• 대한전자공학회논문지SP
• /
• 제48권1호
• /
• pp.54-63
• /
• 2011

본 논문은 저해상도 입력 영상을 고해상도 영상으로 복원하는 고속 학습기반 보간 기법을 제안한다. 일반적인 학습기반 초고해상도 기법은 여러 종류의 저해상도 영상과 고해상도 영상의 상관성을 통해 고주파 정보를 사전에 학습하고, 합성 단계에서 학습한 정보를 이용해 임의의 저해상도 영상을 고해상도 영상으로 복원한다. 이런 기존 학습기반 초 고해상도 기법은 방대한 양의 학습된 정보를 메모리에 저장해야만 하는 단점이 있을 뿐만 아니라 이차원 블록 단위 정합 과정을 거쳐야 하기 때문에 상당한 연산량이 요구된다. 이러한 문제점을 보완하기 위해 본 논문은 일차원 패치 단위 학습을 통해 학습 정보 저장용 메모리 크기 및 연산량을 크게 줄이는 기법을 제안한다. 실험 결과에 따르면, 제안한 기법은 전통적인 bicubic 보간 기법보다 평균 0.7dB 정도 높은 PSNR을 보이며, SSIM도 평균 0.01이상 향상되는 결과를 보인다.

• Nuclear Engineering and Technology
• /
• 제55권10호
• /
• pp.3515-3524
• /
• 2023

A reinforcement learning framework is proposed for the control problem of outlet steam pressure of the once-through steam generator(OTSG) in this paper. The double-layer controller using Proximal Policy Optimization(PPO) algorithm is applied in the control structure of the OTSG. The PPO algorithm can train the neural networks continuously according to the process of interaction with the environment and then the trained controller can realize better control for the OTSG. Meanwhile, reinforcement learning has the characteristic of difficult application in real-world objects, this paper proposes an innovative pretraining method to solve this problem. The difficulty in the application of reinforcement learning lies in training. The optimal strategy of each step is summed up through trial and error, and the training cost is very high. In this paper, the LSTM model is adopted as the training environment for pretraining, which saves training time and improves efficiency. The experimental results show that this method can realize the self-adjustment of control parameters under various working conditions, and the control effect has the advantages of small overshoot, fast stabilization speed, and strong adaptive ability.

• 한국지능시스템학회:학술대회논문집
• /
• 한국퍼지및지능시스템학회 1998년도 The Third Asian Fuzzy Systems Symposium
• /
• pp.134-138
• /
• 1998

Unification of Kohone SOM(Self-Organizing Maps) neural network with the branch-and-bound algorithm is presented for clustering large set of patterns. The branch-and-bound search technique is employed for designing coarse neural network learning paradaim. Those unification can be use for clustering or calssfication of large patterns. For classfication purposes further usefulness is possible, since only two clusters exists in the SOM neural network of each nodes. The result of experiments show the fast learning time, the fast recognition time and the compactness of clustering.

• 한국멀티미디어학회논문지
• /
• 제23권6호
• /
• pp.756-764
• /
• 2020

A typical algorithm used for image analysis is the Convolutional Neural Network(CNN). R-CNN, Fast R-CNN, Faster R-CNN, etc. have been studied to improve the performance of the CNN, but they essentially require large amounts of data and high algorithmic complexity., making them inappropriate for small and medium-sized services. Therefore, in this paper, the image label prediction algorithm based on CNN with collaborative layer with low complexity, high accuracy, and small amount of data was proposed. The proposed algorithm was designed to replace the part of the neural network that is performed to predict the final label in the existing deep learning algorithm by implementing collaborative filtering as a layer. It is expected that the proposed algorithm can contribute greatly to small and medium-sized content services that is unsuitable to apply the existing deep learning algorithm with high complexity and high server cost.

• 한국지능시스템학회논문지
• /
• 제5권3호
• /
• pp.83-91
• /
• 1995

본 연구에서는 종래에 학습법으로 널리 이용되고 있는 역전파학습법의 문제점으로 지적되어 온 학습에 많은 시간이 걸리는 점과 국소적 최적해에 해가 수렴하여 오차가 충분히 작게 되지 않는 등의 문제점을 해결하기 위해, Hu에 의해 고안된 random tabu 탐색법을 이용하여 신경회로망의 연결강도를 최적화하는 학습알고리즘을 새로이 제안하였다. 그리고 이 방법을 배타적 논리합 문제에 적용하여 기존의 역전파학습법과 학습상수 $, $에 tabu탐색법을 이용한 결과와 비교 검토하여 본 방법이 국소적 최적해에 수렴하지 않고 수렴정도를 개선할 수 있음을 확인하였다.

