• 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 Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.37 no.3
• /
• pp.199-208
• /
• 2019

As the number of available satellites increases and technology advances, image information outputs are becoming increasingly diverse and a large amount of data is accumulating. In this study, we propose a change detection method for high-resolution satellite images that uses transfer learning and a deep learning network to overcome the limit caused by insufficient training data via the use of pre-trained information. The deep learning network used in this study comprises convolutional layers to extract the spatial and spectral information and convolutional long-short term memory layers to analyze the time series information. To use the learned information, the two initial convolutional layers of the change detection network are designed to use learned values from 40,000 patches of the ISPRS (International Society for Photogrammertry and Remote Sensing) dataset as initial values. In addition, 2D (2-Dimensional) and 3D (3-dimensional) kernels were used to find the optimized structure for the high-resolution satellite images. The experimental results for the KOMPSAT-3A (KOrean Multi-Purpose SATllite-3A) satellite images show that this change detection method can effectively extract changed/unchanged pixels but is less sensitive to changes due to shadow and relief displacements. In addition, the change detection accuracy of two sites was improved by using 3D kernels. This is because a 3D kernel can consider not only the spatial information but also the spectral information. This study indicates that we can effectively detect changes in high-resolution satellite images using the constructed image information and deep learning network. In future work, a pre-trained change detection network will be applied to newly obtained images to extend the scope of the application.

• Korean Journal of Remote Sensing
• /
• v.39 no.6_1
• /
• pp.1321-1330
• /
• 2023

Normalized Difference Vegetation Index (NDVI) is the most widely used remote sensing data in the agricultural field and is currently provided by most optical satellites. In particular, as high-resolution optical satellite images become available, the selection of optimal optical satellite images according to agricultural applications has become a very important issue. In this study, we aim to define the most optimal optical satellite image when monitoring NDVI in rice fields in Korea and derive the resolution-related requirements necessary for this. For this purpose, we compared and analyzed the spatial distribution and time series patterns of the Dangjin rice paddy in Korea from 2019 to 2022 using NDVI images from MOD13, Landsat-8, Sentinel-2A/B, and PlanetScope satellites, which are widely used around the world. Each data is provided with a spatial resolution of 3 m to 250 m and various periods, and the area of the spectral band used to calculate NDVI also has slight differences. As a result of the analysis, Landsat-8 showed the lowest NDVI value and had very low spatial variation. In comparison, the MOD13 NDVI image showed similar spatial distribution and time series patterns as the PlanetScope data but was affected by the area surrounding the rice field due to low spatial resolution. Sentinel-2A/B showed relatively low NDVI values due to the wide near-infrared band area, and this feature was especially noticeable in the early stages of growth. PlanetScope's NDVI provides detailed spatial variation and stable time series patterns, but considering its high purchase price, it is considered to be more useful in small field areas than in spatially uniform rice paddy. Accordingly, for rice field areas, 250 m MOD13 NDVI or 10 m Sentinel-2A/B are considered to be the most efficient, but high-resolution satellite images can be used to estimate detailed physical quantities of individual crops.

• Journal of Intelligence and Information Systems
• /
• v.24 no.1
• /
• pp.1-23
• /
• 2018

From the 21st century, various high-quality services have come up with the growth of the internet or 'Information and Communication Technologies'. Especially, the scale of E-commerce industry in which Amazon and E-bay are standing out is exploding in a large way. As E-commerce grows, Customers could get what they want to buy easily while comparing various products because more products have been registered at online shopping malls. However, a problem has arisen with the growth of E-commerce. As too many products have been registered, it has become difficult for customers to search what they really need in the flood of products. When customers search for desired products with a generalized keyword, too many products have come out as a result. On the contrary, few products have been searched if customers type in details of products because concrete product-attributes have been registered rarely. In this situation, recognizing texts in images automatically with a machine can be a solution. Because bulk of product details are written in catalogs as image format, most of product information are not searched with text inputs in the current text-based searching system. It means if information in images can be converted to text format, customers can search products with product-details, which make them shop more conveniently. There are various existing OCR(Optical Character Recognition) programs which can recognize texts in images. But existing OCR programs are hard to be applied to catalog because they have problems in recognizing texts in certain circumstances, like texts are not big enough or fonts are not consistent. Therefore, this research suggests the way to recognize keywords in catalog with the Deep Learning algorithm which is state of the art in image-recognition area from 2010s. Single Shot Multibox Detector(SSD), which is a credited model for object-detection performance, can be used with structures re-designed to take into account the difference of text from object. But there is an issue that SSD model needs a lot of labeled-train data to be trained, because of the characteristic of deep learning algorithms, that it should be trained by supervised-learning. To collect data, we can try labelling location and classification information to texts in catalog manually. But if data are collected manually, many problems would come up. Some keywords would be missed because human can make mistakes while labelling train data. And it becomes too time-consuming to collect train data considering the scale of data needed or costly if a lot of workers are hired to shorten the time. Furthermore, if some specific keywords are needed to be trained, searching images that have the words would be difficult, as well. To solve the data issue, this research developed a program which create train data automatically. This program can make images which have various keywords and pictures like catalog and save location-information of keywords at the same time. With this program, not only data can be collected efficiently, but also the performance of SSD model becomes better. The SSD model recorded 81.99% of recognition rate with 20,000 data created by the program. Moreover, this research had an efficiency test of SSD model according to data differences to analyze what feature of data exert influence upon the performance of recognizing texts in images. As a result, it is figured out that the number of labeled keywords, the addition of overlapped keyword label, the existence of keywords that is not labeled, the spaces among keywords and the differences of background images are related to the performance of SSD model. This test can lead performance improvement of SSD model or other text-recognizing machine based on deep learning algorithm with high-quality data. SSD model which is re-designed to recognize texts in images and the program developed for creating train data are expected to contribute to improvement of searching system in E-commerce. Suppliers can put less time to register keywords for products and customers can search products with product-details which is written on the catalog.

