• Journal of the Semiconductor & Display Technology
• /
• v.20 no.3
• /
• pp.6-10
• /
• 2021

Recognition of characters printed or marked on the PCB components from images captured using cameras is an important task in PCB components inspection systems. Previous optical character recognition (OCR) of PCB components typically consists of two stages: character segmentation and classification of each segmented character. However, character segmentation often fails due to corrupted characters, low image contrast, etc. Thus, OCR without character segmentation is desirable and increasingly used via deep neural networks. Typical implementation based on deep neural nets without character segmentation includes convolutional neural network followed by recurrent neural network (RNN). However, one disadvantage of this approach is slow execution due to RNN layers. LPRNet is a segmentation-free character recognition network with excellent accuracy proved in license plate recognition. LPRNet uses a wide convolution instead of RNN, thus enabling fast inference. In this paper, LPRNet was adapted for recognizing characters printed on PCB components with fast execution and high accuracy. Initial training with synthetic images followed by fine-tuning on real text images yielded accurate recognition. This net can be further optimized on Intel CPU using OpenVINO tool kit. The optimized version of the network can be run in real-time faster than even GPU.

• Journal of the Korea Computer Graphics Society
• /
• v.27 no.3
• /
• pp.25-29
• /
• 2021

초박형 라이트필드 카메라 시스템은 이미지 센서 위에 렌즈 어레이를 부착하는 방식으로 만들어진다. 이러한 초박형 라이트필드 카메라는 하나의 이미지 센서를 여러 개의 sub-aperture가 나눠쓰는 방식으로 되어있어 개별 이미지의 분해능이 낮으며, sub-aperture 이미지들을 융합해 추가적인 분해능 향상이 수행되어야 한다. 본 연구에서는 초박형 라이트필드 카메라 시스템을 개발했으며, 개발된 카메라 시스템을 위한 실시간 분해능 향상 알고리즘을 개발, 실험을 통해 검증했다. 개발된 초박형 라이트필드 카메라는 두께 2mm, 24개(6×4)의 551×551 해상도의 sub-aperture로 구성되어 있으며, 임베디드 컴퓨팅 보드를 사용해 휴대가 가능하도록 제작되었다. 실시간 분해능 향상 알고리즘은 임베디드 컴퓨팅 보드의 GPU에서 병렬처리를 통해 라플라시안 피라미드 기반의 이미지 융합 알고리즘을 수행한다. 실험을 통해 검증한 결과로, 개발 시스템은 MTF50값이 평균 35% 정도 개선되었으며, 10.65fps의 처리속도로 실시간 처리가 가능함을 확인했다.

• Journal of KIBIM
• /
• v.9 no.4
• /
• pp.62-74
• /
• 2019

Because the existing BIM platform is based on user services, the focus is on the development of SaaS (Software as a Service), which provides business services online. However, since a harbor is a security facility, the harbor BIM platform is preferably provided in a private form, rather than relying on the infrastructure environment provided by external cloud providers. Therefore, this study analyzes and reviews the main functions to be provided as SaaS services of the harbor BIM platform. The goal is to build a cloud-based harbor BIM platform that can provide this service to users. To this end, we built IaaS (Infrastructure as a Service) environment of the harbor BIM platform based on the open source Open Stack and integrate and develop PaaS environment with Open Shift applied with IaaS. We applied the GPU to the harbor BIM platform to verify the performance of the harbor BIM platform, and found that the rendering and loading times are improved. In particular, it is expected to reduce the cost of introduction and provide it as the basic cloud environment of similar BIM platform for infrastructure facilities.

