• Journal of Multimedia Information System
• /
• 제8권2호
• /
• pp.101-110
• /
• 2021

In computer vision, single-image super resolution has been an area of research for a significant period. Traditional techniques involve interpolation-based methods such as Nearest-neighbor, Bilinear, and Bicubic for image restoration. Although implementations of convolutional neural networks have provided outstanding results in recent years, efficiency and single model multi-scalability have been its challenges. Furthermore, previous works haven't placed enough emphasis on real-number scalability. Interpolation-based techniques, however, have no limit in terms of scalability as they are able to upscale images to any desired size. In this paper, we propose a convolutional neural network possessing the advantages of the interpolation-based techniques, which is also efficient, deeming it suitable in practical implementations. It consists of convolutional layers applied on the low-resolution space, post-up-sampling along the end hidden layers, and additional layers on high-resolution space. Up-sampling is applied on a multiple channeled feature map via bicubic interpolation using a single model. Experiments on architectural structure, layer reduction, and real-number scale training are executed with results proving efficient amongst multi-scale learning (including scale multi-path-learning) based models.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2003년도 ICCAS
• /
• pp.150-155
• /
• 2003

We discussed a quantitative structure-activity relationships (QSAR) technique on incomplete data set. We proposed a new solver that used 2 kinds of multi-layer neural networks. One is to compensate the defect data, and another is to evaluate the QSAR. The solver can predict the defects in model QSAR data. By using them, we get very high precision QSAR. It is 5-10 times higher than that of a traditional method. However, in case of anti-cancer Carboquone, the prediction is not so complete. It was about O(3) wrong than the model calculation. The predicted values would have rather large error. It is caused by noisy observations of Carboquone. However, if we used the uncertain predictions, new data are included in QSAR. If not, they were omitted. The effect would not be little. Therefore, we evaluated the QSAR. The results are contrary to the expectation, are not so wrong. We believe that the wrong effect is suppressed by including information of new data.

• 조명전기설비학회논문지
• /
• 제24권1호
• /
• pp.63-70
• /
• 2010

본 논문에서는 원자력발전소 증기세관 크기 예측을 위한 신경회로망 구조에 대해서 연구한다. 와류탐상 시험(ECT) 신호로부터 특징을 추출한 후, 결함크기 예측을 위해서 다층퍼셉트론 신경회로망을 이용한다. 결함크기 예측성능을 최대화하기 위해서는 신경회로망의 구조, 특히 은닉층 내의 뉴런의 개수를 신중히 결정하여야 한다. 본 논문에서는, 결함크기 예측을 위한 은닉층 내의 뉴런의 개수를 교차검증을 이용하여 매우 효과적으로 결정할 수 있음을 보인다.

• Structural Engineering and Mechanics
• /
• 제51권2호
• /
• pp.299-313
• /
• 2014

In the first part of this paper, the influences of some of crack parameters on natural frequencies of a cracked cantilever Functionally Graded Beam (FGB) are studied. A cantilever beam is modeled using Finite Element Method (FEM) and its natural frequencies are obtained for different conditions of cracks. Then effect of variation of depth and location of cracks on natural frequencies of FGB with single and multiple cracks are investigated. In the second part, two Multi-Layer Feed Forward (MLFF) Artificial Neural Networks (ANNs) are designed for prediction of FGB's Cracks' location and depth. Particle Swarm Optimization (PSO) and Back-Error Propagation (BEP) algorithms are applied for training ANNs. The accuracy of two training methods' results are investigated.

• 대한전기학회논문지:시스템및제어부문D
• /
• 제49권3호
• /
• pp.145-156
• /
• 2000

In this paper, we propose a new methodology which includes the optimal design procedure of Polynomial Neural Networks(PNN) structure for model identification of complex and nonlinear system. The proposed PNN algorithm is based on GMDA(Group Method of Data handling) method and its structure is similar to Neural Networks. But the structure of PNN is not fixed like in conventional Neural Networks and can be generated. The each node of PNN structure uses several types of high-order polynomial such as linear, quadratic and cubic, and is connected as various kinds of multi-variable inputs. In other words, the PNN uses high-order polynomial as extended type besides quadratic polynomial used in GMDH, and the number of input of its node in each layer depends on that of variables used in the polynomial. The design procedure to obtain an optimal model structure utilizing PNN algorithm is shown in each stage. The study is illustrated with the aid of pH neutralization process data besides representative time series data for gas furnace process used widely for performance comparison, and shows that the proposed PNN algorithm can produce the model with higher accuracy than previous other works. And performance index related to approximation and prediction capabilities of model is evaluated and also discussed.

