• Journal of Broadcast Engineering
• /
• v.26 no.2
• /
• pp.125-131
• /
• 2021

In recent years, Convolutional Neural Networks (CNNs) have achieved outstanding performance in the fields of computer vision such as image classification, object detection, visual quality enhancement, etc. However, as huge amount of computation and memory are required in CNN models, there is a limitation in the application of CNN to low-power environments such as mobile or IoT devices. Therefore, the need for neural network compression to reduce the model size while keeping the task performance as much as possible has been emerging. In this paper, we propose a method to compress CNN models by combining matrix decomposition methods of LR (Low-Rank) approximation and CP (Canonical Polyadic) decomposition. Unlike conventional methods that apply one matrix decomposition method to CNN models, we selectively apply two decomposition methods depending on the layer types of CNN to enhance the compression performance. To evaluate the performance of the proposed method, we use the models for image classification such as VGG-16, RestNet50 and MobileNetV2 models. The experimental results show that the proposed method gives improved classification performance at the same range of 1.5 to 12.1 times compression ratio than the existing method that applies only the LR approximation.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.17 no.1
• /
• pp.33-39
• /
• 2017

In mission-critical applications such as vehicular networks, distributed robotics, and other cyber-physical systems, the requirements for latency are more stringent than traditional applications. Among them, autonomous V2V communication is a rapidly emerging domain of applications with a few milliseconds' latency requirements. Today's systems utilizing 802.11p or LTE-direct standards are not primarily designed for ultra-low latency. Because the medium access function contributes to a significant portion of the total latency, it is necessary to modify Layer2 in order to solve the problem. Focusing on MAC layer, we developed a scalable and latency-guaranteed MAC by devising Autonomous TDMA (ATDMA) in which autonomous joining/leaving is allowed without scheduling by coordinator. We also evaluated the performance of the algorithm by comparing with the WAVE protocol.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.11a
• /
• pp.44-45
• /
• 2019

최근 CNN 등 인공신경망은 최근 이미지 분류, 객체 인식, 자연어 처리 등 다양한 분야에서 뛰어난 성능을 보이고 있다. 그러나, 대부분의 분야에서 보다 더 높은 성능을 얻기 위해 사용한 인공신경망 모델들은 파라미터 수 및 연산량 등이 방대하여, 모바일 및 IoT 디바이스 같은 연산량이나 메모리가 제한된 환경에서 추론하기에는 제한적이다. 따라서 연산량 및 모델 파라미터 수를 압축하기 위한 딥러닝 경량화 알고리즘이 연구되고 있다. 본 논문에서는 임베디트 보드에서의 압축된 CNN 모델의 성능을 검증한다. 인공지능 지원 맞춤형 칩인 QCS605 를 내장한 임베디드 보드에서 카메라로 입력한 영상에 대해서 원 CNN 모델과 압축된 CNN 모델의 분류 성능과 동작속도 비교 분석한다. 본 논문의 실험에서는 CNN 모델로 MobileNetV2, VGG16 을 사용했으며, 주어진 모델에서 가지치기(pruning) 기법, 양자화, 행렬 분해 등의 인공신경망 압축 기술을 적용하였을 때 원래의 모델 대비 추론 시간 및 분류의 정확도 성능을 분석하고 인공신경망 압축 기술의 유용성을 확인하였다.

• Bulletin of the Korean Chemical Society
• /
• v.23 no.9
• /
• pp.1257-1262
• /
• 2002

$[(C_2B_9H_{11})Mo(CO)_3]_2PPN_2$ $(2{\cdot}PPN_2)$, a new precursor for the oxidative decarbonylation reaction, was synthesized in high yield by the one-electron oxidation reaction of $[(C_2B_9H_{11})Mo(CO)_3]PPN_2$. $2{\cdot}PPN_2$ was structurally characterized, showing a dimeric nature with long (3.321 ${\AA}$) Mo-Mo bonding. Reaction of $2{\cdot}PPN_2$ with sulfur gave the completely decarbonylated product $[(C_2B_9H_{11})Mo({\mu}-S)(S)]_2PPN_2$ ($3{\cdot}PPN_2$). The ligand substitution of the terminal sulfur ligands in $3{\cdot}PPN_2$ to oxygen ligands was carried out with the use of PhIO to give $[(C_2B_9H_{11})Mo({\mu}-S)(O)]_2PPN_2$ ($4{\cdot}PPN_2$). The structures of $3{\cdot}PPN_2$ and $4{\cdot}PPN_2$ were also studied.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.16 no.1
• /
• pp.157-165
• /
• 2017

