• Journal of Internet Computing and Services
• /
• v.18 no.6
• /
• pp.25-34
• /
• 2017

Autoencoder is a type of deep learning method where input layer and output layer are the same, and effectively extracts and restores characteristics of input vector using constraints of hidden layer. In this paper, we propose methods of Autoencoders to remove a natural background image which is a noise to the CAPTCHA and recover only a numerical images by applying various autoencoder models to a region where one number of CAPTCHA images and a natural background are mixed. The suitability of the reconstructed image is verified by using the softmax function with the output of the autoencoder as an input. And also, we compared the proposed methods with the other method and showed that our methods are superior than others.

• Journal of the Society of Naval Architects of Korea
• /
• v.52 no.3
• /
• pp.255-264
• /
• 2015

Wind load and flow field of a drillship with respect to various super structures were experimentally investigated in KARI 1m-wide wind tunnel with an atmospheric boundary layer simulation. Six-component external balance and Particle image velocimetry technique were used to measure wind load and velocity vectors in the flow-field around the model respectively. The experimental model was an imaginary shaped drillship with an approximated model which has 1/640 scale compared with recent typical drillships. The test Reynolds number based on the overall length was about 1.5×10⁶. It was found that dominant factors influencing on ship wind load are cabin shape and cabin height. Round cabin has smaller axial wind load and narrow boundary layer around the ship than rectangular one, but its yawing moment at certain angles becomes higher. Low cabin height also show positive effects on axial wind load too. Hull shape and forecastle shape show relatively small influences on wind loads except for slight changes around ±45° wind directions.

• Bulletin of the Korean Chemical Society
• /
• v.24 no.5
• /
• pp.593-599
• /
• 2003

The nanoparticle magnetite of which diameter was about 3 nm was synthesized in a homogeneous aqueous solution without a template. The synthesized magnetite nanoparticle was easily oxidized to maghemite in an ambient condition. The magnetic properties of the ferrite nanoparticle show superparamagnetism at room temperature and its blocking temperature is around 93 K. Modifying the sequential adsorption method of metal bisphosphonate, we have prepared a multilayer thin film of the ferrite nanoparticle on planar substrates such as glass, quartz and Si wafer. In this multilayer the ferrite nanoparticle layer and an alkylbisphosphonate layer are alternately placed on the substrates by simple immersion in the solutions of the ferrite nanoparticle and 1, 10-decanediylbis (phosphonic acid) (DBPA), alternately. This is the first example, as far as we know, of nanoparticle/alkyl-bisphosphonate multilayer which is an analogy of metal bisphosphonate multilayer. UV-visible absorption and infrared reflection-absorption studies show that the growth of each layer is very systematic and the film is considerably optically transparent to visible light of 400-700 nm. Atomic force microscopic images of the film show that the surface morphology of the film follows that of the substrate in μm-scale image and the nanoparticle-terminated surface is differentiated from the DBPA-terminated one in nm-scale image. The magnetic properties of this ferrite/DBPA thin film are almost the same as those of the ferrite nanoparticle powder only.

• Journal of Sensor Science and Technology
• /
• v.2 no.1
• /
• pp.29-34
• /
• 1993

a-Si : H photodiodes for image sensor have been fabricated and characterized. Photosensitivity of a ITO/a-Si : H/Al photodiode without blocking layer was 0.7 under the applied voltage of 5 V and peak spectral sensitivity in visible region was found at 620 nm. Dark current of ITO/a-SiN : H/a-Si : H/p-a-Si : H/Al photodiode was suppressed by hole blocking layer and electron blocking layer at the value of lower than 1.5 pA to the applied voltage of 10 V. Also maximum photosensitivity was about 1 under the applied voltage of 3 V and peak spectral sensitivity was found at 540 nm.

• Journal of Sensor Science and Technology
• /
• v.22 no.4
• /
• pp.262-267
• /
• 2013

We developed a real-time screening system using optical coherence tomography (OCT) to distinguish the fruitful melon seeds efficiently. Cross-section images of melon seeds infected with Cucumber green mottle mosaic virus (CGMMV) showed an additional layer that did not appear in normal seeds. Additional layer appeared under $100{\sim}300{\mu}m$ from the surface of the seed. OCT can visualize the micro-structural and morphological changes of the internal seed structure. Real-time OCT seed screening system provided the real-time, non-destructive, cross-section image and quantitative information such as A-scan analysis of selected region in the cross-section image. We can distinguish the viral infection seeds while monitoring the averaged A-scan analysis graph in real-time by considering the second peak value of the graph which refers to the layer that occurred owing to the virus. Real-time OCT seed screening system could assist to distinguish the disease caused by CGMMV.