• 전자공학회논문지B
• /
• 제31B권8호
• /
• pp.211-218
• /
• 1994

In this paper we develope a hybrid learning rule to improve the robustness of multi-layer Perceptions. In most neural networks the activation of a neuron is deternined by a nonlinear transformation of the weighted sum of inputs to the neurons. Investigating the behaviour of activations of hidden layer neurons a new learning algorithm is developed for improved robustness for multi-layer Perceptrons. Unlike other methods which reduce the network complexity by putting restrictions on synaptic weights our method based on error-backpropagation increases the complexity of the underlying proplem by imposing it saturation requirement on hidden layer neurons. We also found that the additional gradient-descent term for the requirement corresponds to the Hebbian rule and our algorithm incorporates the Hebbian learning rule into the error back-propagation rule. Computer simulation demonstrates fast learning convergence as well as improved robustness for classification and hetero-association of patterns.

• 반도체디스플레이기술학회지
• /
• 제17권1호
• /
• pp.88-92
• /
• 2018

Currently drone industry has become one of the fast growing markets and the technology for unmanned aerial vehicles are expected to continue to develop at a rapid rate. Especially small unmanned aerial vehicle systems have been designed and utilized for the various field with their own specific purposes. In these fields the path planning problem to find the shortest path between two oriented points is important. In this paper we introduce a path planning strategy for an autonomous flight of unmanned aerial vehicles through reinforcement learning with self-positioning technique. We perform Q-learning algorithm, a kind of reinforcement learning algorithm. At the same time, multi sensors of acceleraion sensor, gyro sensor, and magnetic are used to estimate the position. For the functional evaluation, the proposed method was simulated with virtual UAV environment and visualized the results. The flight history was based on a PX4 based drones system equipped with a smartphone.

• 한국산학기술학회논문지
• /
• 제19권3호
• /
• pp.693-698
• /
• 2018

본 논문은 최근 물체탐지 분야에서 실시간 물체 탐지 알고리즘으로 주목을 받고 있는 YOLOv2(You Only Look Once) 알고리즘을 이용하여 밀집 영역에 주차되어 있는 자동차 탐지 방법을 제안한다. YOLO의 컨볼루션 네트워크는 전체 이미지에서 한 번의 평가를 통해서 직접적으로 경계박스들을 예측하고 각 클래스의 확률을 계산하고 물체 탐지 과정이 단일 네트워크이기 때문에 탐지 성능이 최적화 되며 빠르다는 장점을 가지고 있다. 기존의 슬라이딩 윈도우 접근법과 R-CNN 계열의 탐지 방법은 region proposal 방법을 사용하여 이미지 안에 가능성이 많은 경계박스를 생성하고 각 요소들을 따로 학습하기 때문에 최적화 및 실시간 적용에 어려움을 가지고 있다. 제안하는 연구는 YOLOv2 알고리즘을 적용하여 기존의 알고리즘이 가지고 있는 물체 탐지의 실시간 처리 문제점을 해결하여 실시간으로 지상에 있는 자동차를 탐지하는 방법을 제안한다. 제안하는 연구 방법의 실험을 위하여 오픈소스로 제공되는 Darknet을 사용하였으며 GTX-1080ti 4개를 탑재한 Deep learning 서버를 이용하여 실험하였다. 실험결과 YOLO를 활용한 자동차 탐지 방법은 기존의 알고리즘 보다 물체탐지에 대한 오버헤드를 감소 할 수 있었으며 실시간으로 지상에 존재하는 자동차를 탐지할 수 있었다.

• 한국산업융합학회 논문집
• /
• 제6권1호
• /
• pp.23-29
• /
• 2003

This paper proposes an efficient feature extraction of the images by using independent component analysis(ICA) based on neural networks of the hybrid learning algorithm. The proposed learning algorithm is the fixed point(FP) algorithm based on Newton method and moment. The Newton method, which uses to the tangent line for estimating the root of function, is applied for fast updating the inverse mixing matrix. The moment is also applied for getting the better speed-up by restraining an oscillation due to compute the tangent line. The proposed algorithm has been applied to the 10,000 image patches of $12{\times}12$-pixel that are extracted from 13 natural images. The 144 features of $12{\times}12$-pixel and the 160 features of $16{\times}16$-pixel have been extracted from all patches, respectively. The simulation results show that the extracted features have a localized characteristics being included in the images in space, as well as in frequency and orientation. And the proposed algorithm has better performances of the learning speed than those using the conventional FP algorithm based on Newton method.

• ETRI Journal
• /
• 제28권1호
• /
• pp.59-66
• /
• 2006

Learning behaviors of a radial basis function network (RBFN) using a singular value decomposition (SVD) and stochastic gradient (SG) algorithm, together named RBF-SVD-SG, for odor sensing systems are analyzed, and a fast training method is proposed. RBF input data is from a conducting polymer sensor array. It is revealed in this paper that the SG algorithm for the fine-tuning of centers and widths still shows ill-behaving learning results when a sufficiently small convergence coefficient is not used. Since the tuning of centers in RBFN plays a dominant role in the performance of RBFN odor sensing systems, our analysis is focused on the center-gradient variance of the RBFN-SVD-SG algorithm. We found analytically that the steadystate weight fluctuation and large values of a convergence coefficient can lead to an increase in variance of the center-gradient estimate. Based on this analysis, we propose to use the least mean square algorithm instead of SVD in adjusting the weight for stable steady-state weight behavior. Experimental results of the proposed algorithm have shown faster learning speed and better classification performance.