Earticle

In this paper a cross-validation algorithm for training probabilistic neural networks (PNNs) is presented in order to be applied to automatic face identification. Actually, standard PNNs perform pretty well for small and medium sized databases but they suffer from serious problems when it comes to using them with large databases like those encountered in biometrics applications. To address this issue, we proposed in this work a new training algorithm for PNNs to reduce the hidden layer's size and avoid over-fitting at the same time. The proposed training algorithm generates networks with a smaller hidden layer which contains only representative examples in the training data set. Moreover, adding new classes or samples after training does not require retraining, which is one of the main characteristics of this solution. Results presented in this work show a great improvement both in the processing speed and generalization of the proposed classifier. This improvement is mainly caused by reducing significantly the size of the hidden layer.

최근 급속한 경제발전과 지역산업의 성장으로 인하여 많은 물류이동 발생과 함께 연약지반에 도로를 건설하는 경우가 많아지면서 연약지반 상에 축조된 구조물과 관련한 제반 문제점들이 대두되고 있다. 말뚝 기초 형식의 교대나 건축물을 연약지반에 시공할 경우 비정상적인 측방유동에 의한 변위가 기초지반에 발생하여 상부 구조물의 안정성과 사용성에 많은 문제를 야기하고 있다. 측방유동은 말뚝의 파손원인과 측방변위에 대한 상관관계 연구, 연약지반상에 설치된 말뚝의 변형과 모멘트에 대한 연구, 수치해석법을 이용한 연약지반상의 성토에 따른 측방변위 특성 및 현장계측을 통한 측방변위 특성 등 많은 연구가 수행되어지고 있으나 측방유동현상은 지금까지도 그 역학적 메커니즘이 정량적으로 파악하기 어렵고, 측방유동에 대한 합리적인 설계법이 확립되어 있지 않는 실정이다. 본 연구에서는 국내 및 일본 측방유동 발생 사례를 바탕으로 효과적이고 보다 정확한 측방유동 판정을 위하여 확률신경망모델을 구축하고 기존의 측방유동 판정식과 비교하였다. 연구결과 제안된 확률신경망 모델들의 측방유동 판정 성공률이 매우 높은 정확도를 가지며 기존 판정식에 비해 측방유동 판정에 매우 효과적임을 알 수 있다.

Recently, there have been many construction projects on soft ground with growth of industry and economy. Therefore foundation piles of abutments and(or) buildings had been suffering from a lot of stability problems of inordinary displacement due to lateral flow of soft ground. Although many researches about lateral flow have been carried out, it is still difficult to assess the mechanism of lateral flow in soft ground quantitatively. And reasonable design method for judgement of lateral flow occurrence in soft ground is not established yet. In this study, six PNN (Probabilistic Neural Network) models were developed according to input variables and database compiled from Korea and Japan for the judgment of lateral flow occurrence. PNN models were compared with present empirical methods. It was found that the developed PNN models can give more precise and reliable judgment of lateral flow occurrence than empirical methods.

This paper presents a different approach by employing probabilistic neural network (PNN) with level set method for segmentation of intensity inhomogeneous magnetic resonance images (MRI). The input image is classified by PNN, and initialise the image with initial zero level set contours. A cascade approach is used in which after processing image by PNN, the output of PNN is fed into Level Set Method (LSM) Algorithm. The LSM is usedto perform the segmentation and to estimate bias field.

Financial investment has become an important issue, there are many trading strategies and parameters based on quantitative models, this paper use neural network algorithm to optimization strategy parameters, various combinations of optimization strategies, as well as the evolution of new strategies to generate better returns. The empirical results show that this method has a stable and substantial return on investment, neural network can be used as an aid for decision making investments in securities.