• Journal of Korea Multimedia Society
• /
• v.25 no.3
• /
• pp.461-467
• /
• 2022

This research is to raise awareness of food safety by designing and supporting a hazard food information notification platform for consumers. To this end, the design was carried out by dividing the process into a data extraction process, an application screen design process, and a CNN-based food inference process. Data was collected through public data APIs and crawling, and it was sent to each activity screen designed for Android studios so that it could be output. As a result, when the platform is executed, information on hazardous food names, registration dates, food classification, manufacturing dates, recovery grades, recovery reasons, recovery methods, company names, barcode numbers, and packaging units can be intuitively and conveniently checked. In addition, CNN-based food inference processes allowed mobile cameras to infer harmful food and applied various quantization techniques such as Dynamic Range, Integer, and Float16 to compare the degree of improvement in inference performance. As a result, the group that applied basic quantization and treated device resources with GPU showed the greatest improvement in inference performance. Through this platform, it is expected that the reliability of food safety will be improved by making it more convenient for consumers to recognize food risks.

• Nuclear Engineering and Technology
• /
• v.53 no.3
• /
• pp.752-762
• /
• 2021

The Smoothed Particle Hydrodynamics is one of the most widely used mesh-free numerical method for thermo-fluid dynamics. Due to its Lagrangian nature and simplicity, it is recently gaining popularity in simulating complex physics with large deformations. In this study, the 3D single/two-phase numerical simulations are performed on the Liquid Metal Reactor (LMR) centralized sloshing benchmark experiment using the SPH parallelized using a GPU. In order to capture multi-phase flows with a large density ratio more effectively, the original SPH density and continuity equations are re-formulated in terms of the normalized-density. Based upon this approach, maximum sloshing height and arrival time in various experimental cases are calculated by using both single-phase and multi-phase SPH framework and the results are compared with the benchmark results. Overall, the results of SPH simulations show excellent agreement with all the benchmark experiments both in qualitative and quantitative manners. According to the sensitivity study of the particle-size, the prediction accuracy is gradually increasing with decreasing the particle-size leading to a higher resolution. In addition, it is found that the multi-phase SPH model considering both liquid and air provides a better prediction on the experimental results and the reality.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.22 no.3
• /
• pp.69-74
• /
• 2022

SNN (Spiking Neural Network) operating in neuromorphic architecture was created by mimicking human neural networks. Neuromorphic computing based on neuromorphic architecture requires relatively lower power than typical deep learning techniques based on GPUs. For this reason, research to support various artificial intelligence models using neuromorphic architecture is actively taking place. This paper conducted a performance analysis of the speech recognition model based on neuromorphic architecture according to the speech data preprocessing technique. As a result of the experiment, it showed up to 84% of speech recognition accuracy performance when preprocessing speech data using the Fourier transform. Therefore, it was confirmed that the speech recognition service based on the neuromorphic architecture can be effectively utilized.

• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.427-430
• /
• 2022

추천시스템에서 콜드스타트 문제를 해결하기 위해 다양한 연구들이 진행되고 있다. 하지만, 대부분의 연구는 아직도 사용자 기반의 히스토리 데이터셋을 반드시 필요로 하여, 콜드스타트 문제를 완벽히 해결하지 못하고 있다. 이에 본 논문은 콜드스타트 문제를 완화할 수 있는 기저속성 기반의 추천시스템을 제안한다. 제안하는 방법론을 검증하기 위해, 직접 수집한 한국어 영화 리뷰 데이터셋을 기반으로 성능을 검증하였으며, 평가 결과 제안한 방법론이 키워드와 사용자의 리뷰 점수를 효과적으로 반영한 추천시스템임을 확인할 수 있었고, 데이터 희소성 및 콜드스타트 문제를 완화하여 기존의 텍스트 기반 랭킹 시스템의 성능을 압도하는 것을 확인하였다. 더 나아가 제안된 기저속성 추천시스템은 추론 시에 GPU 컴퓨팅 자원을 요구하지 않기에 서비스 측면에서도 많은 이점이 있음을 확인하였다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.06a
• /
• pp.67-70
• /
• 2022

