• Proceedings of the Korea Information Processing Society Conference
• /
• 2019.05a
• /
• pp.498-500
• /
• 2019

Point cloud is a flexible set of points that can provide a scalable geometric representation which can be applied in different computer graphic task. We propose a method based on EdgeConv and densely connected layers to aggregate the features for better classification. Our proposed approach shows significant performance improvement compared to the state-of-the-art deep neural network-based approaches.

• Korean Journal of Radiology
• /
• v.20 no.8
• /
• pp.1275-1284
• /
• 2019

Objective: To develop algorithms using convolutional neural networks (CNNs) for automatic segmentation of acute ischemic lesions on diffusion-weighted imaging (DWI) and compare them with conventional algorithms, including a thresholding-based segmentation. Materials and Methods: Between September 2005 and August 2015, 429 patients presenting with acute cerebral ischemia (training:validation:test set = 246:89:94) were retrospectively enrolled in this study, which was performed under Institutional Review Board approval. Ground truth segmentations for acute ischemic lesions on DWI were manually drawn under the consensus of two expert radiologists. CNN algorithms were developed using two-dimensional U-Net with squeeze-and-excitation blocks (U-Net) and a DenseNet with squeeze-and-excitation blocks (DenseNet) with squeeze-and-excitation operations for automatic segmentation of acute ischemic lesions on DWI. The CNN algorithms were compared with conventional algorithms based on DWI and the apparent diffusion coefficient (ADC) signal intensity. The performances of the algorithms were assessed using the Dice index with 5-fold cross-validation. The Dice indices were analyzed according to infarct volumes (< 10 mL, ≥ 10 mL), number of infarcts (≤ 5, 6-10, ≥ 11), and b-value of 1000 (b1000) signal intensities (< 50, 50-100, > 100), time intervals to DWI, and DWI protocols. Results: The CNN algorithms were significantly superior to conventional algorithms (p < 0.001). Dice indices for the CNN algorithms were 0.85 for U-Net and DenseNet and 0.86 for an ensemble of U-Net and DenseNet, while the indices were 0.58 for ADC-b1000 and b1000-ADC and 0.52 for the commercial ADC algorithm. The Dice indices for small and large lesions, respectively, were 0.81 and 0.88 with U-Net, 0.80 and 0.88 with DenseNet, and 0.82 and 0.89 with the ensemble of U-Net and DenseNet. The CNN algorithms showed significant differences in Dice indices according to infarct volumes (p < 0.001). Conclusion: The CNN algorithm for automatic segmentation of acute ischemic lesions on DWI achieved Dice indices greater than or equal to 0.85 and showed superior performance to conventional algorithms.

• Journal of the Korean Society of Fisheries and Ocean Technology
• /
• v.31 no.1
• /
• pp.63-73
• /
• 1995

A remote telesounder system was manufactured to investigate the behavior and distribution characteristics of fish and was tested to verify its performance and effects int he set net fishing ground located off the Namhae island, Oct. 26~27, 1994. The activity pattern of fish was observed at the buoy station installed in a set net and it was also analyzed simultaneously at the remote base station on land located at distance of about 1.5km away from the buoy station. Several findings at both the buoy station and the base station were as follows: The images of color display monitor at the base station on land were very well coincident with the records observed at the buoy station. The lower part of slope net and bagnet fluctuated irregularly, and its variations were about 3.6m and 6.0m, respectively. In the daytime, the fish school entered playground, passed slope net with 5~7m depth, moved dispersedly at night and at twilight the entrance was rare. In the bagnet, in the daytime the fish school swam in the layer of 8m in depth, at night the fishes dispersed in the entire layer and at twilight they formed dense school. The fishes caught by the set net were about 500kg with 20 species. Anchovy was the most dominant species. In conclusion, the remote telesounder was a convenient and powerful system for investigating the distribution pattern and the swimming behavior of fish entering in set net.

• Journal of Internet Computing and Services
• /
• v.20 no.4
• /
• pp.91-102
• /
• 2019

