• Journal of Animal Science and Technology
• /
• v.56 no.4
• /
• pp.15.1-15.10
• /
• 2014

Background: Morphologically classifying embryos is important for numerous laboratory techniques, which range from basic methods to methods for assisted reproduction. However, the standard method currently used for classification is subjective and depends on an embryologist's prior training. Thus, our work was aimed at developing software to classify morphological quality for blastocysts based on digital images. Methods: The developed methodology is suitable for the assistance of the embryologist on the task of analyzing blastocysts. The software uses artificial neural network techniques as a machine learning technique. These networks analyze both visual variables extracted from an image and biological features for an embryo. Results: After the training process the final accuracy of the system using this method was 95%. To aid the end-users in operating this system, we developed a graphical user interface that can be used to produce a quality assessment based on a previously trained artificial neural network. Conclusions: This process has a high potential for applicability because it can be adapted to additional species with greater economic appeal (human beings and cattle). Based on an objective assessment (without personal bias from the embryologist) and with high reproducibility between samples or different clinics and laboratories, this method will facilitate such classification in the future as an alternative practice for assessing embryo morphologies.

• Journal of Information Management
• /
• v.41 no.3
• /
• pp.27-44
• /
• 2010

Most research about patent data analyzes the trend of technologies using a Patent Map(PM), and suggests the frequencies and trend of patents in a certain topic using tables or graphs in Excel. However, more advanced analysis tools are recently needed to compare the trends among national and international industries. This research discussed why statistical analysis is needed to improve the reliability in PM analysis, and the research compares the trends of patents in Korea between 1990 and 2004 by years, International Patent Classification(IPC) sections, and countries using the frequencies and Poisson regression model. The statistical analysis is also suggested and applied to R&D studies.

• Journal of KIISE:Software and Applications
• /
• v.37 no.10
• /
• pp.723-735
• /
• 2010

Evaluating the level of achievement and providing the knowledge which is appropriate at the evaluated level have great influence in studying of the human beings. This shows the importance of the order of training and the training order should be considered in machine learning. In this research, to assess the influence of the order of training, we propose a method of controlling the order of training samples utilizing the experience of supervisor in the training of MLP. The supervisor finds out the current state of MLP using teaching experience and student evaluation, and then selects the most instructive sample for MLP in that state. We use CRF to represent and utilize the experience of supervisor. While the proposed method is similar to active learning in selecting samples, it is basically different in that selection is not to reduce the number of samples to be used but to assist the learning progress. The result from classification problem shows that the method is usually effective in terms of time taken in training in contrast to random selection.

• Journal of KIISE:Software and Applications
• /
• v.30 no.3_4
• /
• pp.358-371
• /
• 2003

Automatic word spacing is a process of deciding correct boundaries between words in a sentence including spacing errors. It is very important to increase the readability and to communicate the accurate meaning of text to the reader. The previous statistical approaches for automatic word spacing do not consider the previous spacing state, and thus can not help estimating inaccurate probabilities. In this paper, we propose two statistical word spacing models which can solve the problem of the previous statistical approaches. The proposed models are based on the observation that the automatic word spacing is regarded as a classification problem such as the POS tagging. The models can consider broader context and estimate more accurate probabilities by generalizing hidden Markov models. We have experimented the proposed models under a wide range of experimental conditions in order to compare them with the current state of the art, and also provided detailed error analysis of our models. The experimental results show that the proposed models have a syllable-unit accuracy of 98.33% and Eojeol-unit precision of 93.06% by the evaluation method considering compound nouns.

• Journal of agricultural medicine and community health
• /
• v.15 no.1
• /
• pp.3-8
• /
• 1990

Seven specimens of salmon(Oncorhynchus keta) purchased directly, in Oct. 10. 1990, in the Taep'o port. Kang-won-do were examined for infection state of anisakid larvae and classification of extracted larvae according to morphological characters. The results were as follows. 1) From seven salmon 202 anisakid larvae were found, and mean infection number of anisakid larvae per individual salmon was 28.86. 2) From total extracted anisakid larvae 198 larvae(98%) were found in muscle. Therefore the distribution of anisakid larvae in intestine was extremely rare compared to in muscle. 3) The percentage of anisakid larvae in II, IV muscle region was 93% and from this we could surmise that most anisakid larvae in salmon did not penetrate to the very distanted muscle from intestine. 4) Three types of anisakid larvae(Anisakis Type I, Contracaecum Type B, Contracaecum Type D) were identified and, among them, Contracaecum Type B was the first recording type in Korea 5) Larvae of Contracaecum it genus were found only in intestine. Therefore it surmised that penetration neture to muscle of Contracaecum larvae was less than that of Anisakis Type I.

