• Journal of Information Processing Systems
• /
• v.14 no.6
• /
• pp.1464-1479
• /
• 2018

In face recognition, sometimes the number of available training samples for single category is insufficient. Therefore, the performances of models trained by convolutional neural network are not ideal. The small sample face recognition algorithm based on novel Siamese network is proposed in this paper, which doesn't need rich samples for training. The algorithm designs and realizes a new Siamese network model, SiameseFacel, which uses pairs of face images as inputs and maps them to target space so that the $L_2$ norm distance in target space can represent the semantic distance in input space. The mapping is represented by the neural network in supervised learning. Moreover, a more lightweight Siamese network model, SiameseFace2, is designed to reduce the network parameters without losing accuracy. We also present a new method to generate training data and expand the number of training samples for single category in AR and labeled faces in the wild (LFW) datasets, which improves the recognition accuracy of the models. Four loss functions are adopted to carry out experiments on AR and LFW datasets. The results show that the contrastive loss function combined with new Siamese network model in this paper can effectively improve the accuracy of face recognition.

• Journal of Plant Biotechnology
• /
• v.45 no.4
• /
• pp.364-368
• /
• 2018

The R2R3-type protein IbMYB1 transcription factor is a key regulator for anthocyanin biosynthesis in the storage roots of sweet potatoes. It was previously demonstrated that the IbMYB1 expression stimulates anthocyanin pigmentation in tobacco leaves, arabidopsis and storage roots of sweet potatoes. In this study, we generated the transgenic potato plants that express the IbMYB1 genes, which accumulated high levels of anthocyanins under the control of either the tuber-specific patatin (PAT) promoter or oxidative stress-inducible peroxidase anionic 2 (SWPA2) promoter. The PAT-MYB1 transgenic lines exhibited higher anthocyanin levels in the tuber than the empty vector control (EV) or SWPA2-MYB1 plants. When combined, our results indicated that overexpression of the IbMYB1 is a highly promising strategy for the generation of transgenic plants with enhanced tissue specific anthocyanin production.

• Journal of Korean Neurosurgical Society
• /
• v.62 no.1
• /
• pp.35-45
• /
• 2019

Objective : To describe our experiences with a fully equipped high-end digital subtraction angiography (DSA) system within a hybrid operating room (OR). Methods : A single-plane DSA system with 3-dimensional rotational angiography, cone-beam computed tomography (CBCT), and real-time navigation software was used in our hybrid OR. Between April 2014 and January 2018, 191 sessions of cerebrovascular procedures were performed in our hybrid OR. After the retrospective review of all cases, the procedures were categorized into three subcategorical procedures : combined endovascular and surgical procedure, complementary rescue procedure during intervention and surgery, and frameless stereotaxic operation. Results : Forty-nine of 191 procedures were performed using hybrid techniques. Four cases of blood blister aneurysms and a ruptured posterior inferior cerebellar artery aneurysm were treated using bypass surgery and endovascular trapping. Eight cases of ruptured aneurysm with intracranial hemorrhage (ICH) were treated by partial embolization and surgical clipping. Six cases of ruptured arteriovenous malformation with ICH were treated by Onyx embolization of nidus and subsequent surgical removal of nidus and ICH. Two (5.4%) of the 37 cases of pre-mature rupture during clipping were secured by endovascular coil embolization. In one (0.8%) complicated case of 103 intra-arterial thrombectomy procedures, emergency surgical embolectomy with bypass surgery was performed. In 27 cases of ICH, frameless stereotaxic hematoma aspiration was performed using $XperGuide^{(R)}$ system (Philips Medical Systems, Best, the Netherlands). All procedures were performed in single sessions without any procedural complications. Conclusion : Hybrid OR with a fully equipped DSA system could provide precise and safe treatment strategies for cerebrovascular diseases. Especially, we could suggest a strategy to cope flexibly in complex lesions or unexpected situations in hybrid OR. CBCT with real-time navigation software could augment the usefulness of hybrid OR.

• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.41 no.4
• /
• pp.1-8
• /
• 2018

Whole body fatigue detection is an important phenomenon and the factors contributing to whole body fatigue can be controlled if a mathematical model is available for its assessment. This research study aims at developing a model that categorizes whole body exertion into fatigued and non-fatigued states based on physiological and perceived variables. For this purpose, logistic regression was used to categorize the fatigued and non-fatigued subject as dichotomous variable. Normalized mean power frequency of eight muscles from 25 subjects was taken as physiological variable along with the heart rate while Borg scale ratings were taken as perceived variables. The logit function was used to develop the logistic regression model. The coefficients of all the variables were found and significance level was checked. The detection accuracy of the model for fatigued and non-fatigues subjects was 83% and 95% respectively. It was observed that the mean power frequency of anterior deltoid and the Borg scale ratings of upper and lower extremities were significant in predicting the whole body fatigued when evaluated dichotomously (p < 0.05). The findings can help in better understanding of the importance of combined physiological and perceived exertion in designing the rest breaks for workers involved in squat lifting tasks in industrial as well as health sectors.

• Journal of Radiation Protection and Research
• /
• v.44 no.2
• /
• pp.53-63
• /
• 2019

Background: As neutron fields are always accompanied by gamma rays, it is essential to distinguish neutrons from gamma rays in the detection of neutrons. Neutrons and gamma rays can be separated by pulse shape discrimination (PSD) methods. Recently, we performed characterization of a stilbene scintillator detector and an EJ-301 liquid scintillator detector with a high-speed digitizer DT5730 and investigated optimized PSD variables for both detectors. This study is for providing a basis for developing fast neutron/gamma-ray dual-particle imager. Materials and Methods: We conducted PSD experiments using stilbene scintillator and EJ-301 liquid scintillator and evaluated neutron and gamma ray discriminability of each PSD method with a $^{137}Cs$ gamma source and a $^{252}Cf$ neutron source. We implemented digital signal processing techniques to apply two PSD methods - the charge comparison (CC) method and the constant time discrimination (CTD) method - to distinguish neutrons from gamma rays. We tried to find optimized PSD variables giving the best discriminability in a given experimental condition. Results and Discussion: For the stilbene scintillator detector, the charge comparison method and the constant time discrimination method both delivered the PSD FOM values of 1.7. For the EJ-301 liquid scintillator detector, both PSD methods delivered the PSD FOM values of 1.79. With the same PSD variables, PSD performance was excellent in $300{\pm}100keVee$, $500{\pm}100keVee$, and $700{\pm}100keVee$ energy regions. This result shows that we can achieve an effective discrimination of neutrons from gamma rays using these scintillator detector systems. Conclusion: We applied both PSD methods to a stilbene and a liquid scintillator and optimized the PSD performance represented by FOM values. We observed a good separation performance of both scintillators combined with a high-speed digitizer and digital PSD. These results will provide reference values for the dual-particle imager we are developing, which can image both fast neutrons and gamma rays simultaneously.

• International Journal of Industrial Entomology and Biomaterials
• /
• v.39 no.1
• /
• pp.14-21
• /
• 2019

Genome editing by CRISPR/Cas9, a third-generation gene scissor in molecular breeding at the genome level, is attracting much attention as one of the breeding techniques of the future. In this study, genetic and phenotypic analysis was used to examine the responsiveness of the Bakokjam variety of the silkworm Bombyx mori to molecular breeding using CRISPR/Cas9 in editing the white egg 2 (w-2) gene. The nucleotide sequence of the w-2 gene was analyzed and three different guide RNAs (gRNA) were prepared. The synthesized gRNA was combined with Cas9 protein and then analyzed by T7 endonuclease I after introduction into the Bm-N silkworm cell line. To edit the silkworm gene, W1N and W2P gRNA and Cas9 complexes were microinjected into silkworm embryos. Based on the results of microinjection, the hatching rate was 16-24% and the incidence of mutation was 33-37%. The gene mutation was verified in the heterozygous F1 generation, but no phenotypic change was observed. In F2 homozygotes generated by F1 self-crosses, a mutant phenotype was observed. These results suggest that silkworm molecular breeding using the CRISPR/Cas9 system is possible and will be a very effective way to shorten the time required than the traditional breeding process.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.2
• /
• pp.751-770
• /
• 2019