In autonomous driving systems, the ability to classify pedestrians in images captured by cameras is very important for pedestrian safety. In the past, after extracting features of pedestrians with HOG(Histogram of Oriented Gradients) or SIFT(Scale-Invariant Feature Transform), people classified them using SVM(Support Vector Machine). However, extracting pedestrian characteristics in such a handcrafted manner has many limitations. Therefore, this paper proposes a method to classify pedestrians reliably and effectively using CNN's(Convolutional Neural Network) deep features and transfer learning. We have experimented with both the fixed feature extractor and the fine-tuning methods, which are two representative transfer learning techniques. Particularly, in the fine-tuning method, we have added a new scheme, called M-Fine(Modified Fine-tuning), which divideslayers into transferred parts and non-transferred parts in three different sizes, and adjusts weights only for layers belonging to non-transferred parts. Experiments on INRIA Person data set with five CNN models(VGGNet, DenseNet, Inception V3, Xception, and MobileNet) showed that CNN's deep features perform better than handcrafted features such as HOG and SIFT, and that the accuracy of Xception (threshold = 0.5) isthe highest at 99.61%. MobileNet, which achieved similar performance to Xception and learned 80% fewer parameters, was the best in terms of efficiency. Among the three transfer learning schemes tested above, the performance of the fine-tuning method was the best. The performance of the M-Fine method was comparable to or slightly lower than that of the fine-tuningmethod, but higher than that of the fixed feature extractor method.

• The Bulletin of The Korean Astronomical Society
• /
• v.44 no.1
• /
• pp.69.3-69.3
• /
• 2019

The CHIMPS2 survey is to extend the JCMT HARP $^{13}CO/C^{18}O$ J=3-2 Inner Milky-Way Plane Survey (CHIMPS) and the ${12}^CO$ J=3-2 survey (COHRS) into the inner Galactic Plane, the Central Molecular Zone (CMZ), and a section of the Outer Plane. When combined with the complementary $^{12}CO/^{13}CO/C^{18}O$ J=1-0 survey at the Nobeyama 45m (FUGIN) at matching 15" resolution and sensitivity, and other current CO surveys, the results will provide a complete set of transition data with which to calculate accurate column densities, gas temperatures and turbulent Mach numbers. These will be used to: analyze molecular cloud properties across a range of Galactic environments; map the star-formation efficiency (SFE) and dense-gas mass fraction (DGMF) in molecular gas as a function of position in the Galaxy and its relation to the nature of the turbulence within molecular clouds; determine Galactic structure as traced by molecular gas and star formation; constrain cloud-formation models; study the relationship of filaments to star formation; test current models of the gas kinematics and stability in the Galactic center region and the flow of gas from the disc. It will also provide an invaluable legacy data set for JCMT that will not be superseded for several decades. In this poster, we will present the current status of the CHIMPS2.

• Journal of the Korean Society for Heat Treatment
• /
• v.10 no.4
• /
• pp.232-236
• /
• 1997

A rectangular type circulating fluidized bed (CFB) with an internal nozzle and two partition walls was proposed. In this modified CFB, an internal nozzle and two slanted partition walls were additionally set in the riser. This cold mode apparatus was made of acrylic resin; the riser was 1500mm high and $1000{\times}1000mm^2$ in the cross sectional area, the internal nozzle was 200mm high and 10mm in the inner diameter, and the partition wall was 7mm thick. Glass beads of $91{\mu}m$ in the mean diameter were employed as bed materials. In the cold mode by using the proposed CFB with an internal nozzle, it was possible to change the particle hold up by changing the gas flow ratio of the nozzle to the total(Qn/Qt). It was found that the inflection point which devided the bed structure between the dense and the dilute phase in the riser varied with Qn/Qt.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2024.05a
• /
• pp.137-140
• /
• 2024

가상 및 증강현실(VR/AR), 원격 제어, 산업 자동화를 위한 실시간 애플리케이션이 증가함에 따라 무선 통신 네트워크의 처리량과 지연시간 성능이 중요해졌다. 이에 따라 Wi-Fi 8 에서는 초고신뢰 (Ultra-High Reliable, UHR)을 목표로 표준화가 진행 중이며 다수의 AP가 협력하여 데이터를 전송하는 다중 AP 협력 기법이 핵심 기술로 논의되고 있다. 하지만 기존의 다중 AP 협력 환경에서 협력 전송을 위한 제어 정보로 인한 간섭 증가와 OBSS(Overlapped Basic Service Set) 간섭 문제를 고려하지 못하고 있다. 따라서 본 논문에서는 효율적인 다중 AP 협력 전송을 위한 단순화된 채널 관리 방법(Simplified Channel Management Mechanism, SCMM)을 제안한다. 실험 결과에 따르면, SCMM 이 종래모델 대비 처리량은 평균 21.23% 증가했고, 지연시간은 평균 51.02% 감소했다.

