• 자동차안전학회지
• /
• 제11권2호
• /
• pp.29-34
• /
• 2019

This paper presents a new method for construction of a static obstacle map. A static obstacle is important since it is utilized to path planning and decision. Several established approaches generate static obstacle map by grid method and counting algorithm. However, these approaches are occasionally ineffective since the density of LiDAR layer is low. Our approach solved this problem by applying probability theory. First, we converted all LiDAR point to Gaussian distribution to considers an uncertainty of LiDAR point. This Gaussian distribution represents likelihood of obstacle. Second, we modeled dynamic transition of a static obstacle map by adopting the Hidden Markov Model. Due to the dynamic characteristics of the vehicle in relation to the conditions of the next stage only, a more accurate map of the obstacles can be obtained using the Hidden Markov Model. Experimental data obtained from test driving demonstrates that our approach is suitable for mapping static obstacles. In addition, this result shows that our algorithm has an advantage in estimating not only static obstacles but also dynamic characteristics of moving target such as driving vehicles.

• 전기학회논문지
• /
• 제67권1호
• /
• pp.114-123
• /
• 2018

This paper concerns a design and learning method of softmax function neural networks based on K-means clustering. The partial discharge data Information is preliminarily processed through simulation using an Epoxy Mica Coupling sensor and an internal Phase Resolved Partial Discharge Analysis algorithm. The obtained information is processed according to the characteristics of the pattern using a Motor Insulation Monitoring System program. At this time, the processed data are total 4 types that void discharge, corona discharge, surface discharge and slot discharge. The partial discharge data with high dimensional input variables are secondarily processed by principal component analysis method and reduced with keeping the characteristics of pattern as low dimensional input variables. And therefore, the pattern classifier processing speed exhibits improved effects. In addition, in the process of extracting the partial discharge data through the MIMS program, the magnitude of amplitude is divided into the maximum value and the average value, and two pattern characteristics are set and compared and analyzed. In the first half of the proposed partial discharge pattern classifier, the input and hidden layers are classified by using the K-means clustering method and the output of the hidden layer is obtained. In the latter part, the cross entropy error function is used for parameter learning between the hidden layer and the output layer. The final output layer is output as a normalized probability value between 0 and 1 using the softmax function. The advantage of using the softmax function is that it allows access and application of multiple class problems and stochastic interpretation. First of all, there is an advantage that one output value affects the remaining output value and its accompanying learning is accelerated. Also, to solve the overfitting problem, L2-normalization is applied. To prove the superiority of the proposed pattern classifier, we compare and analyze the classification rate with conventional radial basis function neural networks.

• Journal of Electrical Engineering and Technology
• /
• 제12권1호
• /
• pp.373-377
• /
• 2017

This paper presents a voice-activity detection (VAD) method for sound sequences with various SNRs. For real-time VAD applications, it is inadequate to employ a post-processing for the removal of burst clippings from the VAD output decision. To tackle this problem, building on the bi-level hidden Markov model, for which a state layer is inserted into a typical hidden Markov model (HMM), we formulated a robust method for VAD not requiring any additional post-processing. In the method, a forward-inference-ratio test was devised to detect the speech endpoints and Mel-frequency cepstral coefficients (MFCC) were used as the features. Our experiment results show that, regarding different SNRs, the performance of the proposed approach is more outstanding than those of the conventional methods.

• 전자공학회논문지B
• /
• 제31B권5호
• /
• pp.141-148
• /
• 1994

Both the neural network classifier utilizing multi-layer perceptron and the linear tree classifier composed of hierarchically structured linear discriminating functions can form arbitrarily complex decision boundaries in the feature space and have very similar decision making processes. In this paper, a new method for automatically choosing the number of neurons in the hidden layers and for initalzing the connection weights between the layres and its supporting theory are presented by mapping the sequential structure of the linear tree classifier to the parallel structure of the neural networks having one or two hidden layers. Experimental results on the real data obtained from the military ship images show that this method is effective, and that three exists no siginificant difference in the classification acuracy of both classifiers.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2001년도 ICCAS
• /
• pp.163.1-163
• /
• 2001

In this paper, we use time-frequency localization analysis method to analize the target function and the area of the target space. When we analize the function with the time and frequency axis simultaneously, the characteristic of the function is shown more precisely and the area is covered by a certain block. After we analize the target function in the time-frequency space, we can decide the activation functions and compose the hidden layer of the RBFN by choosing the radial basis function which can represent the characteristic of the target function, RBFN made by this method, designs the good structure proper to the target problem because we can decide the number of hidden node first.