• Journal of Korea Multimedia Society
• /
• v.21 no.12
• /
• pp.1425-1430
• /
• 2018

In this paper, we propose a CNN based deep learning algorithm for semantic segmentation of images. In order to improve the accuracy of semantic segmentation, we combined pixel level object classification and image level object classification. The image level object classification is used to accurately detect the characteristics of an image, and the pixel level object classification is used to indicate which object area is included in each pixel. The proposed network structure consists of three parts in total. A part for extracting the features of the image, a part for outputting the final result in the resolution size of the original image, and a part for performing the image level object classification. Loss functions exist for image level and pixel level classification, respectively. Image-level object classification uses KL-Divergence and pixel level object classification uses cross-entropy. In addition, it combines the layer of the resolution of the network extracting the features and the network of the resolution to secure the position information of the lost feature and the information of the boundary of the object due to the pooling operation.

• Journal of Korea Multimedia Society
• /
• v.23 no.9
• /
• pp.1155-1163
• /
• 2020

Deep learning approach is emerging as a new way to improve the accuracy of tree species identification using bark image. However, the approach has not been studied enough because it is confronted with the problem of acquiring a large volume of bark image dataset. This study solved this problem by utilizing a pretrained off-the-shelf DCNN model. It compares the discrimination power of bark features extracted by each DCNN model. Then it extracts the features by using a selected DCNN model and feeds them to a multi-layer perceptron (MLP). We found out that the ResNet50 model is effective in extracting bark features and the MLP could be trained well with the features reduced by the principal component analysis. The proposed approach gives accuracy of 99.1% and 98.4% for BarkTex and Trunk12 datasets respectively.

• Journal of the Korean Society for Precision Engineering
• /
• v.27 no.4
• /
• pp.102-108
• /
• 2010

Projection microstereolithography is a process of fabricating a micro-structure by using dynamic mask such as digital micromirror device(DMD). DMD shapes the beam into cross-sectional image of structure. Photocurable resin is cured by the beam and stacked layer on top of layer. It is difficult to deliver the beam from the DMD to the photocurable resin without any distortions. We assume that the beam exposed to the resin by 1 pixel of DMD has Gaussian distribution, so the shaped beam reflected by the DMD affects its neighboring area. Curing pattern corresponding to a cross-sectional images is predicted by superposition of pixels of Gaussian distribution and it is similar to cured shape.

• Journal of the Korean Society of Manufacturing Process Engineers
• /
• v.12 no.3
• /
• pp.36-40
• /
• 2013

Recently, electronic equipments become smaller, more functional, and more complex than before. As these trends, MLC(multi-layer ceramic) circuit has been emerged to a promising technology in semiconductor inspection industry. Especially, multi-layer ceramic which is consisted of many fine-pitch multi-hole is used to produce a semiconductor inspection unit. The hole is processed by UV laser. But, working conditions are changed all the time. Therefore real time measurement of fine-pitch multi-hole is very important method for ensuring performance. In this paper we found the best method for illuminating and auto focusing. And, we verified our equipment.

• 제어로봇시스템학회:학술대회논문집
• /
• 1996.10b
• /
• pp.605-608
• /
• 1996

The aim of this thesis is to develop a SMT-components placement inspection system equipped with a visual sensor. The visual sensor, which consists of a camera and 2-layer LED illuminator, developed to inspect the component placement state such as missing, shift, flipping, polarity and tomb-stone. on PCB in the reflow-process. In practical applications, however, it is too hard to classify component from images mixed pad on PCB, cream solder paste and component. To overcome the problem, this thesis proposes the 2-layer illumination method and the heuristic image processing algorithms according to inspection type. To show the effectiveness of the proposed approach, a series of experiments on the inspection were conducted. The results show that the proposed method is robust to visual noise and variations in component conditions.