Earticle

It is important to understand the exact habitat distribution of endangered species because of their decreasing numbers. In this study, we build a system with a deep learning module that collects the image data of endangered animals, processes the data, and saves the data automatically. The system provides a more efficient way than human effort for classifying images and addresses two problems faced in previous studies. First, specious answers were suggested in those studies because the probability distributions of answer candidates were calculated even if the actual answer did not exist within the group. Second, when there were more than two entities in an image, only a single entity was focused on. We applied an object detection algorithm (YOLO) to resolve these problems. Our system has an average precision of 86.79%, a mean recall rate of 93.23%, and a processing speed of 13 frames per second.

Artificial intelligence, especially deep learning technology, is penetrating the majority of research areas, including the field of bioinformatics. However, deep learning has some limitations, such as the complexity of parameter tuning, architecture design, and so forth. In this study, we analyze these issues and challenges in regards to its applications in bioinformatics, particularly genomic analysis and medical image analytics, and give the corresponding approaches and solutions. Although these solutions are mostly rule of thumb, they can effectively handle the issues connected to training learning machines. As such, we explore the tendency of deep learning technology by examining several directions, such as automation, scalability, individuality, mobility, integration, and intelligence warehousing.

Lane-changing represents not only a common driving behavior but also a potentially hazardous one. Accurately predicting lane change intentions plays a crucial role in enhancing road traffic safety and guiding autonomous vehicle planning. In this study, a Face-mesh model is used to extract salient features from complex driver behavior data. Subsequently, by using the Farneback optical flow algorithm in conjunction with the ResNet-50 neural network, important lane change cues were extracted from the vehicle surroundings. The Transformer model was optimized using the Teacher-forcing training strategy and the Scheduled-sampling method, fostering faster convergence and heightened prediction accuracy. Empirical tests had shown that this model had attained an impressive precision of 98.61%, recall of 98.24 %, and an F1 score of 98.42 % when forecasting lane change intentions 0.5 s ahead.

In the medical fields, many efforts have been made to develop and improve Hospital Information System (HIS) including Electronic Medical Record (EMR), Order Communication System (OCS), and Picture Archiving and Communication System (PACS). However, materials generated and used in medical fields have various types and forms. The current HISs separately store and manage them by different systems, even though they relate to each other and contain redundant data. These systems are not helpful particularly in emergency where medical experts cannot check all of clinical materials in the golden time. Therefore, in this paper, we propose a process to build an integrated data model for medical information currently stored in various HISs. The proposed data model integrates vast information by focusing on medical images since they are most important materials for the diagnosis and treatment. Moreover, the model is disease-specific to consider that medical information and clinical materials including images are different by diseases. Two case studies show the feasibility and the usefulness of our proposed data model by building models about two diseases, acute myocardial infarction (AMI) and ischemic stroke.

This letter proposes a powerful method to reconstruct the spectral image frames for increasing data transmission capability in Display Field Communication (DFC) systems. The proposed method reconstructs the spectral image iteratively using the good data symbol estimates fed back by the camera decoder. To begin with, initial image is estimated by means of known pilot symbols. Then, hard symbol decisions obtained by iterative decoding are used to improve the quality of the reconstructed image. The proposed technique selects good data pixels using hard information from the decoder and performs re-estimation of the image pixels. From numerical simulations, we show that the proposed method significantly boosts the data transmission capacity of the DFC system over the conventional DFC that uses reference frames.

This paper proposes a novel reversible data hiding scheme based on a Vector Quantization (VQ) codebook. The proposed scheme uses the principle component analysis (PCA) algorithm to sort the codebook and to find two similar codewords of an image block. According to the secret to be embedded and the difference between those two similar codewords, the original image block is transformed into a difference number table. Finally, this table is compressed by entropy coding and sent to the receiver. The experimental results demonstrate that the proposed scheme can achieve greater hiding capacity, about five bits per index, with an acceptable bit rate. At the receiver end, after the compressed code has been decoded, the image can be recovered to a VQ compressed image.

Advancements in deep learning have enhanced vision-based aggregate analysis. However, further development and studies have encountered challenges, particularly in acquiring large-scale datasets. Data collection is costly and time-consuming, posing a significant challenge in acquiring large datasets required for training neural networks. To address this issue, this study introduces a simulation that efficiently generates the necessary data and labels for training neural networks. We utilized a genetic algorithm (GA) to create optimized lists of aggregates based on the specified values of weight and particle size distribution for the aggregate sample. This enabled sample data collection without conducting sieving tests. Our evaluation of the proposed simulation and GA methodology revealed errors of 1.3% and 2.7 g for aggregate size distribution and weight, respectively. Furthermore, we assessed a segmentation model trained with data from the simulation, achieving a promising preliminary F1 score of 78.18 on the actual aggregate image.