• ETRI Journal
• /
• 제39권6호
• /
• pp.841-850
• /
• 2017

We investigate neural network image reconstruction for magnetic particle imaging. The network performance strongly depends on the convolution effects of the spectrum input data. The larger convolution effect appearing at a relatively smaller nanoparticle size obstructs the network training. The trained single-layer network reveals the weighting matrix consisting of a basis vector in the form of Chebyshev polynomials of the second kind. The weighting matrix corresponds to an inverse system matrix, where an incoherency of basis vectors due to low convolution effects, as well as a nonlinear activation function, plays a key role in retrieving the matrix elements. Test images are well reconstructed through trained networks having an inverse kernel matrix. We also confirm that a multi-layer network with one hidden layer improves the performance. Based on the results, a neural network architecture overcoming the low incoherence of the inverse kernel through the classification property is expected to become a better tool for image reconstruction.

• 한국환경보건학회지
• /
• 제28권3호
• /
• pp.19-25
• /
• 2002

In this paper, the water quality forecast was performed on the BOD of the Chungju Dam using the ARIMA model, which is a nonlinear statistics model, and the artificial neural network model. The monthly data of water quality were collected from 1991 to 2000. The most appropriate ARIMA model for Chungju dam was found to be the multiplicative seasonal ARIMA(1,0,1)(1,0,1)$_{12}$, model. While the artificial neural network model, which is used relatively often in recent days, forecasts new data by the strength of a learned matrix like human neurons. The BOD values were forecasted using the back-propagation algorithm of multi-layer perceptrons in this paper. Artificial neural network model was com- posed of two hidden layers and the node number of each hidden layer was designed fifteen. It was demonstrated that the ARIMA model was more appropriate in terms of changes around the overall average, but the artificial neural net-work model was more appropriate in terms of reflecting the minimum and the maximum values.s.

• 한국정보통신학회논문지
• /
• 제12권4호
• /
• pp.685-690
• /
• 2008

수치해석적으로 열교환기의 열성능 평가를 하기 위하여는 온도, 압력, 비체적, 엔탈피, 엔트로피 등의 열역학적 성질들의 수치값을 필요로 한다. 그러나 열역학적 성질들 사이의 관계를 나타내는 증기표나 선도를 수치 해석에 직접적으로 이용할 수는 없기 때문에 모델링하여야 한다. 본 연구에서는 2차 스플라인 보간법과 비교함으로써, 물의 과열증기 모델링에 신경회로망의 적용 가능성을 검토하였다. 신경회로망은 온도와 압력 2개의 노드로 구성된 입력층, 각각 15개와 25개의 노드로 구성된 2개의 은닉층, 비체적, 엔탈피, 엔트로피 등 3개의 노드로 구성된 출력층으로 이루어 진다. 스플라인 보간법에는 2차 다항식을 사용하였다. 소구간으로 구성된 스플라인 보간법과 비교하여 신경회로망은 훨씬 더 많은 데이터에 대하여 작은 백분율 오차를 보여 주었으며, 이 결과로부터 신경회로망이 과열증기의 열역학적 성질들을 모델링하는데 아주 강력한 방법이 될 수 있음을 확인하였다.

• 대한원격탐사학회:학술대회논문집
• /
• 대한원격탐사학회 2005년도 Proceedings of ISRS 2005
• /
• pp.34-37
• /
• 2005

An efficient and robust neural network-based scheme is introduced in this paper to perform automatic cloud detection. Unlike many existing cloud detection schemes which use thresholding and statistical methods, we used the artificial neural network methods, the multi-layer perceptrons (MLP) with back-propagation algorithm and radial basis function (RBF) networks for cloud detection from Geostationary satellite images. We have used a simple scene (a mixed scene containing only cloud and clear sky). The main results show that the neural networks are able to handle complex atmospheric and meteorological phenomena. The experimental results show that two methods performed well, obtaining a classification accuracy reaching over 90 percent. Moreover, the RBF model is the most effective method for the cloud classification.

• 대한전기학회논문지:시스템및제어부문D
• /
• 제53권5호
• /
• pp.353-360
• /
• 2004

This paper considers reinforcement learning control with the self-organizing map. Reinforcement learning uses the observable states of objective system and signals from interaction of the system and environment as input data. For fast learning in neural network training, it is necessary to reduce learning data. In this paper, we use the self-organizing map to partition the observable states. Partitioning states reduces the number of learning data which is used for training neural networks. And neural dynamic programming design method is used for the controller. For evaluating the designed reinforcement learning controller, an inverted pendulum on the cart system is simulated. The designed controller is composed of serial connection of self-organizing map and two Multi-layer Feed-Forward Neural Networks.