• 전자공학회논문지B
• /
• 제32B권5호
• /
• pp.756-764
• /
• 1995

In this paper, we proposed MLP(MultiLayer Perceptron) based HMM's(Hidden Markov Models) with bounded state duration for isolated word recognition. The minimum and maximum state duration for each state of a HMM are estimated during the training phase and used as parameters of constraining state transition in a recognition phase. The procedure for estimating these parameters and the recognition algorithm using the proposed HMM's are also described. Speaker independent isolated word recognition experiments using a vocabulary of 10 city names and 11 digits indicate that recognition rate can be improved by adjusting the minimum state durations.

• 한국지능시스템학회논문지
• /
• 제9권6호
• /
• pp.571-576
• /
• 1999

We consider a class of auto-associative memories namely N-Cubes (Neural-network Cubes) in which 2-D gray-level images and hidden sinusoidal 1-D wavelets are stored in cubical memories. First we develop a learning procedure based upon the least-squares algorithm, Therefore each 2-D training image is mapped into the associated 1-D waveform in the training phase. Second we show how the recall procedure minimizes errors among the orthogonal basis functions in the hidden layer. As a 2-D images ould be retrieved in the recall phase. Simulation results confirm the efficiency and the noise-free properties of N-Cubes.

• Journal of Welding and Joining
• /
• 제31권3호
• /
• pp.44-48
• /
• 2013

Artificial neural network (ANN) model is applied to predict arc welding process window for automotive steel plate. Target weldment was various automotive steel plate combination with lap fillet joint. The accuracy of prediction was evaluated through comparison experimental result to ANN simulation. The effect of ANN variables on the accuracy is investigated such as number of hidden layers, perceptrons and transfer function type. A static back propagation model is established and tested. The result shows comparatively accurate predictability of the suggested ANN model. However, it restricts to use nonlinear transfer function instead of linear type and suggests only one single hidden layer rather than multiple ones to get better accuracy. In addition to this, obvious fact is affirmed again that the more perceptrons guarantee the better accuracy under the precondition that there are enough experimental database to train the neural network.

• 한국건축시공학회:학술대회논문집
• /
• 한국건축시공학회 2019년도 춘계 학술논문 발표대회
• /
• pp.30-31
• /
• 2019

Carbonation of the root concrete reduces the durability of the reinforced concrete, and it is important to check the carbonation resistance of the concrete to ensure the durability of the reinforced concrete structure. In this study, a basic study on the prediction of carbonation progress was conducted by considering the mixing conditions of concrete using deep learning algorithm during the theory of artificial neural network theory. The data used in the experiment used values that converted the carbonation velocity coefficient obtained from the mixing conditions of concrete and the accelerated carbonation experiment into the actual environment. The analysis shows that the error rate of the deep learning model according to the Hidden Layer is the best for the model using five layers, and based on the five Hidden layers, we want to verify the predicted performance of the carbonation speed coefficient of the carbonation test specimen in which the exposure experiment took place in the real environment.

• 한국지능시스템학회:학술대회논문집
• /
• 한국퍼지및지능시스템학회 2002년도 추계학술대회 및 정기총회
• /
• pp.337-340
• /
• 2002

본 논문에서는 휠체어 시스템에 화자 독립 고립단어 인식을 위한 임베디드 시스템 설계에 관한 내용을 서술한다. 실제 환경에서는 잡음이 포함되어 있어 인식률을 저하시키므로, 잡음을 제거하는 방식 중 가장 간단한 방식인 스펙트럼 차감법(Spectral subtraction method)을 사용하여 잡음을 제거했다 전처리 단계에서는 12차 LPC&Cepstrum 방식을 사용했고, 인식 알고리즘은 DHMM (Discrete Hidden Markov Model)을 전반부 인식기로 사용했다. 이 알고리즘을 적용하기 위해서는 데이터 간소화를 위해 벡터양자화(Vector Quantization) 처리가 전제되어야한다 또한 인식알고리즘은 인식률을 향상을 위해 후처리 인식기로 신경망(MLP:Multi-layer Perceptron)을 통해서 인식률을 향상시켰다 화자 독립 시스템에 맞는 인식 단어의 구성은 총 7개단어로 남녀 총 25명 목소리로 구성하였다. 그리고 하드웨어 구성은 32-bits floating point 방식인 TMS320C32를 적용했고, 메모리 부분은 4Mbyte로 설계를 했으며, 메인보드의 설계는 현재 완성 단계에 있다.