The OCR technology for Indian documents is in emerging stage and most of these Indian OCR systems can read the documents written in only a single script. As many commercial and official documents of different states of India are tri-lingual in nature, therefore identification of script and/ or language is one of the elementary tasks for multi-script document recognition. A script recognizer simplifies the task of multi-lingual OCR by improving the accuracy and reducing the computational complexity. This script recognition may be at line, word or character level depending on interlacing of different scripts at different levels. This paper presents the effectiveness of Gabor Filter banks with kNN, SVM and PNN classifiers to identify the scripts at line level from such trilingual documents. The experiments show that Gabor features with SVM classifier achieve a recognition rate of 99.85% for trilingual documents.

This paper presents three techniques of plants classification based on their leaf shape the SVM-BDT, PNN and Fourier moment technique for solving multiclass problems. All the three techniques have been applied to a database of 1600 leaf shapes from 32 different classes, where most of the classes have 50 leaf samples of similar kind. In the proposed work three techniques are used for comparing the performance of classification of leaves. Probabilistic Neural Network with principal component analysis, Support Vector Machine utilizing Binary Decision Tree and Fourier Moment. The proposed SVM based Binary Decision Tree architecture takes advantage of both the efficient computation of the decision tree architecture and the high classification accuracy of SVMs. This can lead to a dramatic improvement in recognition speed when addressing problems with large number of classes. Classification results from all the three techniques were compared and it was observed that SVM-BDT performs better than Fourier and PNN technique.

본 논문은 입력 이미지 블록의 클래스 조건부 확률 밀도 함수의 커널 추정에 기반한 공간 영역에서의 다중 초점 이미지 융합 기법을 제안한다. 이미지 융합 문제를 시험 패턴으로부터 추정된 유사 밀도 함수에 의해 사후 클래스 확률, P(wi|Bikl)을 계산하는 분류 임무로 접근하였다. C개의 입력 이미지 Ii에 대하여 제안한 방법은 i 클래스 wi를 정의하고 베이즈 결정 원리에 기초하여 판별 함수를 최대화하는 PxQ 블록 Bikl의 집합에 의해 표현되는 결정 지도로 부터 융합 이미지 Z(k,l)를 형성한다. 출력 화질의 척도로서 RMSE 와 상호 정보량인 MI를 사용하여 제안한 기법의 성능이 평가되었다. 커널 함수의 폭 σ도 변화시키고, 다른 종류의 커널과 블록 크기를 변화시켜 가며 성능 평가를 수행하였다. 제안한 기법은 C=2 와 C=3에 대하여 시험하였고 시험 결과는 좋은 성능을 보였다.

This paper presents a spatial-level nonparametric multi-focus image fusion technique based on kernel estimates of input image blocks' underlying class-conditional probability density functions. Image fusion is approached as a classification task whose posterior class probabilities, P(wi|Bikl), are calculated with likelihood density functions that are estimated from the training patterns. For each of the C input images Ii, the proposed method defines i classes wi and forms the fused image Z(k,l) from a decision map represented by a set of P x Q blocks Bikl whose features maximize the discriminant function based on the Bayesian decision principle. Performance of the proposed technique is evaluated in terms of RMSE and Mutual Information (MI) as the output quality measures. The width of the kernel functions, σ, were made to vary, and different kernels and block sizes were applied in performance evaluation. The proposed scheme is tested with C=2 and C=3 input images and results exhibited good performance.

Recently, structures are becoming longer and higher because of the developments of new materials and construction techniques. However, such modern structures are more susceptible to excessive structural vibrations which cause deterioration in serviceability and structural safety. A modified probabilistic neural network(MPNN) approach is proposed to reduce the structural vibration. In this study, the global probability density function(PDF) of MPNN is reflected by summing the heterogeneous local PDFs automatically determined in the individual standard deviation of each variable. The proposed algorithm is applied for the vibration control of a three-story shear building model under Northridge earthquake. When the control results of the MPNN are compared with those of conventional PNN to verify the control performance, the MPNN controller proves to be more effective than PNN methods in decreasing the structural responses.