• Journal of the Korean Society of Costume
• /
• v.36
• /
• pp.183-203
• /
• 1998

The wave of Avant-garde swept away all in the unique social background so called 'October Revolution' and the early 1900 Russian society which was able to absorb and accept anything. The Russian avant-garde has been affected by the Cubism and the Futurism those had peculiarly appeared in the early twentieth century, spreaded out to three spheres: the Suprematism, the Rayonism and the Constructivism. The Russian Constructivism has appeared in this background, concretely and ideally ex-pressed the ideology of the revolution into the artistic form and made an huge influence to the whole Russian society. The Constructivist like Tatlin, naum Gabo, Pevaner, Rodchenko, Stepanova, Popova and Exter gave great effect on the Soviet Constructive fashion design in 1920's after the Revolution. The Soviet costume in 1920s hold in common the characteristicss of the Constructive graphic as it is, geometrical and abstractive form, energetic and motility. In fashion design, these graphic qualities have been showed as the application of geometrical form and architectural image, physical distortion and transformation. And in textile design, the simple, dynamical presentation has been appeared. We can classify the Soviet costume at this time into three occasions. The first term is from late 1910 th mid 1920, and it is altered from folk costume design to modern one. With Lamanova as the first on the list, using the folk mitif, the Constructive expression of simple form has been gradually revealed in design. Designers like Makarova, Pribylskaia and Mukhina produced the plane, simple chemise style with the decoration of the Russian traditional motif. From early to late 1920 is the second term, and it is at the pick of the most active processing of the Constructive design. Not only at the costume in daily life but also at the theatrical costume and textile, the con-structive design has been represented all avail-able fields. Many Constructivists including Stepanova, Popova, Exter and Rodchenko took part in the textile design and costume design so as to evlvo their aesthetic concept. The third term is from late 1920 to early 1930. The socialistic realism has dominated over the whole culture and art, the revolutionary dynamic motif has been presented also in textile design. The formative features of Soviet Constructive fashion design are; silhouette, from, motif, color and fabric. The first, the silhouette : a straight rectangular silhouetted has been expressed through the whole period and a volumed one with distorted human body shape has introduced in the theatrical costume design. The second, the form: many lengthened rectangular forms have been made at beginnings, but to the middle period, geometrical, architectural forms have been more showed and there are energy and movement in design. At the last period, only a partial feature-division has been seen. The third, the motif; no pattern or ethnic motif has been partly used at beginnings, a figure like circle, tri-angle has gradually appeared in textile design. At latter period, a real-existent motif like an airplane has been represented with graphing and simplicity. The fourth, the color ; because of insufficient dyeing, neutral color like black or grey color has been mainly covered, but after middle term, a primary color or pastel tone has been seen, contrast of the fabric; without much development of textile industry after the Revolution, thick and durable fabrics have been the main stream, but as time had going to the last period, fabrics such as linen, cotton, velvet and silk have been varously choesn. At the theatrical costume, new materials like plastics and metals that were able to accentuate the form. The pursuit of popularity, simplicity and functionalism that the basic concept of Constructive fashion is one of the "beauty" which has been searching in modern fashion. And now we can appreciate how innovative and epochal this Soviet Constructive fashion movement was.ement was.

