• Journal of Biomedical Engineering Research
• /
• v.19 no.5
• /
• pp.455-468
• /
• 1998

For the objective interpretation of cerebral metabolic patterns in epilepsy patients, we developed computer-aided classifier using artificial neural network. We studied interictal brain FDG PET scans of 257 epilepsy patients who were diagnosed as normal(n=64), L TLE (n=112), or R TLE (n=81) by visual interpretation. Automatically segmented volume of interest (VOI) was used to reliably extract the features representing patterns of cerebral metabolism. All images were spatially normalized to MNI standard PET template and smoothed with 16mm FWHM Gaussian kernel using SPM96. Mean count in cerebral region was normalized. The VOls for 34 cerebral regions were previously defined on the standard template and 17 different counts of mirrored regions to hemispheric midline were extracted from spatially normalized images. A three-layer feed-forward error back-propagation neural network classifier with 7 input nodes and 3 output nodes was used. The network was trained to interpret metabolic patterns and produce identical diagnoses with those of expert viewers. The performance of the neural network was optimized by testing with 5~40 nodes in hidden layer. Randomly selected 40 images from each group were used to train the network and the remainders were used to test the learned network. The optimized neural network gave a maximum agreement rate of 80.3% with expert viewers. It used 20 hidden nodes and was trained for 1508 epochs. Also, neural network gave agreement rates of 75~80% with 10 or 30 nodes in hidden layer. We conclude that artificial neural network performed as well as human experts and could be potentially useful as clinical decision support tool for the localization of epileptogenic zones.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.16 no.4
• /
• pp.31-40
• /
• 2016

In vehicular ad-hoc networks (VANETs), vehicles sense information on emergency incidents (e.g., accidents, unexpected road conditions, etc.) and propagate this information to following vehicles and a server to share the information. However, this process of emergency message propagation is based on multiple broadcast messages and can lead to broadcast storms. To address this issue, in this work, we use a novel approach to detect the vehicles that are farthest away but within communication range of the transmitting vehicle. Specifically, we discuss a signal-to-noise ratio (SNR)-based linear back-off (SLB) scheme where vehicles implicitly detect their relative locations to the transmitter with respect to the SNR of the received packets. Once the relative locations are detected, nodes that are farther away will set a relatively shorter back-off to prioritize its forwarding process so that other vehicles can suppress their transmissions based on packet overhearing. We evaluate SLB using a realistic simulation environment which consists of a NS-3 VANET simulation environment, a software-based WiFi-IP gateway, and an ITS server operating on a separate machine. Comparisons with other broadcasting-based schemes indicate that SLB successfully propagates emergency messages with latencies and hop counts that is close to the experimental optimal while reducing the number of transmissions by as much as 1/20.

• Journal of Intelligence and Information Systems
• /
• v.22 no.2
• /
• pp.127-142
• /
• 2016

Deep learning model is a kind of neural networks that allows multiple hidden layers. There are various deep learning architectures such as convolutional neural networks, deep belief networks and recurrent neural networks. Those have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. Among those architectures, convolutional neural networks and recurrent neural networks are classified as the supervised learning model. And in recent years, those supervised learning models have gained more popularity than unsupervised learning models such as deep belief networks, because supervised learning models have shown fashionable applications in such fields mentioned above. Deep learning models can be trained with backpropagation algorithm. Backpropagation is an abbreviation for "backward propagation of errors" and a common method of training artificial neural networks used in conjunction with an optimization method such as gradient descent. The method calculates the gradient of an error function with respect to all the weights in the network. The gradient is fed to the optimization method which in turn uses it to update the weights, in an attempt to minimize the error function. Convolutional neural networks use a special architecture which is particularly well-adapted to classify images. Using this architecture makes convolutional networks fast to train. This, in turn, helps us train deep, muti-layer networks, which are very good at classifying images. These days, deep convolutional networks are used in most neural networks for image recognition. Convolutional neural networks use three basic ideas: local receptive fields, shared weights, and pooling. By local receptive fields, we mean that each neuron in the first(or any) hidden layer will be connected to a small region of the input(or previous layer's) neurons. Shared weights mean that we're going to use the same weights and bias for each of the local receptive field. This means that all the neurons in the hidden layer detect exactly the same feature, just at different locations in the input image. In addition to the convolutional layers just described, convolutional neural networks also contain pooling layers. Pooling layers are usually used immediately after convolutional layers. What the pooling layers do is to simplify the information in the output from the convolutional layer. Recent convolutional network architectures have 10 to 20 hidden layers and billions of connections between units. Training deep learning networks has taken weeks several years ago, but thanks to progress in GPU and algorithm enhancement, training time has reduced to several hours. Neural networks with time-varying behavior are known as recurrent neural networks or RNNs. A recurrent neural network is a class of artificial neural network where connections between units form a directed cycle. This creates an internal state of the network which allows it to exhibit dynamic temporal behavior. Unlike feedforward neural networks, RNNs can use their internal memory to process arbitrary sequences of inputs. Early RNN models turned out to be very difficult to train, harder even than deep feedforward networks. The reason is the unstable gradient problem such as vanishing gradient and exploding gradient. The gradient can get smaller and smaller as it is propagated back through layers. This makes learning in early layers extremely slow. The problem actually gets worse in RNNs, since gradients aren't just propagated backward through layers, they're propagated backward through time. If the network runs for a long time, that can make the gradient extremely unstable and hard to learn from. It has been possible to incorporate an idea known as long short-term memory units (LSTMs) into RNNs. LSTMs make it much easier to get good results when training RNNs, and many recent papers make use of LSTMs or related ideas.

