• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제16권2호
• /
• pp.365-380
• /
• 2022

The target detection algorithm based on supervised learning is the current mainstream algorithm for target detection. A high-quality dataset is the prerequisite for the target detection algorithm to obtain good detection performance. The larger the number and quality of the dataset, the stronger the generalization ability of the model, that is, the dataset determines the upper limit of the model learning. The convolutional neural network optimizes the network parameters in a strong supervision method. The error is calculated by comparing the predicted frame with the manually labeled real frame, and then the error is passed into the network for continuous optimization. Strongly supervised learning mainly relies on a large number of images as models for continuous learning, so the number and quality of images directly affect the results of learning. This paper proposes a dataset STAR-24K (meaning a dataset for Space TArget Recognition with more than 24,000 images) for detecting common targets in space. Since there is currently no publicly available dataset for space target detection, we extracted some pictures from a series of channels such as pictures and videos released by the official websites of NASA (National Aeronautics and Space Administration) and ESA (The European Space Agency) and expanded them to 24,451 pictures. We evaluate popular object detection algorithms to build a benchmark. Our STAR-24K dataset is publicly available at https://github.com/Zzz-zcy/STAR-24K.

• 한국건축시공학회:학술대회논문집
• /
• 한국건축시공학회 2012년도 추계 학술논문 발표대회
• /
• pp.153-154
• /
• 2012

The accuracy of cost estimation at an early stage in school building project is one of the critical factors for successful completion. So many method and techniques have developed that can estimate construction cost using limited information available in the early stage. Among the techniques, Support Vector Machine(SVM) has received attention in various field due to its excellent capacity for self-learning and generalization performance. Therefore, the purpose of this study is to verify the applicability of cost prediction model based on SVM in school building project at the early stage. Data used in this study are 139 school building cost constructed from 2004 to 2007 in Gyeonggi-Do. And prediction error rate of 7.48% in support vector machine is obtained. So the results showed applicability of using SVM model for predicting construction cost of school building projects.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 1989년도 하계종합학술대회 논문집
• /
• pp.101-105
• /
• 1989

The iterative learning controller makes the system output follow the desired output over a finite time interval through iterating trials. In this paper, first we discuss that the design problem of learning controller is originally the design problem of the inverse model. Then we show that the tracking error which is the difference between the desired output and the system output is reduced monotonically by properly modeled inverse system if the magnitude of the learning operator being introduced is bounded within the unit circle in complex domain. Also it would be shown that the conventional learning control method is a kind of extremely simplified inverse model learning control method of the objective controlled system. Hence this control method can be considered as a generalization of the conventional learning control method. The more a designer model the objective controlled system precisely, the better the performance of the approximated inverse model learning controller would be. Finally we compare the performance of the conventional learning control method with that of the approximated inverse model learning control method by computer simulation.

• 전기전자학회논문지
• /
• 제23권4호
• /
• pp.1408-1414
• /
• 2019

전이학습을 수행한 심층 인공신경망을 압축센싱 심혈관 자기공명영상에 적용하였다. 전이학습은 선행학습 신경망의 구조나 필터 커널, 가중치를 현재의 학습이나 응용에 활용하는 방법이다. 전이학습은 학습 속도를 향상시키고, 학습 데이터가 제한적일 때 신경망의 일반화에 도움이 된다. 8명의 건강한 지원자가 참여한 심장 자기공명영상 실험에서 전이학습을 수행한 신경망은 단독학습 신경망에 비해 학습시간이 5배 이상 단축되었다. 시험 데이터에 대해서도 전이학습을 수행한 신경망은 전이학습을 수행하지 않은 신경망에 비하여 낮은 정규화 평균제곱오차와 향상된 재구성 영상화질을 보였다.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• 제2권1호
• /
• pp.43-48
• /
• 2002

This paper discusses the application of fuzzy-ARTMAP neural network to compensate the nonlinearity of satellite communication channel. The fuzzy-ARTMAP is the class of ART(adaptive resonance theory) architectures designed fur supervised loaming. It has capabilities not fecund in other neural network approaches, that includes a small number of parameters, no requirements fur the choice of initial weights, automatic increase of hidden units, and capability of adding new data without retraining previously trained data. By a match tracking process with vigilance parameter, fuzzy-ARTMAP neural network achieves a minimax teaming rule that minimizes predictive error and maximizes generalization. Thus, the system automatically leans a minimal number of recognition categories, or hidden units, to meet accuracy criteria. As a input-converting process for implementing fuzzy-ARTMAP equalizer, the sigmoid function is chosen to convert actual channel output to the proper input values of fuzzy-ARTMAP. Simulation studies are performed over satellite nonlinear channels. QPSK signals with Gaussian noise are generated at random from Volterra model. The performance of proposed fuzzy-ARTMAP equalizer is compared with MLP equalizer.