The compressive strength of concrete is commonly used criterion in producing concrete. However, the tests on the compressive strength are complicated and time-consuming. More importantly, it is too late to make improvement even if the test result does not satisfy the required strength, since the test is usually performed at the 28th day after the placement of concrete at the construction site. Therefore, accurate and realistic strength estimation before the placement of concrete is being highly required. In this study, the estimation of the compressive strength of concrete was performed by probabilistic neural network(PNN) on the basis of concrete mix proportions. The estimation performance of PNN was improved by considering the correlation between input data and targeted output value. Improved probabilistic neural network was proposed to automatically calculate the smoothing parameter in the conventional PNN by using the scheme of dynamic decay adjustment (DDA) algorithm. The conventional PNN and the PNN with DDA algorithm(IPNN) were applied to predict the compressive strength of concrete using actual test data of two concrete companies. IPNN showed better results than the conventional PNN in predicting the compressive strength of concrete.

계층형 신경망은 패턴 분류를 위해 사용되어 왔다. 이것은 주어진 교사패턴들의 학습으로 원하는 입력-출력 간의 매핑을 할 수 있기 때문이다. 신경망은 타겟ㅌ트 패턴이 입력 패턴의 카테고리에 일치할 때 타겟트 패턴을 학습하므로서 사후 확률을 근사화할 수 있다. 그리고 입력 공간을 부분 공간으로 나누어 학습 데이터들의 비율로서 만든 타겟트 벡터들로 학습한 신경망은 확률밀도 함수를 나타낼 수 있다. 본 연구에서는 역전파 학습법을 이용한 계층형 NN 과 코드북으로서 사후 확률과 확률밀도함수의 측정방법을 제안하였다. VQ 로 추정한 사후확률고 확률밀도함수를 이용하여 학습이 필요없는 RBF network 의 일종인 PNN으로 모음 인식을 수행 하였다. 인식 실험에서 PNN 의 결과는 역전파 학습법을 이용항 3층 신경망과 VQ 의 평균 인식율과 비교되었다. VQ-PNN의 인식율이 다른 것보다 우수하게 나타났다.

This paper presents the new methods of fault diagnosis through multiple alarm processing of protective relays and circuit breakers in power systems using probabilistic neural networks. In this paper, fault section detection neural network (FSDNN) for fault diagnosis is designed using the alarm information of relays or circuit breakers. In contrast to conventional methods, the proposed FSDNN determines the fault section directly and fast. To show the possibility of the proposed method, it is simulated through simulation panel for Sinyangsan substation system in KEPCO (Korea Electric Power Corporation) and the case studies show the effectiveness of the probabilistic neural network mehtod for the fault diagnosis.

Offshore structures are subjected to wave, wind, and earthquake loads. The failure of offshore structures can cause sea pollution, as well as losses of property and lives. Therefore, safety of the structure is an important issue. The reduction of the dynamic response of offshore towers, subjected wind generated random ocean waves, is a critical problem with respect to serviceability, fatigue life and safety of the structure. In this paper, a structural control method is proposed to control the vibration of offshore structures by the probabilistic neural network (PNN). The state vectors of the structure and control forces are used for training patterns of the PNN, in which control forces are prepared by linear quadratic regulator (LQR) control algorithm. The proposed algorithm is applied to a fixed offshore structure under random ocean waves. Active control of the fixed offshore structure using the PNN control algorithm shows good results.

As an alternative measure to replace reactive stance with proactive one, a risk based management scheme has been commonly applied to enhance public satisfaction on water service by providing a higher creditable solution to handle a rehabilitation problem of pipe having high potential risk of leaks. This study intended to examine the feasibility of a simulation model to predict a recurrence probability of pipe leaks. As a branch of the data mining technique, probabilistic neural network (PNN) algorithm was applied to infer the extent of leaking recurrence probability of water network. PNN model could classify the leaking level of each unit segment of the pipe network. Pipe material, diameter, C value, road width, pressure, installation age as input variable and 5 classes by pipe leaking probability as output variable were built in PNN model. The study results indicated that it is important to pay higher attention to the pipe segment with the leak record. By increase the hydraulic pipe pressure to meet the required water demand from each node, simulation results indicated that about 6.9% of total number of pipe would additionally be classified into higher class of recurrence risk than present as the reference year. Consequently, it was convinced that the application of PNN model incorporated with a data base management system of pipe network to manage municipal water distribution network could make a promise to enhance the management efficiency by providing the essential knowledge for decision making rehabilitation of network.

