• Tunnel and Underground Space
• /
• v.26 no.1
• /
• pp.46-58
• /
• 2016

In this study, a LIDAR laser scanner was used to scan a rock slope around Mt. Gwanak and to produce point cloud from which directional information of rock joint surfaces shall be extracted. It was analyzed using two different algorithms, i.e. Ball Pivoting and Wrap algorithm, and four sampling intervals, i.e. raw, 2, 5, and 10 cm. The results of Fuzzy K-mean clustering were analyzed on the stereonet. As a result, the Ball Pivoting and Wrap algorithms were considered suitable for extraction of rock surface orientation. In the case of 5 cm sampling interval, both triangulation algorithms extracted the most number of the patch and patched area.

• Journal of Korea Society of Industrial Information Systems
• /
• v.11 no.1
• /
• pp.77-85
• /
• 2006

In this paper, we propose a hierarchical classifier of object images using neural networks for content-based image classification. The images for classification are object images that can be divided into foreground and background. In the preprocessing step, we extract the object region and shape-based texture features extracted from wavelet transformed images. We group the image classes into clusters which have similar texture features using Principal Component Analysis(PCA) and K-means. The hierarchical classifier has five layes which combine the clusters. The hierarchical classifier consists of 59 neural network classifiers learned with the back propagation algorithm. Among the various texture features, the diagonal moment was the most effective. A test with 1000 training data and 1000 test data composed of 10 images from each of 100 classes shows classification rates of 81.5% and 75.1% correct, respectively.

• Journal of the Institute of Electronics Engineers of Korea SC
• /
• v.48 no.1
• /
• pp.31-38
• /
• 2011

Faults of motor drive systems to be used for various industrial applications can cause serious problems. In this paper, a method to diagnose switch open fault of a voltage-fed PWM inverter is proposed. The proposed method normalizes dq current and fault-detection and first classification are performed by mean values of dq phase currents, second classification is performed by features such as the relation of dq phase currents, the ranges of those, the positions of those according to the results, and fault switch is diagnosed with the results. The proposed method performs the simulation for diagnosis of inverter switch open faults with MATLAB and identifies the feasibility of the proposed method. Because the proposed method is implemented by simple algorithms, the proposed algorithm can be embedded in general induction motor drive systems and be used.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.24 no.6
• /
• pp.665-670
• /
• 2014

A SVM is a kind of binary classifier in order to find optimal hyperplane which separates training data into two groups. Due to its remarkable performance, the SVM is applied in various fields such as inductive inference, binary classification or making predictions. Also it is a representative black box model; there are plenty of actively discussed researches about analyzing trained SVM classifier. This paper conducts a study on a method that is automatically detecting the line-shaped echoes, sun strobe echo and radial interference echo, using the SVM algorithm because the line-shaped echoes appear relatively often and disturb weather forecasting process. Using a spatial clustering method and corrected reflectivity data in the weather radar, the training data is made up with mean reflectivity, size, appearance, centroid altitude and so forth. With actual occurrence cases of the line-shaped echoes, the trained SVM classifier is verified, and analyzed its characteristics using the decision tree method.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2015.10a
• /
• pp.183-185
• /
• 2015

Recently, big data analysis based on machine learning has gained popularity and many machine learning techniques have been applied to the field of agriculture. By using machine learning technique to analyze huge number of samples of biological and environmental data, new observations can be found. In this research, we consider the estimation of proportional daily weight increase (PDWI) based on measurement data from experimental swine farm. In order to derive the exact formulation for PDWI estimation, we have used measured value of mean, daily maximum, daily minimum of temperature, humidity, CO2, wind speed and measured PDWI values. Based on collected data, we have derived equation for PDWI estimation using tree-based algorithm. In the derived formulation, we have found that the daily average temperature is the most dominant factor that affects PDWI. Our results can be applied to pig farms to estimate the PDWI of swine.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.38 no.6
• /
• pp.26-37
• /
• 2001

In order to improve the several problems of the general backpropagation, we propose a method using a fuzzy logic system for automatic tuning of the activation function gain in the backpropagation. First, we researched that the changing of the gain of sigmoid function is equivalent to changing the learning rate, the weights, and the biases. The inputs of the fuzzy logic system were the sensitivity of error respect to the last layer and the mean sensitivity of error respect to the hidden layer, and the output was the gain of the sigmoid function. In order to verify the effectiveness of the proposed method, we performed simulations on the parity problem, function approximation, and pattern recognition. The results show that the proposed method has considerably improved the performance compared to the general backpropagation.

