• KIPS Transactions on Software and Data Engineering
• /
• v.7 no.7
• /
• pp.259-266
• /
• 2018

In this paper, we propose a method to classify a document using a Recurrent Neural Network by extracting features considering word sense and contexts. Word2vec method is adopted to include the order and meaning of the words expressing the word in the document as a vector. Doc2vec is applied for considering the context to extract the feature of the document. RNN classifier, which includes the output of the previous node as the input of the next node, is used as the document classification method. RNN classifier presents good performance for document classification because it is suitable for sequence data among neural network classifiers. We applied GRU (Gated Recurrent Unit) model which solves the vanishing gradient problem of RNN. It also reduces computation speed. We used one Hangul document set and two English document sets for the experiments and GRU based document classifier improves performance by about 3.5% compared to CNN based document classifier.

• Journal of Welding and Joining
• /
• v.34 no.2
• /
• pp.59-66
• /
• 2016

The Gas Metal Arc(GMA) welding, called Metal Inert Gas(MIG) welding, has been an important component in manufacturing industries. A key technology for robotic welding processes is seam tracking system, which is critical to improve the welding quality and welding capacities. The objectives of this study were to develop the intelligent and cost-effective algorithms for image processing in GMA welding which based on the laser vision sensor. Welding images were captured from the CCD camera and then processed by the proposed algorithm to track the weld joint location. The proposed algorithms that commonly used at the present stage were verified and compared to obtain the optimal one for each step in image processing. Finally, validity of the proposed algorithms was examined by using weld seam images obtained with different welding environments for image processing. The results proved that the proposed algorithm was quite excellent in getting rid of the variable noises to extract the feature points and centerline for seam tracking in GMA welding and could be employed for general industrial application.

• Proceedings of the Korean Association of Geographic Inforamtion Studies Conference
• /
• 2004.03a
• /
• pp.493-498
• /
• 2004

The objective of the present study is to compare various conventional and recently evolved satellite image-processing techniques and to ascertain the best possible technique that can identify and position of aquaculture farms accurately in and around the Younggwang coastal area. Several conventional techniques performed to extract such information fiom the Landsat-TM imagery do not seem to yield better information about the aquaculture farms, and lead to misclassification. The large errors between the actual and extracted aquaculture farm information are due to existence of spectral confusion and inadequate spatial resolution of the sensor. This leads to possible occurrence of mixture pixels or 'mixels' of the source of errors in the classification techniques. Understanding the confusing and mixture pixel problems requires the development of efficient methods that can enable more reliable extraction of aquaculture farm information. Thus, the more recently evolved methods such as the step-by-step partial spectral end-member extraction and linear spectral unmixing methods are introduced. The farmer one assumes that an end-member, which is often referred to as 'spectrally pure signature' of a target feature, does not appear to be a spectrally pure form, but always mix with the other features at certain proportions. The assumption of the linear spectral unmxing is that the measured reflectance of a pixel is the linear sum of the reflectance of the mixture components that make up that pixel. The classification accuracy of the step-by-step partial end-member extraction improved significantly compared to that obtained from the traditional supervised classifiers. However, this method did not distinguish the aquaculture ponds and non-aquaculture ponds within the region of the aquaculture farming areas. In contrast, the linear spectral unmixing model produced a set of fraction images for the aquaculture, water and soil. Of these, the aquaculture fraction yields good estimates about the proportion of the aquaculture farm in each pixel. The acquired proportion was compared with the values of NDVI and both are positively correlated (R$^2$ =0.91), indicating the reliability of the sub-pixel classification.ixel classification.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.17 no.2
• /
• pp.147-157
• /
• 2017

In recent years, Super-pixels have become very popular for use in computer vision applications. Super-pixel algorithm transforms pixels into perceptually feasible regions to reduce stiff features of grid pixel. In particular, super-pixels are useful to depth estimation, skeleton works, body labeling, and feature localization, etc. But, it is not easy to generate a good super-pixel partition for doing these tasks. Especially, super-pixels do not satisfy more meaningful features in view of the gestalt aspects such as non-sum, continuation, closure, perceptual constancy. In this paper, we suggest an advanced algorithm which combines simple linear iterative clustering with fuzzy clustering concepts. Simple linear iterative clustering technique has high adherence to image boundaries, speed, memory efficient than conventional methods. But, it does not suggest good compact and regular property to the super-pixel shapes in context of gestalt aspects. Fuzzy similarity measures provide a reasonable graph in view of bounded size and few neighbors. Thus, more compact and regular pixels are obtained, and can extract locally relevant features. Simulation shows that fuzzy similarity based super-pixel building represents natural features as the manner in which humans decompose images.

