• The Journal of the Acoustical Society of Korea
• /
• v.13 no.2
• /
• pp.50-59
• /
• 1994

This paper presents a speech recognition method utilizing third-order cumulants as a feature vector and a neural network for recognition. The use of higher-order cumulants provides desirable uncoupling between the gaussian noise and speech, which enables us to estimate the coefficients of AR model without bias. Unlike the conventional method using second-order statistics, the proposed one exhibits low bias even in SNR as low as 0 dB at the expense of higher variance. It is confirmed through computer simulation that recognition rate of korean single-vowels with the cumulant-based method is much higher than the results with the conventional method even in low SNR.

• Geomechanics and Engineering
• /
• v.28 no.3
• /
• pp.313-324
• /
• 2022

Rock properties are important in the design of mines and civil engineering excavations to prevent the imminent failure of slopes and collapse of underground excavations. However, the time, cost, and expertise required to perform experiments to determine those properties are high. Therefore, empirical models have been developed for estimating the mechanical properties of rock that are difficult to determine experimentally from properties that are less difficult to measure. However, the inherent variability in rock properties makes the accurate performance of the empirical models unrealistic and therefore necessitate the use of soft computing models. In this study, Gaussian process regression (GPR), artificial neural network (ANN) and response surface method (RSM) have been proposed to predict the static and dynamic rock properties from the P-wave and rock density. The outcome of the study showed that GPR produced more accurate results than the ANN and RSM models. GPR gave the correlation coefficient of above 99% for all the three properties predicted and RMSE of less than 5. The detailed sensitivity analysis is also conducted using the RSM and the P-wave velocity is found to be the most influencing parameter in the rock mechanical properties predictions. The proposed models can give reasonable predictions of important mechanical properties of sedimentary rock.

• Wind and Structures
• /
• v.38 no.1
• /
• pp.75-91
• /
• 2024

Reliable wind signal reconstruction can be beneficial to the operational safety of long-span bridges. Non-Gaussian characteristics of wind signals make the reconstruction process challenging. In this paper, non-Gaussian wind signals are converted into a combined prediction of two kinds of features, actual wind speeds and wind angles of attack. First, two decomposition techniques, empirical mode decomposition (EMD) and variational mode decomposition (VMD), are introduced to decompose wind signals into intrinsic mode functions (IMFs) to reduce the randomness of wind signals. Their principles and applicability are also discussed. Then, four artificial intelligence (AI) algorithms are utilized for wind signal reconstruction by combining the particle swarm optimization (PSO) algorithm with back propagation neural network (BPNN), support vector regression (SVR), long short-term memory (LSTM) and bidirectional long short-term memory (Bi-LSTM), respectively. Measured wind signals from a bridge site in a deep-cutting gorge are taken as experimental subjects. The results showed that the reconstruction error of high-frequency components of EMD is too large. On the contrary, VMD fully extracts the multiscale rules of the signal, reduces the component complexity. The combination of VMD-PSO-Bi-LSTM is demonstrated to be the most effective among all hybrid models.

• Proceedings of the KIEE Conference
• /
• 2003.07d
• /
• pp.2519-2521
• /
• 2003

In this paper, we propose Self-Organizing Fuzzy Polynomial Neural Networks(SOFPNN) architecture for optimal model identification and discuss a comprehensive design methodology supporting its development. It is shown that this network exhibits a dynamic structure as the number of its layers as well as the number of nodes in each layer of the SOFPNN are not predetermined (as this is the case in a popular topology of a multilayer perceptron). As the form of the conclusion part of the rules, especially the regression polynomial uses several types of high-order polynomials such as linear, quadratic, and modified quadratic. As the premise part of the rules, both triangular and Gaussian-like membership function are studied and the number of the premise input variables used in the rules depends on that of the inputs of its node in each layer. We introduce two kinds of SOFPNN architectures, that is, the basic and modified one with both the generic and the advanced type. The superiority and effectiveness of the proposed SOFPNN architecture is demonstrated through nonlinear function numerical example.

• Smart Structures and Systems
• /
• v.23 no.1
• /
• pp.1-14
• /
• 2019

Control algorithms are the most important aspects in successful control of structures against earthquakes. In recent years, intelligent control methods rather than classical control methods have been more considered by researchers, due to some specific capabilities such as handling nonlinear and complex systems, adaptability, and robustness to errors and uncertainties. However, due to lack of learning ability of fuzzy controller, it is used in combination with a genetic algorithm, which in turn suffers from some problems like premature convergence around an incorrect target. Therefore in this research, the introduction and design of the Fuzzy Cooperative Coevolution (Fuzzy CoCo) controller and Adaptive Neural-Fuzzy Inference System (ANFIS) have been innovatively presented for semi-active seismic control. In this research, in order to improve the seismic behavior of structures, a semi-active control of building using Magneto Rheological (MR) damper is proposed to determine input voltage of Magneto Rheological (MR) dampers using ANFIS and Fuzzy CoCo. Genetic Algorithm (GA) is used to optimize the performance of controllers. In this paper, the design of controllers is based on the reduction of the Park-Ang damage index. In order to assess the effectiveness of the designed control system, its function is numerically studied on a 9-story benchmark building, and is compared to those of a Wavelet Neural Network (WNN), fuzzy logic controller optimized by genetic algorithm (GAFLC), Linear Quadratic Gaussian (LQG) and Clipped Optimal Control (COC) systems in terms of seismic performance. The results showed desirable performance of the ANFIS and Fuzzy CoCo controllers in considerably reducing the structure responses under different earthquakes; for instance ANFIS and Fuzzy CoCo controllers showed respectively 38 and 46% reductions in peak inter-story drift ($J_1$) compared to the LQG controller; 30 and 39% reductions in $J_1$ compared to the COC controller and 3 and 16% reductions in $J_1$ compared to the GAFLC controller. When compared to other controllers, one can conclude that Fuzzy CoCo controller performs better.