• Journal of the Korea institute for structural maintenance and inspection
• /
• v.10 no.5
• /
• pp.179-186
• /
• 2006

In this study, Multi-Layer Perceptron(MLP) among models of Artificial Neural Network(ANN) is used for the development of a model that evaluates the bending capacities of reinforced concrete beams strengthened by FRP Rebar. And the data of the existing researches are used for materials of ANN model. As the independent variables of input layer, main components of bending capacities, width, effective depth, compressive strength, reinforcing ratio of FRP, balanced steel ratio of FRP are used. And the moment performance measured in the experiment is used as the dependent variable of output layer. The developed model of ANN could be applied by GFRP, CFRP and AFRP Rebar and the model is verified by using the documents of other previous researchers. As the result of the ANN model presumption, comparatively precise presumption values are achieved to presume its bending capacities at the model of ANN(0.05), while observing remarkable errors in the model of ANN(0.1). From the verification of the ANN model, it is identified that the presumption values comparatively correspond to the given data ones of the experiment. In addition, from the Sensitivity Analysis of evaluation variables of bending performance, effective depth has the highest influence, followed by steel ratio of FRP, balanced steel ratio, compressive strength and width in order.

• The Transactions of the Korea Information Processing Society
• /
• v.2 no.6
• /
• pp.937-946
• /
• 1995

We proposed and efficient block segmentation and classification method for the document analysis using SGLDM(spatial gray level dependence matrix) and BP (back Propagation) neural network. Seven texture features are extracted directly from the SGLDM of each gray-level block image, and by using the nonlinear classifier of neural network BP, we can classify document blocks into 9 categories. The proposed method classifies the equation block, the table block and the flow chart block, which are mostly composed of the characters, out of the blocks that are conventionally classified as non-character blocks. By applying Sobel operator on the gray-level document image beforebinarization, we can reduce the effect of the background noises, and by using the additional horizontal-vertical smoothing as well as the vertical-horizontal smoothing of images, we can obtain an effective block segmentation that does not lead to the segmentation into small pieces. The result of experiment shows that a document can be segmented and classified into the character blocks of large fonts, small fonts, the character recognigible candidates of tables, flow charts, equations, and the non-character blocks of photos, figures, and graphs.

• Journal of the Institute of Convergence Signal Processing
• /
• v.10 no.1
• /
• pp.7-12
• /
• 2009

In this paper, we propose a directional feature extraction method for off-line handwritten numerals by using the morphological operations. Direction features are obtained from four directional line images, each of which contains horizontal, vertical, right-diagonal and left-diagonal lines in entire numeral lines. Conventional method for extracting directional features uses Kirsch masks which generate edge-shaped double line images for each direction, whereas our method uses directional erosion operations and generate single line images for each direction. To apply these directional erosion operations to the numeral image, preprocessing steps such as thinning and dilation are required, but resultant directional lines are more similar to numeral lines themselves. Our four [$4{\times}4$] directional features of a numeral are obtained from four directional line images through a zoning method. For obtaining the higher recognition rates of the handwrittern numerals, we use the multiple feature which is comprised of our proposed feature and the conventional features of a kirsch directional feature and a concavity feature. For recognition test with given features, we use a multi-layer perceptron neural network classifier which is trained with the back propagation algorithm. Through the experiments with the CENPARMI numeral database of Concordia University, we have achieved a recognition rate of 98.35%.