• The Journal of Society for e-Business Studies
• /
• v.22 no.2
• /
• pp.61-77
• /
• 2017

Recently, the number of academic conference information on the Internet has rapidly increased, the automatic classification of academic conference information according to research subjects enables researchers to find the related academic conference efficiently. Information provided by most conference listing services is limited to title, date, location, and website URL. However, among these features, the only feature containing topical words is title, which causes information insufficiency problem. Therefore, we propose methods that aim to resolve information insufficiency problem by utilizing web contents. Specifically, the proposed methods the extract main contents from a HTML document collected by using a website URL. Based on the similarity between the title of a conference and its main contents, the topical keywords are selected to enforce the important keywords among the main contents. The experiment results conducted by using a real-world dataset showed that the use of additional information extracted from the conference websites is successful in improving the conference classification performances. We plan to further improve the accuracy of conference classification by considering the structure of websites.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.15 no.2
• /
• pp.245-252
• /
• 2020

Unexpected equipment defects can cause a huge economic losses in the society at large. Although condition monitoring can provide solutions, the signal processing algorithms must be developed to predict mechanical failures using data acquired from various sensors attached to the equipment. The signal processing algorithms used in a condition monitoring requires high computing efficiency and resolution. To improve condition monitoring on a wireless sensor network(WSN), data visualization can maximize the expressions of the data characteristics. Thus, this paper proposes the extraction of visual feature from temperature data over time using condition monitoring based on a WSN to identify environmental conditions of equipment in a large-scale infrastructure. Our results show that time-frequency analysis can visually track temperature changes over time and extract the characteristics of temperature data changes.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.24 no.6
• /
• pp.622-627
• /
• 2014

This paper proposes a new method which improves recognition rate on the gait recognition system using wavelet transform, linear discriminant analysis and genetic algorithm. We use wavelet transform to obtain the four sub-bands from the gait energy image. In order to extract feature data from sub-bands, we use linear discriminant analysis. Distance values between training data and four sub-band data are calculated and four weights which are calculated by genetic algorithm is assigned at each sub-band distance. Based on a new fusion distance value, we conducted recognition experiments using k-nearest neighbors algorithm. Experimental results show that the proposed weight fusion method has higher recognition rate than conventional method.

• Proceedings of the KIEE Conference
• /
• 2000.07d
• /
• pp.2843-2845
• /
• 2000

A lane detection based on a road model or feature all need correct acquirement of information on the lane in a image, It is inefficient to implement a lane detection algorithm through the full range of a image when being applied to a real road in real time because of the calculating time. This paper defines two searching range of detecting lane in a road, First is searching mode that is searching the lane without any prior information of a road, Second is recognition mode, which is able to reduce the size and change the position of a searching range by predicting the position of a lane through the acquired information in a previous frame. It is allow to extract accurately and efficiently the edge candidates points of a lane as not conducting an unnecessary searching. By means of removing the perspective effect of the edge candidate points which are acquired by using the inverse perspective transformation, we transform the edge candidate information in the Image Coordinate System(ICS) into the plane-view image in the World Coordinate System(WCS). We define linear approximation filter and remove the fault edge candidate points by using it This paper aims to approximate more correctly the lane of an actual road by applying the least-mean square method with the fault-removed edge information for curve fitting.

• Journal of Digital Contents Society
• /
• v.12 no.3
• /
• pp.279-289
• /
• 2011

The latest hardware and software techniques are combined to give safe walking guidance and convenient service of realtime walking assistance system for visually impaired person. This system consists of obstacle detection and perception, place recognition, and sign recognition for pedestrian can safely walking to arrive at their destination. In this paper, we exploit the sign object detection system in subway station for sign recognition that one of the important factors of walking assistance system. This paper suggest the adaptive feature map that can be robustly extract the sign object region from complexed environment with light and noise. And recognize a sign using fast zernike moment features which is invariant under translation, rotation and scale of object during walking. We considered three types of signs as arrow, restroom, and exit number and perform the training and recognizing steps through adaboost classifier. The experimental results prove that our method can be suitable and stable for real-time system through yields on the average 87.16% stable detection rate and 20 frame/sec of operation time for three types of signs in 5000 images of sign database.

• Journal of Service Research and Studies
• /
• v.5 no.2
• /
• pp.103-110
• /
• 2015

Recently, Classification of characters in a 3D FPS game has emerged as a very significant issue. In this study, We propose the game character Classification method using Image Region Segmentation of the extracting meaningful object in a simple operation. In this method, first used a non-linear RGB color model and octree color quantization scheme. The input image represented a less than 20 quantized color and uses a small number of meaningful color histogram. And then, the image divided into small blocks, calculate the degree of similarity between the color histogram intersection and adjacent block in block units. Because, except for the block boundary according to the texture and to extract only the boundaries of the object block. Set a region by these boundary blocks as a game object and can be used for FPS game play. Through experiment, we obtain accuracy of more than 80% for Classification method using each feature. Thus, using this property, characters could be classified effectively and it draws the game more speed and strategic actions as a result.