Data is critical in deep learning but the scarcity of data often occurs in research, especially in the preparation of the paired training data. In this paper, document image binarization with unpaired data is studied by introducing adversarial learning, excluding the need for supervised or labeled datasets. However, the simple extension of the previous unpaired training to binarization inevitably leads to poor performance compared to paired data training. Thus, a new deep learning approach is proposed by introducing a multi-diversity of higher quality generated images. In this paper, a two-stage model is proposed that comprises the generative adversarial network (GAN) followed by the U-net network. In the first stage, the GAN uses the unpaired image data to create paired image data. With the second stage, the generated paired image data are passed through the U-net network for binarization. Thus, the trained U-net becomes the binarization model during the testing. The proposed model has been evaluated over the publicly available DIBCO dataset and it outperforms other techniques on unpaired training data. The paper shows the potential of using unpaired data for binarization, for the first time in the literature, which can be further improved to replace paired data training for binarization in the future.

High-resolution image data taken by drones is being used as spatial information in various industries such as building maintenance, facility diagnosis, and urban planning modeling. In the industrial field, it is easy to visualize 3D on the web without installing separate software, and it is easy to run on various devices such as desktops and tablets and tablets regardless of platform environment, so it is used for visualization of drone high-resolution image data. However, in the process of processing high-resolution image data on a web browser, it has structural limitations such as rendering limitations, resampling, and downscaling, which degrade the actual image quality. This study proposes a method to restore drone high-resolution image quality using SRCNN (Super Resolution Convolutional Neural Network) so that the degraded image quality can be expressed at the original level on a web browser. We compared the restoration performance of applying SRCNN in a local environment for some of the 91 images taken by a drone at the Industry-Academic Cooperation Center building of *** University in Busan, Korea. The results show that SRCNN can be applied to restore image quality. In this study, a possible method for improving the image quality degradation problem in web browsers using AI-based post-processing was presented, and it is expected to contribute to web visualization and platform implementation of large amounts of high-resolution image data in the future.

We report processing technique in the MO image measurement system. Calibration procedure is not only considered to perpendicular field but also in-plane field. Current density and field profiles are obtained by Biot-savart law and inversion method. We show example of (Gd,Y)1Ba2Cu3O7-δ-BaZrO3 film that have tilted nano rod pinning centers about 13° from the c-axis.

3D scanning is a technique to measure the 3D shape information of the object. Shape information obtained by 3D scanning is expressed either as point cloud or as polygon mesh type data that can be widely used in various areas such as reverse engineering and quality inspection. 3D scanning should be performed as accurate as possible since the scanned data is highly required to detect the features on an object in order to scan the shape of the object more precisely. In this study, we propose the method on finding the location of feature more accurately, based on the extended Biplane SNAKE with global optimization. In each iteration, we project the feature lines obtained by the extended Biplane SNAKE into each image plane and move the feature lines to the features on each image. We have applied this approach to real models to verify the proposed optimization algorithm.

수자원의 계절적 편중이 심한 한반도에서 농업용 저수지는 이를 효과적으로 유지 및 관리하기 위한 필수적인 구조물이다. 저수지 모니터링을 위한 수단으로 광학 및 합성개구레이더(Synthetic Aperture Radar, SAR) 위성영상이 활용되고 있으나, 광학영상은 기상현상에 의한 간섭이 심하다는 한계점이 존재하며, SAR 영상은 짙은 식생에서 일어나는 다중 산란 및 노이즈에 의한 오탐지 및 미탐지가 발생하기 쉽다. 이에 본 연구에서는 광학 영상과 SAR 영상의 융합을 통해 저수지 수체 탐지 정확도를 높이고 상호보완적 작용에 대해 정량적으로 분석하고자 하였다. 경기도 이동저수지, 충청남도 천태 저수지를 대상으로, 국내 고해상도 위성인 차세대중형위성 1호, 다목적실용위성 3호 및 3A호, 그리고 유럽우주국의 Sentinel-2 영상 기반 Normalized Difference Water Index (NDWI)와 SAR 탑재 위성인 Sentinel-1 단일 영상에 비지도학습 기법인 K-means 클러스터링 기법을 사용하여 수체를 탐지하고, NDWI-SAR 후방산란계수로 이루어진 2-D grid space에 동일 기법을 활용하여 정확도의 향상 정도를 파악하였다. 전반적인 정확도는 다목적실용위성이 가장 높은 것으로 나타났으며(두 저수지 모두 0.98), 이후 Sentinel-1(두 저수지 모두 0.93), Sentinel-2(이동: 0.83, 천태: 0.97), 차세대중형위성(이동: 0.69, 천태: 0.78) 순서로 감소하였다. 천태저수지에서 2-D K-means 클러스터링 기법을 적용한 결과 차세대중형위성의 수체탐지 정확도는 약 85%의 정밀도 향상과 14%의 재현율 감소와 함께 약 22% 향상되었으며(정확도 약 0.95), 다목적실용위성 및 Sentinel-2의 수체탐지 정밀도는 3-5% 향상되었고, 재현율은 4-7% 감소하였다. 추후 차세대중형위성 5호인 수자원위성 등 고해상도 SAR 위성과 이를 활용할 수 있는 고도화된 영상 융합기술, 수체 탐지 기술이 개발된다면 국내 수자원에 대한 매우 정확한 모니터링이 가능할 것으로 기대된다.