• Journal of the Korean Society of Fisheries and Ocean Technology
• /
• v.27 no.1
• /
• pp.1-12
• /
• 1991

An experiment has been carefully designed and performed to verify the theory for the echointergration technique of estimating the density of fish school by the use of steel spheres in a laboratory tank. The spheres used to simulate a fish school were randomly distributed throughout the insonified volume to produce the acoustic echoes similar to those scattered from real fish schools. The backscattered echoes were measured as a function of target density at tow frequencies of 50kHz and 200kHz. Data acquisition, processing and analysis were performed by means of the microcomputer-based sonar-echo processor including a FFT analyzer. Acoustic scattering characteristics of a 36cm mackerel was investigated by measuring fish echoes with frequencies ranging from 47.8kHz to 52.0kHz. The fluctuation of bottom echoes caused by the effects of fish-school attenuation and multiple scattering which occurred in dense aggregations of fishes was also examined by analyzing the echograms of sardine schools obtained by a 50kHz telesounder in the set-net's bagnet, and the echograms obtained by a scientific echo sounder of 50kHz in the East China Sea, respectively. The results obtained can be summarized as follows: 1. The measured and the calculated echo shapes on the steel sphere used to simulate a fish school were in close agreement. 2. The waveform and amplitude of echo signals by a mackerel without swimbladder fluctuated irregularly with the measuring frequency. 3. When a collection of 30 targets/m super(3) lied the shadow region behind another collection of 5 targets/m super(3), the mean losses in echo energy for the 30 targets/m super(3) were about -0.4dB at 50kHz and about -0.2dB at 200kHz, respectively. 4. In the echograms obtained in the East China Sea, the bottom echoes fluctuated remarkably when the dense aggregations of fish appeared between transducer and seabed. Especially, in the case of the echograms of sardine school obtained in a set-net's bagnet, the disappearance of bottom echoes and the lengthening of the echo trace by fish aggregations were observed. Then the mean density of the sardine school was estimated as 36 fish/m super(3). It suggests that when the distribution density of fishes in oceans is greater than this density, the effects of fish-school attenuation and multiple scattering must be taken into account as a possible source of error in fish abundance estimates. 5. The relationship between mean backscattering strength (, dB) and target density ($\rho$, No./m super(3)) were expressed by the equations: =-46.2+13.7 Log($\rho$) at 50kHz and =-43.9+13.4 Log($\rho$) at 200kHz. 6. The difference between the experimentally derived number and the actual number of targets gradually decreased with an increase in the target density and was within 20% when the density was 30 targets/m super(3). From these results, we concluded that when the number of targets in the insonified volume is large, the validity of the echo-integration technique of estimating the density of fish schools could be expected.

• Journal of the Korean Geotechnical Society
• /
• v.34 no.8
• /
• pp.51-64
• /
• 2018

Korean peninsula is known to be far from the plate boundary and not to generate large-scale earthquakes. However, earthquakes recently occurred in Gyeongju (2016/09/12, $M_L=5.8$) and Pohang (2017/11/15, $M_L=5.4$). The interest in earthquake engineering has increased, and various studies are actively underway by recently events. However, the seismic station network in Korea is less dense than that of the western U.S., resulting in the lack of data for detailed analyses of earthquakes. Therefore, KMA (Korea Meteorological Administration) set up temporary seismic stations and recorded ground motions from aftershocks. In this study, characteristics of Pohang seismic propagation and generation of bedrock motion are analyzed through the aftershock ground motion records at both permanent and temporary stations, as well as through the collected geological structure and site information. As a result, the response at Mangcheon-Li shows evidences of basin effects from both geology structures and measured aftershock motions.

• Structural Engineering and Mechanics
• /
• v.69 no.1
• /
• pp.95-104
• /
• 2019

The Standard ASCE 61-14 proposes the Substitute Structure Method (SSM) as a Nonlinear Static Procedure (NSP) to estimate nonlinear displacement demands at the center of mass of piers or wharves under seismic actions. To account for bidirectional earthquake excitation according to the Standard, results from independent pushover analyses in each orthogonal direction should be combined using either a 100/30 directional approach or a procedure referred to as the Dynamic Magnification Factor, DMF. The main purpose of this paper is to present an evaluation of these NSPs in relation to four wharf model structures on soil conditions ranging from soft to medium dense clay. Results from nonlinear static analyses were compared against benchmark values of relevant Engineering Design Parameters, EDPs. The latter are defined as the geometric mean demands that are obtained from nonlinear dynamic analyses using a set of 30 two-component ground motion records. It was found that SSM provides close estimates of the benchmark displacement demands at the center of mass of the wharf structures. Furthermore, for the most critical pile connection at a landside corner of the wharf the 100/30 and DMF approaches produced displacement, curvature, and force demands that were reasonably comparable to corresponding benchmark values.