• Proceedings of the KIEE Conference
• /
• 2000.11d
• /
• pp.800-802
• /
• 2000

In this paper, we propose the Self-Organizing Networks(SON) based on competitive Fuzzy Polynomial Neuron(FPN) for the optimal design of nonlinear process system. The SON architectures consist of layers with activation nodes based on fuzzy inference rules. Here each activation node is presented as FPN which includes either the simplified or regression Polynomial fuzzy inference rules. The proposed SON is a network resulting from the fusion of the Polynomial Neural Networks(PNN) and a fuzzy inference system. The conclusion part of the rules, especially the regression polynomial uses several types of high-order polynomials such as liner, quadratic and modified quadratic. As the premise part of the rules, both triangular and Gaussian-like membership functions are studied. Chaotic time series data used to evaluate the performance of our proposed model.

• Proceedings of the Korean Information Science Society Conference
• /
• 2002.10d
• /
• pp.541-543
• /
• 2002

본 논문에서는 HOLA(고차국소자동상관계수)를 이용한 특징추출과 BP(Backpropagation Network) 알고리즘을 이용하여 얼굴을 인식하는 방법을 제안한다. 이를 위해 동일한 환경, 즉 일정한 조도 하에서 카메라로부터 동일거리에 있는 영상을 256$\times$256 크기의 그레이 스케일(Gray Scale)로 취득하여 영상내의 잡음을 가우시안(Gaussian) 필터를 이용하여 제거한다. 차영상을 이용하여 얼굴영역을 분리한 후 얼굴영역의 특징벡터를 구하기 위하여 HOLA(고차 국소 자동 상관함수)를 사용한다. 계산된 특징벡터는 BP 신경망의 학습을 통하여 얼굴인식을 위한 데이터로 사용된다. 시뮬레이션을 통해 제안된 알고리즘에 의한 인식률향상과 속도 향상을 입증한다.

• Annual Conference on Human and Language Technology
• /
• 2015.10a
• /
• pp.231-233
• /
• 2015

본 논문에서는 심층 신경망 기반의 내장형 음성 인식 시스템에서 음성 인식 속도를 개선하기 위한 최적화 방법에 대해 논한다. 심층 신경망 기반의 음성 인식은 기존의 Gaussian Mixture Model (GMM) 기반에 비해 좋은 인식 성능을 보이지만 높은 연산량으로 인해 리소스가 제약된 내장형 단말기에 적용하기에는 어려움이 따른다. 따라서, 본 연구에서는 심층 신경망의 계산량 문제를 해결하고자 ARM 코어에 내장된 병렬 명령어를 사용한 최적화 기법과 특이값 분해를 통해 심층 신경망 매트릭스 연산량 감소 방안에 대해 제안한다.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.21 no.12
• /
• pp.2291-2297
• /
• 2017

This paper describe to extract speech measure algorithm for evaluating a speech database, and presents generating method of a speech quality measure using DNN(Deep Neural Network). In our previous study, to produce an effective speech quality measure, we propose a combination of various speech measures which are highly correlated with WER(Word Error Rate). The new combination of various types of speech quality measures in this study is more effective to predict the speech recognition performance compared to each speech measure alone. In this paper, we describe the method of extracting measure using DNN, and we change one of the combined measure from GMM(Gaussican Mixture Model) score used in the previous study to DNN score. The combination with DNN score shows a higher correlation with WER compared to the combination with GMM score.

• Transactions on Control, Automation and Systems Engineering
• /
• v.3 no.4
• /
• pp.283-288
• /
• 2001

In this paper, we proposed a method for extracting facial characteristics of human being in an image. Given a pair of gray level sample images taken with and without human being, the face of human being is segmented from the image. Noise in the input images is removed with the help of Gaussian filters. Edge maps are found of the two input images. The binary edge differential image is obtained from the difference of the two input edge maps. A mask for face detection is made from the process of erosion followed by dilation on the resulting binary edge differential image. This mask is used to extract the human being from the two input image sequences. Features of face are extracted from the segmented image. An effective recognition system using the discrete wave let transform (DWT) is used for recognition. For extracting the facial features, such as eyebrows, eyes, nose and mouth, edge detector is applied on the segmented face image. The area of eye and the center of face are found from horizontal and vertical components of the edge map of the segmented image. other facial features are obtained from edge information of the image. The characteristic vectors are extrated from DWT of the segmented face image. These characteristic vectors are normalized between +1 and -1, and are used as input vectors for the neural network. Simulation results show recognition rate of 100% on the learned system, and about 92% on the test images.