Agricultural reservoirs are crucial structures for water resources monitoring especially in Korea where the resources are seasonally unevenly distributed. Optical and Synthetic Aperture Radar (SAR) satellites, being utilized as tools for monitoring the reservoirs, have unique limitations in that optical sensors are sensitive to weather conditions and SAR sensors are sensitive to noises and multiple scattering over dense vegetations. In this study, we tried to improve water body detection accuracy through optical-SAR data fusion, and quantitatively analyze the complementary effects. We first detected water bodies at Edong, Cheontae reservoir using the Compact Advanced Satellite 500(CAS500), Kompsat-3/3A, and Sentinel-2 derived Normalized Difference Water Index (NDWI), and SAR backscattering coefficient from Sentinel-1 by K-means clustering technique. After that, the improvements in accuracies were analyzed by applying K-means clustering to the 2-D grid space consists of NDWI and SAR. Kompsat-3/3A was found to have the best accuracy (0.98 at both reservoirs), followed by Sentinel-2(0.83 at Edong, 0.97 at Cheontae), Sentinel-1(both 0.93), and CAS500(0.69, 0.78). By applying K-means clustering to the 2-D space at Cheontae reservoir, accuracy of CAS500 was improved around 22%(resulting accuracy: 0.95) with improve in precision (85%) and degradation in recall (14%). Precision of Kompsat-3A (Sentinel-2) was improved 3%(5%), and recall was degraded 4%(7%). More precise water resources monitoring is expected to be possible with developments of high-resolution SAR satellites including CAS500-5, developments of image fusion and water body detection techniques.

이 논문에서는 비정형 의료이미지 정형화 및 패턴 추출을 위해, 의료기기 또는 의료전문가로부터 생성 또는 감지되는 의료이미지 저장을 위한 데이터베이스 구축한다. 이러한 비정형 이미지의 특징을 정형화된 디지털 데이터로 변환한 후 정형화된 디지털 이미지 데이터로부터 의미있는 패턴 정보를 생성한다. 이러한 경험 기술 소개를 통해 많은 연구자들은 의료이미지 데이터베이스를 보다 쉽게 접근할 수 있고 다양한 분야에서 정형화된 의료이미지를 활용할 수 있다.

This paper suggested that medical image database construction technique that generated and recognized from variable medical device and professional medical experts for the formalization and pattern extraction from informal medical images. And then we transformed informal image characteristics to digital data, and generated the meaningful pattern matching informations. Through this experienced works, so many related researchers can easily access the medical images database and use this formalized image informations on the variable fields.

Shape reconstruction is considered as a new technology to be useful and important in many areas such as RPD (Rapid Product Development) and reverse engineering, compared with the conventional design and manufacturing. In shape reconstruction, it becomes possible to reconstruct objects not by their measured shape data but those data extracted from the original shape. The goal of this research is to realize 3D shape construction by showing a possible way to analyze the input image data and reconstruct that original shape. The main 2 steps of the reconstructing process are getting cross-section data from image processing and linking loops between one slice and the next one. And the reconstructed object in this way is compared with the other object using a laser scanner and modelled by an commercial software.