• The Korean Journal of Malacology
• /
• v.24 no.3
• /
• pp.229-241
• /
• 2008

Changes of growth and histopathological feature in the mantle structure of the equilateral venus, Gomphina veneriformis exposed to tribultyltin chloride (TBTCl) for 36 weeks were observed. Concentrations of TBTCl were 0, 0.4, 0.6, and $0.8{\mu}g/L$. A regression analysis by power function of SPSS was shown that the growth of experimental groups was significantly decreased after 12 weeks of exposure. For histological analysis, mantle tissues were characterized using H-E stain, AB-PAS (pH 2.5) reaction and Masson's trichrome stain, and epidermal layer thickness and mucous cell distribution were analysed using the image analyser. The mantle had 4-folds (inner-inner, inner-outer, middle, and outer) and its epidermal layer consisted of simple epithlia. A periostracum was observed in the periostracal groove between middle and outer fold. Inner epidermal layer consisted of simple ciliated columnar epithelia, but the outer epidermal layer consisted of simple non-ciliated columnar epithelia. Alcian blue positive mucous cells showed blue color (7462c, 653c) in the inner fold, violet color (2583c) in the middle fold, and blue color (647c, 7455c) in inner epidermal layer (numbers in the parenthesis are codes of Pantone process coated color). Hemolymph sinus in the mantle was extended, and mucous cells in inner plica of the middle fold were stained as blue (7455c) and violet (2587c), after 12 weeks of TBTCI exposure. Cilia and striated border were disappeared, and number of mucous cells in the inner epidermal layer was reduced. Serious histopathological changes in middle and outer fold near the periostracum were observed after 36 weeks. Moreover, epidermal layer thickness and mucous cell distribution were showed decreasing tendency as exposure time to TBTCI was increased. Results of this study suggested that TBTCl induced growth disorder with histopathological changes.

• Investigative Magnetic Resonance Imaging
• /
• v.18 no.1
• /
• pp.1-6
• /
• 2014

Purpose : $T_2{^*}$ relaxation time which includes susceptibility information represents unique feature of tissue. The objective of this study was to investigate $T_2{^*}$ relaxation times of the normal glandular tissue and fat of breast using a 3T MRI system. Materials and Methods: Seven-echo MR Images were acquired from 52 female subjects (age $49{\pm}12 $years; range, 25 to 75) using a three-dimensional (3D) gradient-echo sequence. Echo times were between 2.28 ms to 25.72 ms in 3.91 ms steps. Voxel-based $T_2{^*}$ relaxation times and $R_2{^*}$ relaxation rate maps were calculated by using the linear curve fitting for each subject. The 3D regions-of-interest (ROI) of the normal glandular tissue and fat were drawn on the longest echo-time image to obtain $T_2{^*}$ and $R_2{^*}$ values. Mean values of those parameters were calculated over all subjects. Results: The 3D ROI sizes were $4818{\pm}4679$ voxels and $1455{\pm}785$ voxels for the normal glandular tissue and fat, respectively. The mean $T_2{^*}$ values were $22.40{\pm}5.61ms$ and $36.36{\pm}8.77ms$ for normal glandular tissue and fat, respectively. The mean $R_2{^*}$ values were $0.0524{\pm}0.0134/ms$ and $0.0297{\pm}0.0069/ms$ for the normal glandular tissue and fat, respectively. Conclusion: $T_2{^*}$ and $R_2{^*}$ values were measured from human breast tissues. $T_2{^*}$ of the normal glandular tissue was shorter than that of fat. Measurement of $T_2{^*}$ relaxation time could be important to understand susceptibility effects in the breast cancer and the normal tissue.

• Cartoon and Animation Studies
• /
• s.48
• /
• pp.359-383
• /
• 2017

3,000 years ago, since the first poster of humanity appeared in Egypt, the invention of printing technique and the development of civilization have accelerated the poster production technology. In keeping with this, the expression of poster has also been developed as an attempt to express artistic sensibility in a simple arrangement of characters, and now it has become an art form that has become a domain of professional designers. However, the technological development in the expression of poster is keep staying in two-dimensional, and is dependent on printing only that it is irrelevant to the change of ICT environment based on modern multimedia. Especially, among the many kinds of posters, the style of movie posters, which are the only objects for video, are still printed on paper, and many attempts have been made so far, but the movie industry still does not consider ICT integration at all. This study started with the feature that the object of the movie poster dealt with the video and attempted to introduce the augmented reality to apply the dynamic image of the movie to the static poster. In the graduation work of the media design major of a university in Korea, the poster of each works for promoting the visual work of the students was designed and printed in the form of a commercial film poster. Among them, 6 artworks that are considered to be suitable for augmented reality were selected and augmented reality was introduced and exhibited. Content that appears matched to the poster through the mobile device is reproduced on a poster of a scene of the video, but the text informations of the original poster are kept as they are, so that is able to build a moving poster looked like a wanted from the movie "Harry Potter". In order to produce this augmented reality poster, we applied augmented reality to posters of existing commercial films produced in two different formats, and found a way to increase the characteristics of AR contents. Through this, we were able to understand poster design suitable for AR representation, and technical expression for stable operation of augmented reality can be summarized in the matching process of augmented reality contents production.