• Journal of Sensor Science and Technology
• /
• v.6 no.6
• /
• pp.476-482
• /
• 1997

A portable electronic nose system has been fabricated and characterized using an oxide semiconductor gas sensor array and pattern recognition techniques such as principal component analysis and back-propagation artificial neural network. The sensor array consists of six thick-film gas sensors whose sensing layers are Pd-doped $WO_{3}$, Pt-doped $SnO_{2}$, $TiO_{2}-Sb_{2}O_{5}-Pd$-doped $SnO_{2}$, $TiO_{2}-Sb_{2}O_{5}-Pd$-doped $SnO_{2}$ + Pd coated layer, $Al_{2}O_{3}$-doped ZnO and $PdCl_{2}$-doped $SnO_{2}$. The portable electronic nose system consists of an 16bit Intel 80c196kc as CPU, an EPROM for storing system main program, an EEPROM for containing optimized connection weights of artificial neural network, an LCD for displaying gas concentrations. As an application the system has been used to identify 26 carbon monoxide/hydrocarbon (CO/HC) car exhausting gases in the concentration range of CO 0%/HC 0 ppm to CO 7.6%/HC 400 ppm and the identification has been successfully demonstrated.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1996.10a
• /
• pp.239-242
• /
• 1996

본 논문에서는 BP(Back Propagation)에 비해서 빠른 학습시간과 다른 경쟁학습 신경회로망 알고리즘에 비해서 비교적 우수한 성능으로 패턴인식 등에 많이 이용되고 있는 LVQ(Learning Vector Quantization) 알고리즘의 성능을 향상시키기 위한 방법을 논의하고자 한다. 일반적으로 LVQ는 음(negative)의 학습을 하기 때문에 초기 가중치가 제대로 설정되지 않으면 발산할 수 있다는 단점이 있으며, 경쟁학습 계열의 신경망이기 때문에 출력 층의 뉴런 수에 따라 성능에 큰 영향을 받는다고 알려져 있다.[1]. 지도학습 형태를 지닌 LVQ의 경우에 학습패턴이 n개의 클래스를 가지고, 각 클래스 별로 학습패턴의 수가 같은 경우에 일반적으로 전체 출력뉴런에 대해서 (출력뉴런수/n)개의 뉴런을 각 클래스의 목표(desired) 클러스터로 할당하여 학습을 수행하는데, 본 논문에서는 각 클래스에 동일한 수의 출력뉴런을 할당하지 않고, 학습데이터에서 각 클래스의 분산을 추정하여 각 클래스의 분산을 추정분산에 비례하게 목표 출력뉴런을 할당하고, 초기 가중치도 추정분산에 비례하게 각 클래스의 초기 임의 위치 입력백터를 사용하여 학습을 수행하는 방법을 제안한다. 본 논문에서 제안하는 방법은 분류하고자 하는 데이터에 대해서 필요한 최적의 출력뉴런 수를 찾는 것이 아니라 이미 결정되어 있는 출력뉴런 수에 대해서 각 클래스에 할당할 출력 뉴런 수를 데이터의 추정분산에 의해서 결정하는 것으로, 추정분산이 크면 상대적으로 많은 출력 뉴런을 할당하고 작으면 상대적으로 적은 출력뉴런을 할당하고 초기 가중치도 마찬가지 방법으로 결정하며, 이렇게 하면 정해진 출력뉴런 개수 안에서 각 클래스 별로 분류의 어려움에 따라서 출력뉴런을 할당하기 때문에 미학습 뉴런이 줄어들게 되어 성능의 향상을 기대할 수 있으며, 실험적으로 제안된 방법이 더 나은 성능을 보임을 확인했다.initially they expected a more practical program about planting than programs that teach community design. Many people are active in their own towns to create better environments and communities. The network system "Alpha Green-Net" is functional to support graduates of the course. In the future these educational programs for citizens will becomes very important. Other cities are starting to have their own progrms, but they are still very short term. "Alpha Green-Net" is in the process of growing. Many members are very keen to develop their own abilities. In the future these NPOs should become independent. To help these NPOs become independent and active the educational programs should consider and teach about how to do this more in the future.단하였는데 그 결과, 좌측 촉각엽에서 제4형의 신경연접이 퇴행성 변화를 나타내었다. 그러므로 촉각의 지각신경세포는 뇌의 같은 족 촉각엽에 뻗어와 제4형 신경연접을 형성한다고 결론되었다.$/ 값이 210 $\mu\textrm{g}$/$m\ell$로서 효과적인 저해 활성을 나타내었다 따라서, 본 연구에서 빈