사회의 발달에 따라 정보의 양이 폭발적으로 증가하게 되어 이를 효율적으로 유지, 보관하기 위해 디지털 아카이브의 필요성이 대두되었다. 현재 아카이브의 표준으로는 OAIS(Open Archival Information System) 참조 모형이 있다. 이 모형은 장기보존에만 관심이 있고, 주로 포괄적인 콘텐츠에만 신경을 쓰고 있다. 본 논문에서는 이미지 자료를 대상으로 장기보존만이 아닌 효율적인 활용을 위한 디지털 아카이브 워크플로우와 메타데이터에 대한 모델을 제시하고자 한다. 이미지 자료 디지털 아카이브 구축을 위해서는 OAIS 참조 모형을 통해 서지 정보와 미디어 형태의 메타데이터 표준이 추가되어야 한다는 필요성이 제기된다. 이미지 정보의 효율적인 디지털 아카이빙 및 활용을 위한 클래스 기반 다단계 메타데이터 모델 관리를 제안한다.

As the amount of information increases explosively with the development of society, the need for a digital archive has emerged in order to maintain information efficiently. There is a current standard of digital archive is OAIS(Open Archival Information System) reference model. OAIS reference model is mainly interested in long-term preservation and concerned about comprehensive content. This paper propose digital archive workflow that is used not only for long-term preservation but also for the efficient utilization with the image data. The OAIS reference model and the metadata standard should be added surge information to build a digital archive. Therefore we propose metadata model should be managed based on multi-level classes not only for effective digital image archive but also for utilization.

North Korea reported that tideland reclamation projects had been successfully constructed and/or under construction during the period of the third development scheme(1987∼1993), which were 28,400ha in 9 project areas: 8 projects along the western coast and one in the eastern coast. In this study eight projects located in western coast were investigated in order to confirm the detail of works, construction stages and difference from our project formulation methods using the topographic maps published in different years and the recent sattelite image data especially Lansat TM and SPOT PN. Intensity-hue-saturation (IHS) method was adopted to merge two sattelite data for the image enhancement of remote sensing. Construction stages of sea-dikes, land consolidation for paddy and salt pan, reservoir for irrigation and desalinization and the present land use were investigated and estimated the acreage of the development areas. The total gross project areas of 38,105 ha: 16,555 ha completed for paddy or salt pan, 16,826 ha under construction, and 4,724 ha under planning were confirmed, although the area of 27,100 ha in 8 projects were reported to be completed or ongoing on the bimonthly journal of N. Korean Trend published in 1994.

농업용 저수지는 농업용수 공급에 있어서 매우 중요한 생산기반시설로, 우리나라 농업용수의 60% 정도를 공급하고 있다. 다만, 여러 문제로 인해 농업용수의 효율적인 공급에 어려움이 발생하고 있으며, 효과적인 공급 및 관리 체계 구현을 위한 정확한 실시간 저수위 혹은 저수량 추정이 필요하다. 본 연구에서는 영상정보를 활용한 딥러닝 기반 농업용 저수지 수위 인식 모델을 제안하였다. 개발한 모델은 (1) CCTV 영상정보 자료 수집 및 분석, (2) U-Net 이미지 분할 방법을 통한 입력 자료 생성, 그리고 (3) CNN과 ResNet 모델을 통한 수위 인식 세 단계로 구성된다. 모델은 두 농업용 저수지(G저수지와 M저수지)의 영상자료와 저수위 시계열자료를 활용하여 구현하였다. 적용 결과 이미지 분할 모델의 성능은 매우 우수한 것으로 나타났으며, 수위 인식 모델의 경우 수위 분류 계급구간에 따라 성능이 상이한 것으로 나타났다. 특히 영상자료의 픽셀 변동이 클수록 정확도 80% 이상이 확보 가능한 것으로 확인되었으나, 그렇지 않은 경우, 정확도가 50% 수준인 것으로 나타났다. 본 연구에서 개발한 모델은 향후 이미지 자료가 추가로 확보될 경우, 그 활용도 및 정확도가 더 높아질 것으로 기대한다.

The agricultural reservoir is a critical water supply system in South Korea, providing approximately 60% of the agricultural water demand. However, the reservoir faces several issues that jeopardize its efficient operation and management. To address this issues, we propose a novel deep-learning-based water level recognition model that uses CCTV image data to accurately estimate water levels in agricultural reservoirs. The model consists of three main parts: (1) dataset construction, (2) image segmentation using the U-Net algorithm, and (3) CCTV-based water level recognition using either CNN or ResNet. The model has been applied to two reservoirs G-reservoir and M-reservoir with observed CCTV image and water level time series data. The results show that the performance of the image segmentation model is superior, while the performance of the water level recognition model varies from 50 to 80% depending on water level classification criteria (i.e., classification guideline) and complexity of image data (i.e., variability of the image pixels). The performance of the model can be improved if more numbers of data can be collected.