• Korean Journal of Remote Sensing
• /
• v.31 no.5
• /
• pp.449-459
• /
• 2015

The conventional National Forest Inventory(NFI)-based forest carbon stock estimation method is suitable for national-scale estimation, but is not for regional-scale estimation due to the lack of NFI plots. In this study, for the purpose of regional-scale carbon stock estimation, we created grid-based forest carbon stock maps using spatial ancillary data and two types of up-scaling methods. Chungnam province was chosen to represent the study area and for which the $5^{th}$ NFI (2006~2009) data was collected. The first method (method 1) selects forest type map as ancillary data and uses regression model for forest carbon stock estimation, whereas the second method (method 2) uses satellite imagery and k-Nearest Neighbor(k-NN) algorithm. Additionally, in order to consider uncertainty effects, the final AGB carbon stock maps were generated by performing 200 iterative processes with Monte Carlo simulation. As a result, compared to the NFI-based estimation(21,136,911 tonC), the total carbon stock was over-estimated by method 1(22,948,151 tonC), but was under-estimated by method 2(19,750,315 tonC). In the paired T-test with 186 independent data, the average carbon stock estimation by the NFI-based method was statistically different from method2(p<0.01), but was not different from method1(p>0.01). In particular, by means of Monte Carlo simulation, it was found that the smoothing effect of k-NN algorithm and mis-registration error between NFI plots and satellite image can lead to large uncertainty in carbon stock estimation. Although method 1 was found suitable for carbon stock estimation of forest stands that feature heterogeneous trees in Korea, satellite-based method is still in demand to provide periodic estimates of un-investigated, large forest area. In these respects, future work will focus on spatial and temporal extent of study area and robust carbon stock estimation with various satellite images and estimation methods.

• Tuberculosis and Respiratory Diseases
• /
• v.54 no.1
• /
• pp.80-90
• /
• 2003

Background : Goblet cell hyperplasia is a critical pathological feature in hypersecretory diseases of the airways. A bacterial infection of the lung is also known to induce inflammatory responses, which can lead to the overproduction of mucus. Recently, mucin synthesis in the airways has been reported to be regulated by neutrophilic inflammation-induced epidermal growth factor receptor (EGFR) expression and activation. In addition, it was reported that migration of the activated neutrophils is dependent on the matrix metalloproteinases (MMPs), especially MMP-9. In this study, bacterial lipopolysaccharide (LPS)-induced goblet cell hyperplasia and mucus hypersecretion by EGFR cascade, resulting from the MMPs-dependent neutrophilic inflammation were investigated in the rat airways. Methods : Pathogen-free Sprague-Dawley rats were studied in vivo. Various concentrations of LPS were instilled into the trachea in $300{\mu}{\ell}$ PBS (LPS group). Sterile PBS ($300{\mu}{\ell}$) was instilled into the trachea of the control animals (control group). The airways were examined on different days after instilling LPS. For an examination of the relationship between the LPS-induced goblet cell hyperplasia and MMPs, the animals were pretreated 3 days prior to the LPS instillation and daily thereafter with the matrix metalloproteinase inhibitor (MMPI; 20 mg/Kg/day of CMT-3; Collagenex Pharmaceuticals, USA). The neutrophilic infiltration was quantified as a number in five high power fields (HPF). The alcian blue/periodic acid-Schiff (AB/PAS) stain were performed for the mucus glycoconjugates and the immunohistochemical stains were performed for MUC5AC, EGFR and MMP-9. Their expressions were quantified by an image analysis program and were expressed by the percentage of the total bronchial epithelial area. Results : The instillation of LPS induced AB/PAS and MUC5AC staining in the airway epithelium in a time- and dose-dependent manner. Treatment with the MMPI prevented the LPS-induced goblet cell hyperplasia significantly. The instillation of LPS into the trachea induced also EGFR expression in the airway epithelium. The control airway epithelium contained few leukocytes, but the intratracheal instillation of LPS resulted in a neutrophilic recruitment. A pretreatment with MMPI prevented neutrophilic recruitment, EGFR expression, and goblet cell hyperplasia in the LPS-instilled airway epithelium. Conclusion : Matrix metalloproteinase is involved in LPS-induced mucus hypersecretion, resulting from a neutrophilic inflammation and EGFR cascade. These results suggest a potential therapeutic role of MMPI in the treatment of mucus hypersecretion that were associated with a bacterial infection of the airways.