• KIPS Transactions on Software and Data Engineering
• /
• v.9 no.12
• /
• pp.403-410
• /
• 2020

In addition to the research on noise removal and super-resolution using the data restoration (Output result) function of Autoencoder, research on the performance improvement of clustering using the dimension reduction function of autoencoder are actively being conducted. The clustering function and data restoration function using Autoencoder have common points that both improve performance through the same learning. Based on these characteristics, this study conducted an experiment to see if the autoencoder model designed to have excellent data recovery performance is superior in clustering performance. Skip connection technique was used to design autoencoder with excellent data recovery performance. The output result performance and clustering performance of both autoencoder model with Skip connection and model without Skip connection were shown as graph and visual extract. The output result performance was increased, but the clustering performance was decreased. This result indicates that the neural network models such as autoencoders are not sure that each layer has learned the characteristics of the data well if the output result is good. Lastly, the performance degradation of clustering was compensated by using both latent code and skip connection. This study is a prior study to solve the Hanja Unicode problem by clustering.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.43 no.3 s.309
• /
• pp.82-89
• /
• 2006

Palmprint-based personal identification system, as a new member in the biometrics system family, has become an active research topic in recent years. Although lots of methods have been made, how to represent palmprint for effective classification is still an open problem and conducting researches. In this paper, the palmprint classification and recognition method based on PCA (Principal Components Analysis) using the dimension reduction of singular vector is proposed. And the 135dpi palmprint image which is obtained by the palmprint acquisition device is used for the effectual palmprint recognition system. The proposed system is consists of the palmprint acquisition device, DB generation algorithm and the palmprint recognition algorithm. The palmprint recognition step is limited 2 times. As a results GAR and FAR are 98.5% and 0.036%.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.53 no.7
• /
• pp.49-55
• /
• 2016

In digital forensic images, there is a serious problem that is distributed with various image types. For the problem solution, this paper proposes a classification algorithm of the forensic image types. The proposed algorithm extracts the 21-dim. feature vector with the contrast and energy from GLCM (Gray Level Co-occurrence Matrix), and the entropy of each image type. The classification test of the forensic images is performed with an exhaustive combination of the image types. Through the experiments, TP (True Positive) and FN (False Negative) is detected respectively. While it is confirmed that performed class evaluation of the proposed algorithm is rated as 'Excellent(A)' because of the AUROC (Area Under Receiver Operating Characteristic Curve) is 0.9980 by the sensitivity and the 1-specificity. Also, the minimum average decision error is 0.1349. Also, at the minimum average decision error is 0.0179, the whole forensic image types which are involved then, our classification effectiveness is high.

• Journal of the Korean earth science society
• /
• v.38 no.4
• /
• pp.263-268
• /
• 2017

The determination algorithm of the location of a line source with strike and dip using the gravity gradient tensor on a single profile is proposed. We already proposed the determination of strike and dip in the previous paper and then, now we improved the algorithm to locate a line source after determining strike and dip. The strike and dip of the line source can be determined by rotating the gravity gradient tensor matrix as reducing 2 independent components. Using the ratio of remaining 2 components, the location can be determined by the least square manner of the pointing vectors on each observation point. A synthetic model is tested for proving the usefulness of the proposed algorithm.

• The Journal of Korean Institute of Electromagnetic Engineering and Science
• /
• v.29 no.7
• /
• pp.508-514
• /
• 2018

An electroencephalogram (EEG) evaluates the electrical activity generated by brain cell interactions that occur during brain activity, and an EEG can evaluate the brain activity caused by hand movement. In this study, a 16-channel EEG was used to measure the EEG generated before and after hand movement. The measured data can be classified as a supervised learning model, a support vector machine (SVM). To shorten the learning time of the SVM, a feature extraction and vector dimension reduction by filtering is proposed that minimizes motion-related information loss and compresses EEG information. The classification results showed an average of 72.7% accuracy between the sitting position and the hand movement at the electrodes of the frontal lobe.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.19 no.6
• /
• pp.109-116
• /
• 2019

This study proposed a robust detection algorithm. It detects face more stably with respect to changes in light and rotation for the identification of a face shape. Also it satisfies both efficiency of calculation and the function of detection. The algorithm proposed segmented the face area through pre-processing using a face shape as input information in an environment with a single camera and then identified the shape using a Self Organized Feature Map(SOFM). However, as it is not easy to exactly recognize a face area which is sensitive to light, it has a large degree of freedom, and there is a large error bound, to enhance the identification rate, rotation information on the face shape was made into a database and then a principal component analysis was conducted. Also, as there were fewer calculations due to the fewer dimensions, the time for real-time identification could be decreased.

• Convergence Security Journal
• /
• v.12 no.6
• /
• pp.69-75
• /
• 2012

MANET(Mobile Ad-hoc Network) is target of many attacks because of dynamic topology and hop-by-hop data transmission method. In MANET, location setting of intrusion detection system is difficult and attack detection using information collected locally is more difficult. The amount of traffic grow, intrusion detection performance will be decreased. In this paper, MANET is composed of zone form and we used random projection technique which reduces dimension without loss of information in order to perform stable intrusion detection in even massive traffic. Global detection node is used to detect attacks which are difficult to detect using only local information. In the global detection node, attack detection is performed using received information from IDS agent and pattern of nodes. k-NN and ZBIDS were experimented to evaluate performance of the proposed technique in this paper. The superiority of performance was confirmed through the experience.