Intelligent transportation systems are currently being developed for elemental technology development of cooperative intelligent transport systems, which enable vehicles to communicate with each other or reduce the risk of traffic accidents, We have been defining and standardizing services according to the purpose of solving traffic safety problems depending on countries. Therefore, in this study, the developed countries of V2X(vehicle-to-everything) based on USA, Europe, Japan, etc., analyzed the service cases selected in the field demonstration stage after completion of the element technology devanalyzed the service cases selected in the field demonstration stage after completion of the element technology development, and to suggest the direction of futureelopment, and to suggest the direction of future policy direction.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.66 no.1
• /
• pp.249-254
• /
• 2017

This paper presents an assessment system of safety level of 22.9kV grade high voltage electrical facilities considering environmental factors. The assessment system was developed based on the following procedure. Firstly, assessment structure was determined by consulting standards regarding inspection and diagnosis of electrical facilities. Secondly, contents of items and sub items of assessment system were developed. Thirdly, in order to quantify the importance of the assessment system, the weight was calculated using Analytic Hierarchy Process(AHP). Lastly, assessment table of safety level was developed including environmental factors such as period of use and load factor. The developed system can evaluate the safety level of high voltage facilities in an objective way. Therefore it can be applicable to electrical safety management system based on Internet of Things(IoT).

• Advances in aircraft and spacecraft science
• /
• v.9 no.6
• /
• pp.525-536
• /
• 2022

The current state of telecommunications infrastructure in the Arctic does not allow providing a wide range of required services for people, businesses and other categories, which necessitates the use of non-traditional approaches to its organization. The paper proposes an innovative approach to building a combined communication network based on tethered high-altitude platform station (HAPS) located at an altitude of 1-7 km and connected via radio channels with terrestrial and satellite communication networks. Network configuration and composition of telecommunication equipment placed on HAPS and located on the terrestrial and satellite segment of the network was justified. The availability of modern equipment and the distributed structure of such an integrated network will allow, unlike existing networks (Iridium, Gonets, etc.), to organize personal mobile communications, data transmission and broadband Internet up to 100 Mbps access for mobile and fixed subscribers, rapid transmission of information from Internet of Things (IoT) sensors and unmanned aerial vehicles (UAV). A substantiation of the possibility of achieving high network capacity in various paths is presented: inter-platform radio links, subscriber radio links, HAPS feeder lines - terrestrial network gateway, HAPS radio links - satellite retransmitter (SR), etc. The economic efficiency of the proposed solution is assessed.

• Journal of the Korean Society of Food Science and Nutrition
• /
• v.44 no.6
• /
• pp.904-911
• /
• 2015

The objective of this study was to investigate the quality characteristics of traditional wine using an radio-frequency identification (RFID) system annexed to a fermenter. In this study, we proposed an RFID-based data transmission scheme for monitoring fermentation of traditional alcoholic beverages. The pH, total acidity, total sugar, soluble sugar, free sugar, alcohol content, and organic acids of were investigated and subjected to fermentation of traditional alcoholic beverages three times. The pH ranged from 7.98, 7.95, and 7.68 at day 0, decreased drastically to 3.31~2.96 at day 2, and then slowly increased to the end point, finally reaching 3.34 at day 20. Acidity tended to increase quickly with time, especially for all samples after day 2. The fermentation environment induced a sudden increase acidity in reactants and indicated a low pH. The total sugars during fermentation quickly decreased to the range of 20.3, 22.43, and 19.2% at day 2, and the slope of reduction steadily decreased to 5.1, 6.1, and 4.8% at day 10. On the other hand, the alcohol content showed the reverse trend as total sugars. The alcohol content also showed the same pattern as total acids, showing the highest alcohol content of 17.3% (v/v) on day 20. In this study on traditional wine fermentation using an RFID system, we showed that pH, soluble sugar, and alcohol content can be adopted as key indicators for quality control and standardization of traditional wine manufacturing.

