• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2021.06a
• /
• pp.348-351
• /
• 2021

최근 자동화의 필요성이 증가함에 따라 얼굴 표정 인식 분야(face expression recognition)가 인공지능과 이미지 처리 분야에서 활발히 연구되고 있다. 본 논문에서는 기존 인공신경망에서 요구되었던 고성능 GPU 환경과 높은 연산량을 극복하고자 모델 경량화(Light weighted Model) 기법을 적용하여 드론 및 모바일 기기에서 적용될 수 있는 얼굴 표정 인식 신경망을 제안한다. 제안하는 방법은 미세한 얼굴의 표정 인식을 위한 방법으로, 입력 이미지의 receptive field 를 늘려 특징 맵의 표현력을 높이는 방법을 제안한다. 또한 효과적인 신경망의 경량화를 위하여, 파라미터의 연산량을 줄일 때 발생하는 문제점을 극복하기 위한 방법을 제시한다. 따라서 제안하는 네트워크를 적용하면 많은 연산량과 느린 연산속도로 인해 제한되었던 네트워크 환경을 극복할 수 있을 뿐만 아니라, UAV(Unmanned Aerial Vehicle, 무인항공기) 및 모바일 기기에서 신경망을 이용한 실시간 얼굴 표정 인식을 할 수 있다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.05a
• /
• pp.79-80
• /
• 2023

CPU 와 GPU 의 성능이 지속적으로 발전함에 따라 객체 인식 인공지능의 정확도와 추론 속도는 점차 향상되고 있으나 이러한 성능을 Edge Device 와 같은 제한된 환경에서 구현하기에 아직 여러 한계점이 존재한다. 본 논문에서는 여러가지 Edge Device 에서 객체 인식을 위한 경량화 된 모델 중 하나인 SSD-Mobilenet-V2 를 활용하여 결과값을 통해 각 Device 간 경향성을 분석하였다. 본 결과를 바탕으로 다양한 환경에서의 객체인식 인공지능 모델의 성능 개선을 위한 연구에 활용할 수 있다.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.22 no.10
• /
• pp.1300-1306
• /
• 2018

The purpose of this study is to see possibility of Char2Vec as alternative of Word2Vec that most famous word embedding model in Sentence Similarity Measure Problem by Deep-Learning. In experiment, we used the Siamese Ma-LSTM recurrent neural network architecture for measure similarity two random sentences. Siamese Ma-LSTM model was implemented with tensorflow. We train each model with 200 epoch on gpu environment and it took about 20 hours. Then we compared Word2Vec based model training result with Char2Vec based model training result. as a result, model of based with Char2Vec that initialized random weight record 75.1% validation dataset accuracy and model of based with Word2Vec that pretrained with 3 million words and phrase record 71.6% validation dataset accuracy. so Char2Vec is suitable alternate of Word2Vec to optimize high system memory requirements problem.

• Transactions of the Korean Society of Mechanical Engineers
• /
• v.16 no.4
• /
• pp.787-794
• /
• 1992

To predict performances of Stirling Engines, a second-order analysis method has been developed. The present method which is based on the approximate analytical solution to the Ideal Adiabatic Model includes major loss mechanisms due to finite heat transfer and flow friction. Comparison of calculated results with previously reported study for a specific engine shows reasonable agreements and a possibility of being used for basic designs. Also, predicted performances with repect to engine speeds are consistent with experimental data in trend. To improve the prediction capability of this method, it is needed that not only additional losses should be taken into account, but also fundamental characteristics of oscillating flow and heat transfer should be better understood.

• Journal of IKEEE
• /
• v.20 no.3
• /
• pp.303-306
• /
• 2016

In this paper, we propose a method to accelerate convolutional neural network by utilizing a GPGPU. Convolutional neural network is a sort of the neural network learning features of images. Convolutional neural network is suitable for the image processing required to learn a lot of data such as images. The convolutional layer of the conventional CNN required a large number of multiplications and it is difficult to operate in the real-time on the embedded environment. In this paper, we reduce the number of multiplications through Winograd convolution operation and perform parallel processing of the convolution by utilizing SIMT-based GPGPU. The experiment was conducted using ModelSim and TestDrive, and the experimental results showed that the processing time was improved by about 17%, compared to the conventional convolution.