• Journal of the Korea Concrete Institute
• /
• v.14 no.4
• /
• pp.457-466
• /
• 2002

An artificial neural network was applied to predict compressive strength, slump value and mix proportion of a concrete. Standard mixed tables were trained and estimated, and the results were compared with those of the experiments. To consider variabilities of material properties, the standard mixed fables from two companies of Ready Mixed Concrete were used. And they were trained with the neural network. In this paper, standard back propagation network was used. The mix proportion factors such as water cement ratio, sand aggregate ratio, unit water, unit cement, unit weight of sand, unit weight of crushed sand, unit coarse aggregate and air entraining admixture were used. For the arrangement on the approval of prediction of mix proportion factor, the standard compressive strength of $180kgf/cm^2{\sim}300kgf/cm^2$, and target slump value of 8 cm, 15 cm were used. For the arrangement on the approval of prediction of compressive strength and slump value, the standard compressive strength of $210kgf/cm^2{\sim}240kgf/cm^2$, and target slump value of 12 cm and 15 cm wore used because these ranges are most frequently used. In results, in the prediction of mix proportion factor, for all of the water cement ratio, sand aggregate ratio, unit water, unit cement, unit weight of sand, unit weight of crushed sand, unit coarse aggregate, air entraining admixture, the predicted values and the values of standard mixed tables were almost the same within the target error of 0.10 and 0.05, regardless of two companies. And in the prediction of compressive strength and slump value, the predicted values were converged well to the values of standard mixed fables within the target error of 0.10, 0.05, 0.001. Finally artificial neural network is successfully applied to the prediction of concrete mixture and compressive strength.

• Journal of Institute of Control, Robotics and Systems
• /
• v.16 no.12
• /
• pp.1150-1158
• /
• 2010

In this study, the Polynomial-based Radial Basis Function Neural Networks is proposed as one of the recognition part of overall face recognition system that consists of two parts such as the preprocessing part and recognition part. The design methodology and procedure of the proposed pRBFNNs are presented to obtain the solution to high-dimensional pattern recognition problem. First, in preprocessing part, we use a CCD camera to obtain a picture frame in real-time. By using histogram equalization method, we can partially enhance the distorted image influenced by natural as well as artificial illumination. We use an AdaBoost algorithm proposed by Viola and Jones, which is exploited for the detection of facial image area between face and non-facial image area. As the feature extraction algorithm, PCA method is used. In this study, the PCA method, which is a feature extraction algorithm, is used to carry out the dimension reduction of facial image area formed by high-dimensional information. Secondly, we use pRBFNNs to identify the ID by recognizing unique pattern of each person. The proposed pRBFNNs architecture consists of three functional modules such as the condition part, the conclusion part, and the inference part as fuzzy rules formed in 'If-then' format. In the condition part of fuzzy rules, input space is partitioned with Fuzzy C-Means clustering. In the conclusion part of rules, the connection weight of pRBFNNs is represented as three kinds of polynomials such as constant, linear, and quadratic. Coefficients of connection weight identified with back-propagation using gradient descent method. The output of pRBFNNs model is obtained by fuzzy inference method in the inference part of fuzzy rules. The essential design parameters (including learning rate, momentum coefficient and fuzzification coefficient) of the networks are optimized by means of the Particle Swarm Optimization. The proposed pRBFNNs are applied to real-time face recognition system and then demonstrated from the viewpoint of output performance and recognition rate.