Action recognition is an essential task in computer vision due to the variety of prospective applications, such as security surveillance, machine learning, and human-computer interaction. The availability of more video data than ever before and the lofty performance of deep convolutional neural networks also make it essential for action recognition in video. Unfortunately, limited crafted video features and the scarcity of benchmark datasets make it challenging to address the multi-person action recognition task in video data. In this work, we propose a deep convolutional neural network-based Effective Hybrid Learning (EHL) framework for two-person interaction classification in video data. Our approach exploits a pre-trained network model (the VGG16 from the University of Oxford Visual Geometry Group) and extends the Faster R-CNN (region-based convolutional neural network a state-of-the-art detector for image classification). We broaden a semi-supervised learning method combined with an active learning method to improve overall performance. Numerous types of two-person interactions exist in the real world, which makes this a challenging task. In our experiment, we consider a limited number of actions, such as hugging, fighting, linking arms, talking, and kidnapping in two environment such simple and complex. We show that our trained model with an active semi-supervised learning architecture gradually improves the performance. In a simple environment using an Intelligent Technology Laboratory (ITLab) dataset from Inha University, performance increased to 95.6% accuracy, and in a complex environment, performance reached 81% accuracy. Our method reduces data-labeling time, compared to supervised learning methods, for the ITLab dataset. We also conduct extensive experiment on Human Action Recognition benchmarks such as UT-Interaction dataset, HMDB51 dataset and obtain better performance than state-of-the-art approaches.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.3
• /
• pp.1260-1283
• /
• 2019

Low earth orbit (LEO) satellite broadband network is a crucial part of the space information network. LEO satellite constellation design is a top-level design, which plays a decisive role in the overall performance of the LEO satellite network. However, the existing works on constellation design mainly focus on the coverage criterion and rarely take network performance into the design process. In this article, we develop a unified framework for constellation optimization design in LEO satellite broadband networks. Several design criteria including network performance and coverage capability are combined into the design process. Firstly, the quality of service (QoS) metrics is presented to evaluate the performance of the LEO satellite broadband network. Also, we propose a network stability model for the rapid change of the satellite network topology. Besides, a mathematical model of constellation optimization design is formulated by considering the network cost-efficiency and stability. Then, an optimization algorithm based on non-dominated sorting genetic algorithm-II (NSGA-II) is provided for the problem of constellation design. Finally, the proposed method is further evaluated through numerical simulations. Simulation results validate the proposed method and show that it is an efficient and effective approach for solving the problem of constellation design in LEO satellite broadband networks.

• The Journal of the Acoustical Society of Korea
• /
• v.40 no.2
• /
• pp.109-120
• /
• 2021

This paper proposes the H∞ norm based data association gate method to apply robustly the data association gate of passive sonar automatic target tracking which is on non-linear targets in dense cluttered environment. For target tracking, data association method selects the measurements within validated gate, which means validated measuring extent, as candidates for the data association. If the extent of the validated gate in the data association is not proper or the data association executes under dense cluttered environment, it is difficult to maintain the robustness of target tracking due to interference of clutter measurements. To resolve this problem, this paper proposes a novel gating method which applies H∞ norm based bisection algorithm combined with 3-σ gate method under Gaussian distribution assumption and tracking error covariance. The proposed method leads to alleviate the interference of clutters and to track the non-linear maneuvering target robustly. Through analytic method and simulation to utilize simulated data of horizontal and vertical bearing measurements, improvement of data association robustness is confirmed contrary to the conventional method.

• Journal of Korean Society on Water Environment
• /
• v.36 no.6
• /
• pp.621-635
• /
• 2020

Water quality models are scientific tools that simulate and interpret the relationship between physical, chemical and biological reactions to external pollutant loads in water systems. They are actively used as a key technology in environmental water management. With recent advances in computational power, water quality modeling technology has evolved into a coupled three-dimensional modeling of hydrodynamics, water quality, and ecological inputs. However, there is uncertainty in the simulated results due to the increasing model complexity, knowledge gaps in simulating complex aquatic ecosystem, and the distrust of stakeholders due to nontransparent modeling processes. These issues have become difficult obstacles for the practical use of water quality models in the water management decision process. The objectives of this paper were to review the theoretical background, needs, and development status of water quality modeling technology. Additionally, we present the potential future directions of water quality modeling technology as a scientific tool for national environmental water management. The main development directions can be summarized as follows: quantification of parameter sensitivities and model uncertainty, acquisition and use of high frequency and high resolution data based on IoT sensor technology, conjunctive use of mechanistic models and data-driven models, and securing transparency in the water quality modeling process. These advances in the field of water quality modeling warrant joint research with modeling experts, statisticians, and ecologists, combined with active communication between policy makers and stakeholders.