• 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.

• 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.

• Journal of the Korea Computer Graphics Society
• /
• v.22 no.3
• /
• pp.21-29
• /
• 2016

Cell segmentation is an important but time-consuming and laborious task in biological image analysis. An automated, robust, and fast method is required to overcome such burdensome processes. These needs are, however, challenging due to various cell shapes, intensity, and incomplete boundaries. A precise cell segmentation will allow to making a pathological diagnosis of tissue samples. A vast body of literature exists on cell segmentation in microscopy images [1]. The majority of existing work is based on input images and predefined feature models only - for example, using a deformable model to extract edge boundaries in the image. Only a handful of recent methods employ data-driven approaches, such as supervised learning. In this paper, we propose a novel data-driven cell segmentation algorithm for bright-field microscopy images. The proposed method minimizes an energy formula defined by two dictionaries - one is for input images and the other is for their manual segmentation results - and a common sparse code, which aims to find the pixel-level classification by deploying the learned dictionaries on new images. In contrast to deformable models, we do not need to know a prior knowledge of objects. We also employed convolutional sparse coding and Alternating Direction of Multiplier Method (ADMM) for fast dictionary learning and energy minimization. Unlike an existing method [1], our method trains both dictionaries concurrently, and is implemented using the GPU device for faster performance.

• Journal of the Korean Institute of Gas
• /
• v.22 no.6
• /
• pp.136-143
• /
• 2018

Cellular automata have been applied to simulations in many fields such as astrophysics, social phenomena, fire spread, and evacuation. Using cellular automata, this study develops a model for consequence analysis of the dispersion of hazardous chemicals, which is required for risk assessments of and emergency responses for frequent chemical accidents. Unlike in cases of detailed plant safety design, real-time accident responses require fast and iterative calculations to reduce the uncertainty of the distribution of damage within the affected area. EPA ALOHA and KORA of National Institute of Chemical Safety have been popular choices for these analyses. However, this study proposes an initiative to supplement the model and code continuously and is different in its development of free software, specialized for small and medium enterprises. Compared to the full-scale computational fluid dynamics (CFD), which requires large amounts of computation time, the relative accuracy loss is compromised, and the convenience of the general user is improved. Using Python open-source libraries as well as meteorological information linkage, it is made possible to expand and update the functions continuously. Users can easily obtain the results by simply inputting the layout of the plant and the materials used. Accuracy is verified against full-scale CFD simulations, and it will be distributed as open source software, supporting GPU-accelerated computing for fast computation.

• Smart Media Journal
• /
• v.12 no.11
• /
• pp.57-66
• /
• 2023

In this paper, we present a study on an AI-based system utilizing the CCTV system within city buses to predict the intentions of boarding and alighting passengers, with the aim of preventing accidents. The proposed system employs the YOLOv7 Pose model to detect passengers, while utilizing an LSTM model to predict intentions of tracked passengers. The system can be installed on the bus's CCTV terminals, allowing for real-time visual confirmation of passengers' intentions throughout driving. It also provides alerts to the driver, mitigating potential accidents during passenger transitions. Test results show accuracy rates of 0.81 for analyzing boarding intentions and 0.79 for predicting alighting intentions onboard. To ensure real-time performance, we verified that a minimum of 5 frames per second analysis is achievable in a GPU environment. his algorithm enhance the safety of passenger transitions during bus operations. In the future, with improved hardware specifications and abundant data collection, the system's expansion into various safety-related metrics is promising. This algorithm is anticipated to play a pivotal role in ensuring safety when autonomous driving becomes commercialized. Additionally, its applicability could extend to other modes of public transportation, such as subways and all forms of mass transit, contributing to the overall safety of public transportation systems.