최근 딥러닝 방법의 발전하면서 영상처리 및 컴퓨터 비전의 다양한 분야에서 딥러닝 기반의 알고리즘들이 그 이전의 방법들에 비하여 큰 성능 향상을 보이고 있다. 손실 영상 압축의 경우 최근 encoder-decoder 형태의 네트웍이 영상 압축에서 사용되는 transform을 대체하고 있고, transform 결과들의 엔트로피 코딩을 위한 추가적인 encoder-decoder 네트웍을 사용하여 HEVC 수준에 버금가는 성능을 내고 있다. 무손실 압축의 경우에도 매 픽셀 예측을 CNN으로 수행하는 경우, 기존의 예측방법들에 비하여 예측성능이 크게 향상되어 JPEG-2000 Lossless, FLIF, JEPG-XL 등의 딥러닝을 사용하지 않는 방법들에 비하여 우수한 성능을 내는 것으로 보고되고 있다. 그러나 모든 픽셀에 대하여 예측값을 CNN을 통하여 계산하는 방법은, 영상의 픽셀 수 만큼 CNN을 수행해야 하므로 HD 크기 영상에 대하여 지금까지 알려진 가장 빠른 방법이 한 시간 이상 소요되는 등 비현실적인 것으로 알려져 있다. 따라서 최근에는 성능은 이보다 떨어지지만 속도를 현실적으로 줄인 방법들이 제안되고 있다. 이러한 방법들은 초기에는 FLIF나 JPEG-XL에 비하여 성능이 떨어져서, GPU를 사용하면서도 기존의 방법보다 좋지 않은 성능을 보인다는 면에서 여전히 비현실적이었다. 최근에는 신호의 특성을 더 잘 활용하는 방법들이 제안되면서 매 픽셀마다 CNN을 수행하는 방법보다는 성능이 떨어지지만, 짧은 시간 내에 FLIF나 JPEG-XL보다는 좋은 성능을 내는 현실적인 방법들이 제안되었다. 본 연구에서는 이러한 최근의 몇 가지 방법들을 살펴보고 이들보다 성능을 더 좋게 할 수 있는 보조적인 방법들과 raw image에 대한 성능을 평가한다.

• Nuclear Engineering and Technology
• /
• v.54 no.11
• /
• pp.4310-4321
• /
• 2022

The concept of dense granular-flow target (DGT) for the China Initiative Accelerator Driven Subcritical system (CiADS) is an attractive choice for high heat removal ability, low chemical toxicity, and radiotoxicity. A wobbling hollow beam is proposed to enhance the homogeneity of temperature rise of flowing particles in beam-target coupling zone. In this paper, the design procedure of target and beam parameters was discussed firstly. Then we simulated the heat deposition and transfer of the scanning beam in DGT to study the effect of beam parameters. The results show the flux density of proton beam plays a crucial role in the distribution of temperature rise while the contributions from scanning frequency heat transfer are also obvious. Moreover, heat transfer in transversal directions is insignificant, resulting in a low heat flux towards the sidewalls of DGT. This work not only contributes to the design of DGT, but also beneficial for understanding the beam-target coupling in porous materials.

• ETRI Journal
• /
• v.44 no.1
• /
• pp.24-37
• /
• 2022

Virtual view synthesis, which generates novel views similar to the characteristics of actually acquired images, is an essential technical component for delivering an immersive video with realistic binocular disparity and smooth motion parallax. This is typically achieved in sequence by warping the given images to the designated viewing position, blending warped images, and filling the remaining holes. When considering 6DoF use cases with huge motion, the warping method in patch unit is more preferable than other conventional methods running in pixel unit. Regarding the prior case, the quality of synthesized image is highly relevant to the means of blending. Based on such aspect, we proposed a novel blending architecture that exploits the similarity of the directions of rays and the distribution of depth values. By further employing the proposed method, results showed that more enhanced view was synthesized compared with the well-designed synthesizers used within moving picture expert group (MPEG-I). Moreover, we explained the GPU-based implementation synthesizing and rendering views in the level of real time by considering the applicability for immersive video service.