• Tuberculosis and Respiratory Diseases
• /
• v.52 no.1
• /
• pp.14-23
• /
• 2002

Background: The opitmal ventilator setting during partial liquid ventilation(PLV) is controversial. This study investigated the effects of various gas exchange parameters during PLV in normal rabbit lungs in order to aid in the development of an optimal ventilator setting during PLV. Methods: Seven New-Zealand white rabbits were ventilated in pressure-controlled mode with the following settings; tidal volume($V_T$) 8 mL/kg, positive end-expiratory pressure(PEEP) 4 $cmH_2O$, inspiratory-to-expiratory ratio(I:E ratio) 1:2, fraction of inspired oxygen($F_TO_2$) 1.0. The respiration rate(RR) was adjusted to keep $PaCO_2$ between 35~45 mmHg. The ventilator settings were changed every 30 min in the following sequence : (1) Baseline, as the basal ventilator setting, (2) Inverse ratio, I:E ratio 2:1, (3) high PEEP, adjust PEEP to achieve the same mean inspiratory pressure (MIP) as in the inverse ratio, (4) High $V_T$, $V_T$ 15 mL/kg, (5) high RR, the same minute ventilation (MV) as in the High $V_T$. Subsequently, the same protocol was repeated after instilling 18 mL/kg of perfluorodecalin for PLV. The parameters of gas exchange, lung mechanics, and hemodynamics were examined. Results: (1) The gas ventilation(GV) group showed no significant changes in the $PaO_2$ at all phases. The $PaCO_2$ was lower and the pH was higher at the high $V_T$ and high RR phases(p<0.05). No significant changes in the lung mechanics and hemodynamics parameters were observed. (2) The baseline $PaO_2$ for the PLV was $312{\pm}$ mmHg. This was significantly lower when decreased compared to the baseline $PaO_2$ for GV which was $504{\pm}81$ mmHg(p=0.001). During PLV, the $PaO_2$, was significantly higher at the high PEEP($452{\pm}38$ mmHg) and high $V_T$ ($461{\pm}53$ mmHg) phases compared with the baseline phase. However, it did not change significantly during the inverse I:E ratio or the high RR phases. (3) The $PaCO_2$ was significantly lower at high $V_T$ and RR phases for both the GV and PLV. During the PLV, $PaCO_2$ were significantly higher compared to the GV (p<0.05). (4) There were no important or significant changes in of baseline and high RR phases lung mechanics and hemodynamics parameters during the PLV. Conclusion: During PLV in the normal lung, adequate $V_T$ and PEEP are important for optimal oxygenation.

• Journal of the Korea institute for structural maintenance and inspection
• /
• v.26 no.2
• /
• pp.28-36
• /
• 2022

As port structures are exposed to various extreme external loads such as wind (typhoons), sea waves, or collision with ships; it is important to evaluate the structural safety periodically. To monitor the port structure, especially the rubber fender, a fender segmentation system using a vision sensor and deep learning method has been proposed in this study. For fender segmentation, a new deep learning network that improves the encoder-decoder framework with the receptive field block convolution module inspired by the eccentric function of the human visual system into the DenseNet format has been proposed. In order to train the network, various fender images such as BP, V, cell, cylindrical, and tire-types have been collected, and the images are augmented by applying four augmentation methods such as elastic distortion, horizontal flip, color jitter, and affine transforms. The proposed algorithm has been trained and verified with the collected various types of fender images, and the performance results showed that the system precisely segmented in real time with high IoU rate (84%) and F1 score (90%) in comparison with the conventional segmentation model, VGG16 with U-net. The trained network has been applied to the real images taken at one port in Republic of Korea, and found that the fenders are segmented with high accuracy even with a small dataset.