구조 재료와 시공기술의 발달로 구조물은 높고 길게 설계할 수 있게 되었으나, 그에 따른 진동 문제와 사용성에 관한 문제가 발생하였고 구조물의 과다한 변위는 구조물에 심각한 손상을 발생 시켰다. 이러한 구조물의 진동 문제를 해결하기 위하여 본 논문에서는 확률신경망이론을 사용한 구조물의 능동제어방법을 제안하였다. 구조물의 제어를 위하여 LQR 제어알고리즘을 이용하여 구조물의 상태벡터와 제어력을 구한 후, 상태벡터를 입력으로 제어력을 출력으로 하는 확률신경망의 훈련패턴을 구성하였다. 제안된 방법을 사용하여 지진하중을 받는 3층 빌딩구조물을 제어하였고, 기존의 인공신경망의 제어 결과와 비교하였다.

The compressive strength of concrete is commonly used criterion in producing concrete. However, the tests on the compressive strength are complicated and time-consuming. More importantly, it is too late to make improvement even if the test result does not satisfy the required strength, since the test is usually performed at the 28th day after the placement of concrete at the construction site. Therefore, accurate and realistic strength estimation before the placement of concrete is being highly required. In this study, the estimation of the compressive strength of concrete was performed by probabilistic neural network (PNN) on the basis of concrete mix proportions. The estimation performance of PNN was improved by considering the correlation between input data and targeted output value. Adaptive probabilistic neural network (APNN) was proposed to automatically calculate the smoothing parameter in the conventional PNN by using the scheme of dynamic decay adjustment algorithm. The conventional PNN and APNN were applied to predict the compressive strength of concrete using actual test data of a concrete company. APNN showed better results than the conventional PNN in predicting the compressive strength of concrete.

To classify seafloor sediments using a probabilistic neural network (PNN), the frequency-dependent characteristics of broadband acoustic scattering, which make it possible to qualitatively categorize seabed type, were collected from three different geographical areas in Korea. The echo data samples from three types of seafloor sediment were measured using a chirp sonar system operating over a frequency range of 20-220 kHz. The spectrum amplitudes for frequency responses of 35-75 kHz were fed into the PNN as input feature parameters. The PNN algorithm could successfully identify three seabed types: mud, mud/shell and concrete sediments. The percentage probabilities of the three seabed types being correctly classified were 86% for mud, 66% for mud/shell and 72% for concrete sediment.

A new neuro-control scheme for active control of structures is proposed. It utilizes lattice pattern of state vector as training data of probabilistic neural network (PNN). Therefore, it is the so-called lattice probabilistic neural network (LPNN). PNN makes control forces by using all the training patterns. Therefore, it takes much time to obtain a control force in application. This inevitably may delay the control action. However, control force of LPNN is calculated by using only the adjacent information of LPNN input. So, the response of LPNN is greatly faster than PNN. The proposed control algorithm is applied for one story building under California and El Centro earthquakes. Also, control results of the LPNN are compared with those of the conventional PNN. The structural responses have been suppressed effectively by the proposed algorithm.

화자확인은 기본적으로 각 입력 음성에 대해 하나의 임계치를 기준으로 수락과 거부의 두 가지 결정을 내리나, 본 논문은 네 자리의 비밀번호를 음성으로 입력하였을 때 각 숫자음성에 대한 지역적인 결정을 두 개의 임계치를 이용하여 수락, 거부, 결정유보의 세 가지로 구분하고, 비밀번호 전체에 대한 판단 규칙을 제안하였다. 지역적 결정에 필요한 화자에 대한 신뢰척도의 측정치는 확률신경망을 통해 구하였다. 다섯 명의 화자를 대상으로 수행한 실험에서 하나의 임계치를 이용한 기존의 방식은 5.3%의 오류를 나타냈고, 본 논문에서 제안한 방식은 2.1%의 오류를 보였다.