• 전자공학회논문지B
• /
• 제32B권3호
• /
• pp.522-532
• /
• 1995

Multilayer neural networks with backpropagation learning algorithm are widely used for pattern classification problems. For many real applications, it is more important to reduce the misclassification rate than to increase the rate of successful classification. But multilayer perceptrons(MLP's) have drawbacks of slow learning speed and false convergence to local minima. In this paper, we propose a new method for character recognition problems with a single-layer network and double rejection mechanisms, which guarantees a very low misclassification rate. Comparing to the MLP's, it yields fast learning and requires a simple hardware architecture. We also introduce a new coding scheme to reduce the misclassification rate. We have prepared two databases: one with 135,000 digit patterns and the other with 117,000 letter patterns, and have applied the proposed method for printed character recognition, which shows that the method reduces the misclassification rate significantly without sacrificing the correct recognition rate.

• Journal of the Korean Data and Information Science Society
• /
• 제14권4호
• /
• pp.817-824
• /
• 2003

We employed a hierarchical document classification method to classify a massive collection of electronic medical records(EMR) written in both Korean and English. Our experimental system has been learned from 5,000 records of EMR text data and predicted a newly given set of EMR text data over 68% correctly. We expect the accuracy rate can be improved greatly provided a dictionary of medical terms or a suitable medical thesaurus. The classification system might play a key role in some clinical decision support systems and various interpretation systems for clinical data.

• 대한원격탐사학회:학술대회논문집
• /
• 대한원격탐사학회 2003년도 Proceedings of ACRS 2003 ISRS
• /
• pp.107-109
• /
• 2003

Optical sensor-based land cover categories produce spectral signature confusion along with degraded classification accuracy. In the classification tasks, the goal of fusing data from different sensors is to reduce the classification error rate obtained by single source classification. This paper describes the result of land cover/land use classification derived from solely of Landsat TM (TM) and multisensor image fusion between JERS 1 SAR (JERS) and TM data. The best radar data manipulation is fused with TM through various techniques. Classification results are relatively good. The highest Kappa Coefficient is derived from classification using principal component analysis-high pass filtering (PCA+HPF) technique with the Overall Accuracy significantly high.

• 한국방송∙미디어공학회:학술대회논문집
• /
• 한국방송공학회 2009년도 IWAIT
• /
• pp.525-528
• /
• 2009

Recently, image classification has been an important task in various fields. Generally, the performance of image classification is not good without the adjustment of image features. Therefore, it is desired that the way of automatic feature extraction. In this paper, we propose an image classification method which adjusts image features automatically. We assume that texture features are useful in image classification tasks because natural images are composed of several types of texture. Thus, the classification accuracy rate is improved by using distribution of texture features. We obtain texture features by calculating image features from a current considering pixel and its neighborhood pixels. And we calculate image features from distribution of textures feature. Those image features are adjusted to image classification tasks using Genetic Algorithm. We apply proposed method to classifying images into "head" or "non-head" and "male" or "female".

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제18권2호
• /
• pp.420-437
• /
• 2024

There are some problems in network traffic classification (NTC), such as complicated statistical features and insufficient training samples, which may cause poor classification effect. A NTC architecture based on one-dimensional Convolutional Neural Network (CNN) and transfer learning is proposed to tackle these problems and improve the fine-grained classification performance. The key points of the proposed architecture include: (1) Model classification--by extracting normalized rate feature set from original data, plus existing statistical features to optimize the CNN NTC model. (2) To apply transfer learning in the classification to improve NTC performance. We collect two typical network flows data from Youku and YouTube, and verify the proposed method through extensive experiments. The results show that compared with existing methods, our method could improve the classification accuracy by around 3-5%for Youku, and by about 7 to 27% for YouTube.

• 한국초등과학교육학회지:초등과학교육
• /
• 제24권3호
• /
• pp.281-291
• /
• 2005

The purpose of this study was to investigate elementary school pupil's classification ability that appears in classification activity. For this study, we developed 2 suitable tools in classification activity achievement. One is artificial stimulus card that comes into view clearly. The other is natural stimulus card that does not come into view well. The test was administrated to 376 pupils of 2, 4, and 6 grade in D elementary School in Yeongdeungpo-gu, Seoul. The result proved in this study was as following. First, elementary school pupil's classification ability showed the developmental change as the grade level rises. Second, there was no statistical difference between boys and girls. Third, there was high correlation between sort artificial category and natural category in their ability. Fourth, classification achievement rate of constant level by grade was seen regardless of the items. The findings above gives following guidance in science classification learning. First, if teacher understands the development of students' classification ability, more effective classification guidance is available. Second, to cultivate students' classification ability, we should devise and apply program depending on their classification ability by grade.