• Journal of Digital Contents Society
• /
• v.16 no.5
• /
• pp.671-680
• /
• 2015

The Joint Bayesian[1] method was published in 2012. Since then, it has been used for binary classification in almost all state-of-the-art face recognition methods. However, no improved methods have been published so far except 2D-JB[2]. In this paper we propose an improved version of the JB method that considers the features of both the given face image and its mirror image. In pattern classification, it is very likely to make a mistake when the value of the decision function is close to the decision boundary or the threshold. By making the value of the decision function far from the decision boundary, the proposed method reduces the errors. The experimental results show that the proposed method outperforms the JB and 2D-JB methods by more than 1% in the challenging LFW DB. Many state-of-the-art methods required tons of training data to improve 1% in the LFW DB, but the proposed method can make it in an easy way.

• Vascular Specialist International
• /
• v.34 no.4
• /
• pp.109-116
• /
• 2018

Purpose: Diabetic foot wound (DFW) is known as a major contributor of nontraumatic lower extremity amputation. We aimed to evaluate overall amputation rates and risk factors for amputation in patients with DFW. Materials and Methods: From January 2014 to December 2017, 141 patients with DFW were enrolled. We determined rates and risk factors of major amputation in DFW and in DFW with peripheral arterial occlusive disease (PAOD). In addition, we investigated rates and predictors for amputation in diabetic foot ulcer (DFU). Results: The overall rate of major amputation was 26.2% in patients with DFW. Among 141 DFWs, 76 patients (53.9%) had PAOD and 29 patients (38.2%) of 76 DFWs with PAOD underwent major amputation. Wound state according to Wagner classification, congestive heart failure, leukocytosis, dementia, and PAOD were the significant risk factors for major amputation. In DFW with PAOD, Wagner classification grades and leukocytosis were the predictors for major amputation. In addition, amputation was performed for 28 patients (38.4%) while major amputation was performed for 5 patients (6.8%) of 73 DFUs. Only the presence of osteomyelitis (OM) showed significant difference for amputation in DFU. Conclusion: This study represented that approximately a quarter of DFWs underwent major amputation. Moreover, over half of DFW patients had PAOD and about 38.2% of them underwent major amputation. Wound state and PAOD was major predictors for major amputation in DFW. Systemic factors, such as CHF, leukocytosis, and dementia were identified as risk factors for major amputation. In terms of DFU, 38.4% underwent amputation and the presence of OM was a determinant for amputation.

• 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.

• Database Research
• /
• v.34 no.3
• /
• pp.45-57
• /
• 2018

The drowsiness that occurs in the driving is a very dangerous driver condition that can be directly linked to a major accident. In order to prevent drowsiness, there are traditional drowsiness detection methods to grasp the driver's condition, but there is a limit to the generalized driver's condition recognition that reflects the individual characteristics of drivers. In recent years, deep learning based state recognition studies have been proposed to recognize drivers' condition. Deep learning has the advantage of extracting features from a non-human machine and deriving a more generalized recognition model. In this study, we propose a more accurate state recognition model than the existing deep learning method by learning image and PPG at the same time to grasp driver's condition. This paper confirms the effect of driver's image and PPG data on drowsiness detection and experiment to see if it improves the performance of learning model when used together. We confirmed the accuracy improvement of around 3% when using image and PPG together than using image alone. In addition, the multimodal deep learning based model that classifies the driver's condition into three categories showed a classification accuracy of 96%.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.32 no.1
• /
• pp.89-98
• /
• 2022

As deep learning technology was applied to various fields, research on adversarial attack techniques, a security problem of deep learning models, was actively studied. adversarial attacks have been mainly studied in the field of images. Recently, they have even developed a complete decision-based attack technique that can attack with just the classification results of the model. However, in the case of the audio field, research is relatively slow. In this paper, we applied several decision-based attack techniques to the audio field and improved state-of-the-art attack techniques. State-of-the-art decision-attack techniques have the disadvantage of requiring many queries for gradient approximation. In this paper, we improve query efficiency by proposing a method of reducing the vector search space required for gradient approximation. Experimental results showed that the attack success rate was increased by 50%, and the difference between original audio and adversarial examples was reduced by 75%, proving that our method could generate adversarial examples with smaller noise.