar accuracy and robustness to the hydrological model without the ecological module. 2) Using the remotely sensed leaf area index without filtering as input data is not helpful in estimating streamflow. 3) The integrated eco-hydrological model can provide an excellent estimate of the seasonal variability of the leaf area index.

• Journal of Korean Society of Forest Science
• /
• v.110 no.2
• /
• pp.189-197
• /
• 2021

This study aims to optimize the appropriate concentration of fertilizers for Sageretia thea by analyzing growth performances (height and root collar diameter) and physiological characteristics (photosynthesis, chlorophyll contents, and chlorophyll fluorescence reaction). As fertilizer concentration was increased to 1.5 g·L^-1, growth increased, but it decreased at 2.0 g·L^-1 treatment. Root collar diameter growth was reduced because of higher fertilizer concentrations. Photosynthesis reactions showed the highest CO₂ reaction curves, maximum photosynthesis rate, and maximum carboxylation rate in the 1.5 g·L^-1 fertilizer treatment. The chlorophyll fluorescence reaction and SPAD values revealed that fertilizer treatment improves photosynthesis efficiency and robustness compared with untreated control. Therefore, the appropriate fertilizer concentration for producing good seedling quality of Sageretia thea is 1.0~1.5 g·L^-1.

• Korean Journal of Radiology
• /
• v.21 no.7
• /
• pp.869-879
• /
• 2020

Objective: To evaluate the performance of a convolutional neural network (CNN) model that can automatically detect and classify rib fractures, and output structured reports from computed tomography (CT) images. Materials and Methods: This study included 1079 patients (median age, 55 years; men, 718) from three hospitals, between January 2011 and January 2019, who were divided into a monocentric training set (n = 876; median age, 55 years; men, 582), five multicenter/multiparameter validation sets (n = 173; median age, 59 years; men, 118) with different slice thicknesses and image pixels, and a normal control set (n = 30; median age, 53 years; men, 18). Three classifications (fresh, healing, and old fracture) combined with fracture location (corresponding CT layers) were detected automatically and delivered in a structured report. Precision, recall, and F1-score were selected as metrics to measure the optimum CNN model. Detection/diagnosis time, precision, and sensitivity were employed to compare the diagnostic efficiency of the structured report and that of experienced radiologists. Results: A total of 25054 annotations (fresh fracture, 10089; healing fracture, 10922; old fracture, 4043) were labelled for training (18584) and validation (6470). The detection efficiency was higher for fresh fractures and healing fractures than for old fractures (F1-scores, 0.849, 0.856, 0.770, respectively, p = 0.023 for each), and the robustness of the model was good in the five multicenter/multiparameter validation sets (all mean F1-scores > 0.8 except validation set 5 [512 x 512 pixels; F1-score = 0.757]). The precision of the five radiologists improved from 80.3% to 91.1%, and the sensitivity increased from 62.4% to 86.3% with artificial intelligence-assisted diagnosis. On average, the diagnosis time of the radiologists was reduced by 73.9 seconds. Conclusion: Our CNN model for automatic rib fracture detection could assist radiologists in improving diagnostic efficiency, reducing diagnosis time and radiologists' workload.

• KDI Journal of Economic Policy
• /
• v.31 no.2
• /
• pp.127-153
• /
• 2009

There is a growing concern about potential harmful effect of second-hand or environmental tobacco smoking. As a result, smoking bans in workplace become more prevalent worldwide. In Korea, workplace smoking ban policy become more restrictive in 2003 when National health enhancing law was amended. The new law requires all office buildings larger than 3,000 square meters (multi-purpose buildings larger than 2,000 square meters) should be smoke free. Therefore, a lot of indoor office became non smoking area. Previous studies in other counties often found contradicting answers for the effects of workplace smoking ban on smoking behavior. In addition, there was no study in Korea yet that examines the causal impacts of smoking ban on smoking behavior. The situation in Korea might be different from other countries. Using 2001 and 2005 Korea National Health and Nutrition surveys which are representative for population in Korea we try to examine the impacts of law change on current smoker and cigarettes smoked per day. The amended law impacted the whole country at the same time and there was a declining trend in smoking rate even before the legislation update. So, the challenge here is to tease out the true impact only. We compare indoor working occupations which are constrained by the law change with outdoor working occupations which are less impacted. Since the data has been collected before (2001) and after (2005) the law change for treated (indoor working occupations) and control (outdoor working occupations) groups we will use difference in difference method. We restrict our sample to working age (between 20 and 65) since these are the relevant population by the workplace smoking ban policy. We also restrict the sample to indoor occupations (executive or administrative and administrative support) and outdoor occupations (sales and low skilled worker) after dropping unemployed and someone working for military since it is not clear whether these occupations are treated group or control group. This classification was supported when we examined the answers for workplace smoking ban policy existing only in 2005 survey. Sixty eight percent of indoor occupations reported having an office smoking ban policy compared to forty percent of outdoor occupation answering workplace smoking ban policy. The estimated impacts on current smoker are 4.1 percentage point decline and cigarettes per day show statistically significant decline of 2.5 cigarettes per day. Taking into account consumption of average sixteen cigarettes per day among smokers it is sixteen percent decline in smoking rate which is substantial. We tested robustness using the same sample across two surveys and also using tobit model. Our results are robust against both concerns. It is possible that our measure of treated and control group have measurement error which will lead to attenuation bias. However, we are finding statistically significant impacts which might be a lower bound of the true estimates. The magnitude of our finding is not much different from previous finding of significant impacts. For cigarettes per day previous estimates varied from 1.37 to 3.9 and for current smoker it showed between 1%p and 7.8%p.