• 한국통신학회논문지
• /
• 제17권11호
• /
• pp.1175-1184
• /
• 1992

본 논문에서는 한자를 포함하는 한글 문서 인식을 위한 인쇄체 한글, 한자의 구분과 인식 방법에 대하여 연구하였다. 제안된 한글, 한자 구분 방법은 한글의 수직모음과 수평모음의 구조적 특징을 이용하였다. 한글은 6가지 형태로 분류하고 분류된 각 형태에 대하여 세선화 과정을 거치지 않고 모음 우선추출에 의한 자모분리를 행하고 분리된 자음에 대하여 변형된 교차거리 특징을 이용하여 인식하였다. 한자에 대해서는 획교차수의 평균치를 이용하여 전체 한자 대상문자에 대해 분류를 하였으며, 문자의 획교차수와 흑점비율 특징을 이용하여 인식하였다. 한글과 한자의 구분에서는 90.5%의 분류율을 얻었다. 한글인식에 있어서는 대상문자 명조체 2512자에 대하여 90.0%의 형태 분류율을 얻었다. 인식 결과 실험 데이타 1278자에 대하여 92.2%의 인식율을 얻었다. 한자인식에 있어서는 대상문자 4585자에 대하여 분류한 결과 최대밀집 구간은 124자로서 약 1/40 정도로 분류되었음을 알 수 있었고, 인식실험 결과 89.2%의 인식율을 얻었다.

• 전자공학회논문지B
• /
• 제31B권7호
• /
• pp.170-178
• /
• 1994

The nearest neighbor classification rule is widely used because it is not only simple but the error rate is asymptotically less than twice Bayes theoretical minimum error. But the method basically use the whole training patterns as the reference vectors. so that both storage and classification time increase as the number of training patterns increases. LVQ(Learning Vector Quantization) resolved this problem by training the reference vectors instead of just storing the whole training patterns. But it is a heuristic algorithm which has no theoretic background there is no terminating condition and it requires a lot of iterations to get to meaningful result. This paper is to propose a new training method of the reference vectors. which minimize the given error function. The nearest neighbor network,the network version of the nearest neighbor classification rule is proposed. The network is funtionally identical to the nearest neighbor classification rule is proposed. The network is funtionally identical to the nearest neighbor classification rule and the reference vectors are represented by the weights between the nodes. The network is trained to minimize the error function with respect to the weights by the steepest descent method. The learning algorithm is derived and it is shown that the proposed method can adjust more reference vectors than LVQ in each iteration. Experiment showed that the proposed method requires less iterations and the error rate is smaller than that of LVQ2.

• 한국정보과학회논문지:소프트웨어및응용
• /
• 제33권5호
• /
• pp.489-498
• /
• 2006

본 논문은 루프검지기를 이용한 차종분류 방법의 성능 향상을 위해 신경망 패턴인식 기술을 이용한 차종분류 알고리즘을 제안하였다. 기존의 루프검지기 차종분류 방법은 차량의 길이 정보만을 이용해서 차종을 분류하는 것이다. 그러나 루프검지기의 특성상 차종에 따른 길이 정보가 정확하지 않으므로 길이가 비슷한 차종에 대해서는 차종분류 오류가 자주 발생하고 있는 실정이다. 이와 같은 문제점을 개선하기 위해 본 연구에서는 루프검지기 시스템에 신경망 패턴 인식 기술을 적용하였다. 제안된 알고리즘은 차량이 검지영역을 통과할 때 발생하는 루프검지기 공진주파수 값 변화율과 점유시간 정보를 신경망의 입력자료로 활용하여 차량을 5가지 종류로 분류하는 방식이다. 개발된 알고리즘의 성능을 평가하기 위하여, 현장실험을 통해 자료를 수집하고 신경망 학습 및 실험을 실시한 결과 차종분류 정확도가 91.3%였으며, 이는 기존의 연구결과와 비교할 때 매우 높은 것이다.

• 한국멀티미디어학회논문지
• /
• 제6권5호
• /
• pp.824-830
• /
• 2003

메모리 기반 추론에서 기억공간의 효율적인 사용과 분류성능의 향상을 위하여 제안되었던 RPA(Recursive Partition Averaging)알고리즘은 대상 패턴 공간을 분할 한 후 대표 패턴을 추출하여 분류 기준 패턴으로 사용한다. 이 기법은 구성된 초월 평면상에서 단순히 대표패턴을 추출하여 분류 성능 저하의 원인이 되는 단점을 가지고 있었다. 여기에서는 기존 RPA의 단점을 보완하기 위해 FPD (Feature-based Population Densimeter)를 이용한 OHC (Optimized Hyperrectangle Calving) 알고리즘을 제안한다. 제안된 알고리즘은 RPA분할 종료 후 OHC를 이용하여 초월 평면을 최적화한 후 패턴 평균 기법을 적용하여 학습 결과를 산출한다. 제안된 알고리즘은 k-NN분류기에서 필요로 하는 메모리 공간의 40%정도를 사용하며 분류에 있어서도 RPA보다 우수한 인식 성능을 보이고 있다. 또한 저장된 패턴의 감소로 인하여, 실제 분류에 소요되는 시간비교에 있어서도 k-NN보다 월등히 우수한 성능을 보이고 있다.