빅데이터와 사회연결망 분석기법을 이용하여 사회적 이슈인 '노인 이미지'를 분석 하였다. '노인' 키워드를 입력하여 텍스트마이닝 기법으로 추출된 단어를 분석한 결과 대중의 트렌드를 대표하는 카페, 블로그 등의 매체를 통해 본 노인 이미지는 '어르신'이라는 단어를 가장 많이 사용하고 있었다. 상위 10위 빈도수를 보인 단어를 이용하여 노인의 이미지를 표현하면, "노인은 사회의 존경을 받는 어르신이며 돈을 벌기위해 자격증을 따려고 하고 건강을 챙기며 고령에도 불구하고 100세까지 건강하게 일을 하기를 원하는 어르신"으로 정리되었다. 본 연구는 방대한 양의 데이터를 수집하여 이를 사회연결망 기법으로 분석함으로써 사회적 담론을 포함한 거시적 수준의 '노인 이미지' 분석을 통해 기존의 분석방법과 차별화하고자 하였다. 대중이 느끼는 노인에 대한 이미지가 '어르신'으로 긍정적으로 표현되는 것을 볼 때, 현재 추진하는 노인정책의 방향이 바람직한 방향으로 평가 받고 있다고 할 수 있으며, 한편으로는 그렇게 평가받기를 원하는 대중의 '욕구'를 느낄 수 있었다. 따라서 향후에 적용할 노인 정책 방향은, 노인들이 사회적 역할을 감당하여 사회에서 '필요한 존재'로 인식될 수 있도록 하는 정책이 우선되어야 한다. 또한 건강을 유지하고 활동할 수 있는 일자리 창출과 복지, 소외에 대한 대책 등의 우선순위가 반영된 노인 정책을 추진할 것을 제언하였다.

We analyzed the social issue 'image of the elderly' using Big Data and Social Network Analysis. First, we analyzed the words extracted by the text mining technique by inputting the keyword 'elderly'. As a result of analysis, the image of the elderly viewed through media such as cafes, blogs, etc. Representing the trend of the public was using the word 'Senior' the most. The image of the elderly is expressed using the word having the highest frequency in the top 10, "The elderly are 'Senior' people who are respected by society, they are organized to earn money, to earn their qualifications, to health, and to 'Seniors' who desire to work healthy up to 100 years old". The purpose of this study is to differentiate from the existing analysis method by analyzing the macro-level image of the elderly including the social discourse by collecting vast amount of data and analyzing it with the social networking technique. When the image of the elderly that the public perceives is positively expressed as 'Senior', it can be said that the direction of the current elderly policy is evaluated as a desirable direction. On the other hand, it was able to feel the 'desire' of the public who wanted to be evaluated. Therefore, the policy direction of the elderly to be applied in the future should be the policy that enables the elderly to be perceived as 'Necessary existence' in society by taking on social roles. In addition, we proposed to implement the policy of the elderly that reflects priorities such as job creation, welfare, and alienation that can activity and maintain health.

지금까지의 정보검색 연구에서 데이터 융합 기법을 이용한 문서 검색은 하나의 알고리즘에 의한 검색에 비하여 많은 경우에 효율성이 높은 결과를 얻을 수 있었다. 하지만 이미지 검색에서 상이한 알고리즘을 이용한 다수의 검색 결과를 합쳐 하나의 검색결과를 얻는 데이터 융합 기법의 사용은 많지 않았다. 이 연구에서는 소벨 연산자를 이용한 윤곽선 검출과 자기조직화 지도 알고리즘에 의한 두 검색 결과를 융합하여 각각의 알고리즘에 의한 검색결과 보다 높은 효율성을 보여주는 방법을 제시하였다. 이 연구에서는 상용 클립아트 이미지를 이용하여 사람의 주관적인 적합성 판단을 배제한 검색 실험 데이터를 만들어 사용하였다.

In many information retrieval experiments, the data fusion techniques have been used to achieve higher effectiveness in comparison to the single evidence-based retrieval. However, there had not been many image retrieval studies using the data fusion techniques especially in combining retrieval results based on multiple retrieval methods. In this paper, we describe how the image retrieval effectiveness can be improved by combining two sets of the retrieval results using the Sobel operator-based edge detection and the Self Organizing Map(SOM) algorithms. We used the clip art images from a commercial collection to develop a test data set. The main advantage of using this type of the data set was the clear cut relevance judgment, which did not require any human intervention.