• Journal of KIISE
• /
• v.44 no.11
• /
• pp.1194-1208
• /
• 2017

A traditional average smoothing method is designed for smoothing out noise, which, however, unintentionally results in smooth corner points on the curvature accompanied with a shrinkage of curves. In this paper, we propose a novel curve smoothing method via polygonal approximation of the input curve. The proposed method determines the smoothing weight for each point of the input curve based on the angle and approximation error between the approximated polygon and the input curve. The weight constrains a displacement of the point after smoothing not to significantly exceed the average noise error of the region. In the experiment, we observed that the resulting smoothed curve is close to the original curve since the point moves toward the average position of the noise after smoothing. As an application to digital cartography, for the same amount of smoothing, the proposed method yields a less area reduction even on small curve segments than the existing smoothing methods.

• Proceedings of the Korea Technical Association of the Pulp and Paper Industry Conference
• /
• 2001.04a
• /
• pp.168-168
• /
• 2001

제지공정의 지절 현상은 많은 공정 변수들이 복합적으로 작용하여 발생하는 가장 큰 공정 트러블 중의 하나이다. 지절은 생산량 감소 뿐만 아니라 발생 후 공정의 복구 와 정리, 생산재가동 및 공정의 재안정화를 위해 많은 시간과 비용, 그리고 노력이 투 입되어야 하므로 공정의 효율과 생산성을 크게 저하시키는 요인이다. 그러나 지절 현상 의 복잡성 때문에 이에 대해 쉽게 접근하거나 해결하지 못하고 있는 것이 현실이지만 그 필요성은 더욱 더 증대되고 있다. 본 연구에서는 최근 들어 각종 산업분야에서 복잡 한 공정상의 결점 발견 및 진단에 효과적이라고 인정받고 있는 예측 분석기법인 인공 신경망(artificial neural network) 시율레이션과 일반적인 통계기법 중의 하나인 주성분 분석을 이용하여 제지 공정의 지절 현상의 검토 가능성을 타진하였다. 인공신경망이란 인간두뇌에서 일어나는 자극-반응-학습과정을 모사하여 현실세계에 존재하는 다양한 현상들의 업력벡터와 출력상태 간의 비선형 mapping올 컴퓨터 시율 레이션을 통하여 분석하고자 하는 기법으로, 여러 가지 현상들을 학습을 통해서 인식하 는 신경망 내의 신경단위들이 병렬처리에 의해 많은 양의 자료에 대한 추론이나 판단 을 신속하고 정확하게 해주는 특징이 있으며 실시간 패턴인식이나 분류 응용분야에도 매우 매력적으로 이용되고 있는 방법이다. 이러한 인공 신경망 기법 중에서도 본 연구 에서는 퍼셉트론의 한계점을 극복하기 위하여 입력총과 출력층에 한 개 이상의 은닉층 ( (hidden layer)을 사용하여 다층 네트워으로 구성하고, 모든 입력패턴에 대하여 발생하 는 오차함수를 최소화하는 방향으로 연결강도를 조정하는 back propagation 학습 알고 리즘을 사용하였다. 지절의 원인으로 추정 가능한 공정인자들을 변수로 하여 최적의 인 공신경망을 구축하기 위해 학습률과 모멘트 상수의 변화 및 은닉층의 수와 출력층의 뉴런 수를 조절하는 동의 작업을 거쳐 네트워크의 정확도가 높은 인공신경망을 설계하 였다. 또한 이러한 인공신경망과의 비교분석을 위해 동일한 공정 데이터들올 이용하여 보편적으로 사용하는 통계기법 중의 하나인 주성분회귀분석을 실시하였다. 주성분 분석은 여러 개의 반응변수에 대하여 얻어진 다변량 자료의 다차원적인 변 수들을 축소, 요약하는 차원의 단순화와 더불어 서로 상관되어있는 반응변수들 상호간 의 복잡한 구조를 분석하는 기법이다. 본 발표에서는 공정 자료를 활용하여 인공신경망 과 주성분분석을 통해 공정 트러블의 발생에 영향 하는 인자들을 보다 현실적으로 추 정하고, 그 대책을 모색함으로써 이를 최소화할 수 있는 방안을 소개하고자 한다.

• The Korean Journal of Applied Statistics
• /
• v.33 no.1
• /
• pp.61-74
• /
• 2020

Interval-valued data, one type of symbolic data, are observed in the form of intervals rather than single values. Each interval-valued observation has an internal variation. Principal component analysis reduces the dimension of data by maximizing the variance of data. Therefore, the principal component analysis of the interval-valued data should account for the variance between observations as well as the variation within the observed intervals. In this paper, three principal component analysis methods for interval-valued data are summarized. In addition, a new method using a truncated normal distribution has been proposed instead of a uniform distribution in the conventional quantile method, because we believe think there is more information near the center point of the interval. Each method is compared using simulations and the relevant data set from the OECD. In the case of the quantile method, we draw a scatter plot of the principal component, and then identify the position and distribution of the quantiles by the arrow line representation method.