• Journal of rehabilitation welfare engineering & assistive technology
• /
• v.10 no.3
• /
• pp.199-206
• /
• 2016

Most Koreans walk having their toes in or out, because of their sedentary lifestyles. In addition, using smartphone while walking makes having a desirable walking posture even more difficult. The goal of this study is to make a simple system which easily analyze and inform any person his or her personal walking habit. To discriminate gait patterns, we developed a gait monitoring system using a 3-axis accelerometer and a foot pressure monitoring system. The developed system, with an accelerometer and a few pressure sensors, can acquire subject's foot pressure and how tilted his or her torso is. We analyzed the relationship between type of gate and sensor data using this information. As the result of analysis, we could find out that statistical parameters like standard deviation and root mean square are good for discriminating among torso postures, and k-nearest neighbor algorithm is good at clustering gait patterns. The developed system is expected to be applicable to medical or athletic fields at a low price.

• The Journal of Korean Society for Radiation Therapy
• /
• v.29 no.2
• /
• pp.9-17
• /
• 2017

Purpose: The tissue description and electron density indicated by the Computed Tomography(CT) number (also known as Hounsfield Unit) in radiotherapy are important in ensuring the accuracy of CT-based computerized radiotherapy planning. The internal metal implants, however, not only reduce the accuracy of CT number but also introduce uncertainty into tissue description, leading to development of many clinical algorithms for reducing metal artifacts. The purpose of this study was, therefore, to investigate the accuracy and the clinical applicability by analyzing date from SMART MAR (GE) used in our institution. Methode: and material: For assessment of images, the original images were obtained after forming ROIs with identical volumes by using CIRS ED phantom and inserting rods of six tissues and then non-SMART MAR and SMART MAR images were obtained and compared in terms of CT number and SD value. For determination of the difference in dose by the changes in CT number due to metal artifacts, the original images were obtained by forming PTV at two sites of CIRS ED phantom CT images with Computerized Treatment Planning (CTP system), the identical treatment plans were established for non-SMART MAR and SMART MAR images by obtaining unilateral and bilateral titanium insertion images, and mean doses, Homogeneity Index(HI), and Conformity Index(CI) for both PTVs were compared. The absorbed doses at both sites were measured by calculating the dose conversion constant (cCy/nC) from ylinder acrylic phantom, 0.125cc ionchamber, and electrometer and obtaining non-SMART MAR and SMART MAR images from images resulting from insertions of unilateral and bilateral titanium rods, and compared with point doses from CTP. Result: The results of image assessment showed that the CT number of SMART MAR images compared to those of non-SMART MAR images were more close to those of original images, and the SD decreased more in SMART compared to non-SMART ones. The results of dose determinations showed that the mean doses, HI and CI of non-SMART MAR images compared to those of SMART MAR images were more close to those of original images, however the differences did not reach statistical significance. The results of absorbed dose measurement showed that the difference between actual absorbed dose and point dose on CTP in absorbed dose were 2.69 and 3.63 % in non-SMRT MAR images, however decreased to 0.56 and 0.68 %, respectively in SMART MAR images. Conclusion: The application of SMART MAR in CT images from patients with metal implants improved quality of images, being demonstrated by improvement in accuracy of CT number and decrease in SD, therefore it is considered that this method is useful in dose calculation and forming contour between tumor and normal tissues.

• 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.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2010.11a
• /
• pp.14-17
• /
• 2010

Following the proliferation of three-dimensional video contents and displays, many terrestrial broadcasting companies prepare for starting stereoscopic 3DTV service. In terrestrial stereoscopic broadcast, it is a difficult task to code and transmit two video sequences while sustaining as high quality as 2DTV broadcast attains due to the limited bandwidth defined by the existing digital TV standards such as ATSC. Thus, a terrestrial 3DTV broadcasting system with heterogeneous video coding systems is considered for terrestrial 3DTV broadcast where the left image and right images are based on MPEG-2 and H.264/AVC, respectively, in order to achieve both high quality broadcasting service and compatibility for the existing 2DTV viewers. Without significant change in the current terrestrial broadcasting systems, we propose a joint rate control scheme for stereoscopic 3DTV service. The proposed joint rate control scheme applies to the MPEG-2 encoder a quadratic rate-quantization model which is adopted in the H.264/AVC. Then the controller is designed for the sum of two bit streams to meet the bandwidth requirement of broadcasting standards while the sum of image distortions is minimized by adjusting quantization parameter computed from the proposed optimization scheme. Besides, we also consider a condition on quality difference between the left and right images in the optimization. Experimental results demonstrate that the proposed bit rate control scheme outperforms the rate control method where each video coding standard uses its own bit rate control algorithm in terms of minimizing the mean image distortion as well as the mean value and the variation of absolute image quality differences.