• 정보처리학회논문지D
• /
• 제10D권1호
• /
• pp.23-32
• /
• 2003

대규모 데이터 마이닝 환경에서는 이질적인 데이터베이스 혹은 파일 시스템으로부터 분석 대상 데이터를 수집하는 경우가 일반적이므로, 수집된 데이터가 서로 다른 추상화 수준(abstraction level)으로 표현되기 마련이다, 본 논문에서는 기존의 결정 트리(decision tree)를 서로 다른 추상화 수준으로 표현된 데이터에 적용할 때, 분류상 모순이 일어날 수 있음을 보이고, 그에 대한 해결방안을 제시한다. 제안하는 방법은 데이터 간에 존재하는 일반화/세분화 관련성을 결정 트리의 구축 단계는 물론, 클래스 할당 단계에도 반영하여 데이터간의 의미적 연관성을 효과적으로 활용할 수 있도록 한다. 아울러 실제 데이터에 기반을 둔 실험을 통해, 제안한 방법이 기존 방법보다 분류 오류율을 현저히 줄일 수 있음을 보인다.

• 한국지능시스템학회:학술대회논문집
• /
• 한국퍼지및지능시스템학회 2000년도 춘계학술대회 학술발표 논문집
• /
• pp.47-50
• /
• 2000

In this paper, an optimal identification method using Multi-FNN(Fuzzy-Neural Network) is proposed for model ins of nonlinear complex system. In order to control of nonlinear process with complexity and uncertainty of data, proposed model use a HCM clustering algorithm which carry out the input-output data preprocessing function and Genetic Algorithm which carry out optimization of model. The proposed Multi-FNN is based on Yamakawa's FNN and it uses simplified inference as fuzzy inference method and Error Back Propagation Algorithm as learning rules. HCM clustering method which carry out the data preprocessing function for system modeling, is utilized to determine the structure of Multi-FNN by means of the divisions of input-output space. Also, the parameters of Multi-FNN model such as apexes of membership function, learning rates and momentum coefficients are adjusted using genetic algorithms. Also, a performance index with a weighting factor is presented to achieve a sound balance between approximation and generalization abilities of the model, To evaluate the performance of the proposed model, we use the time series data for gas furnace and the numerical data of nonlinear function.

• 전기학회논문지
• /
• 제57권12호
• /
• pp.2255-2262
• /
• 2008

This paper presents a design parameter calculation methodology and its realization to construction for the damped oscillatory impulse current generator(ICG) modelled as damping factor $\alpha$. Matlab internal functions, "fzero" and "polyfit" are applied to find a which are solutions of second order nonlinear equation related with three wave parameters $T_{1},T_{2}$ and $I_{os}$. The calculation results for standard impulse current waveforms such as 4/10${\mu}s$, 8/20${\mu}s$ and 30/80${\mu}s$ show very good accuracy and this results make it possible to extend to generalization in the design of damped oscillatory lCG with any capacitor. 8/20${\mu}s$ ICG based on the calculated design circuit parameters is fabricated in consideration of the nonlinear load(MOV) variation. Comparisons of the tested waveforms with the designed estimation show error within 10% for the waveform tolerance recommended in IEC 60060-1 and IEEE std. C62.45.

• 대한전기학회논문지:시스템및제어부문D
• /
• 제52권1호
• /
• pp.64-75
• /
• 2003

Medical image segmentation is the process by which an original image is partitioned into some homogeneous regions like bones, soft tissues, etc. This study demonstrates an automatic medical image segmentation technique based on independent component analysis. Independent component analysis is a generalization of principal component analysis which encodes the higher-order dependencies in the input in addition to the correlations. It extracts statistically independent components from input data. Use of automatic medical image segmentation technique using independent component analysis under the assumption that medical image consists of some statistically independent parts leads to a method that allows for more accurate segmentation of bones from CT data. The result of automatic segmentation using independent component analysis with square test data was evaluated using probability of error(PE) and ultimate measurement accuracy(UMA) value. It was also compared to a general segmentation method using threshold based on sensitivity(True Positive Rate), specificity(False Positive Rate) and mislabelling rate. The evaluation result was done statistical Paired-t test. Most of the results show that the automatic segmentation using independent component analysis has better result than general segmentation using threshold.

• Journal of Communications and Networks
• /
• 제5권1호
• /
• pp.33-42
• /
• 2003

In this paper a two-stage maximum-likelihood (ML) detection structure for group detection in DS/CDMA systems is presented. The first stage of the receiver is a linear filter, aimed at suppressing the effect of the unwanted (i.e., out-of-grout) users' signals, while the second stage is a non-linear block, implementing a ML detection rule on the set of desired users signals. As to the linear stage, we consider both the decorrelating and the minimum mean square error approaches. Interestingly, the proposed detection structure turns out to be a generalization of Varanasi's group detector, to which it reduces when the system is synchronous, the signatures are linerly independent and the first stage of the receiver is a decorrelator. The issue of blind adaptive receiver implementation is also considered, and implementations of the proposed receiver based on the LMS algorithm, the RLS algorithm and subspace-tracking algorithms are presented. These adaptive receivers do not rely on any knowledge on the out-of group users' signals, and are thus particularly suited for rejection of out-of-cell interference in the base station. Simulation results confirm that the proposed structure achieves very satisfactory performance in comparison with previously derived receivers, as well as that the proposed blind adaptive algorithms achieve satisfactory performance.