• IEMEK Journal of Embedded Systems and Applications
• /
• v.16 no.5
• /
• pp.179-186
• /
• 2021

For stereo matching based on deep learning, the design of network structure is crucial to the calculation of matching cost, and the time-consuming problem of convolutional neural network in image processing also needs to be solved urgently. In this paper, a method of stereo matching using sparse loss volume in parallax dimension is proposed. A sparse 3D loss volume is constructed by using a wide step length translation of the right view feature map, which reduces the video memory and computing resources required by the 3D convolution module by several times. In order to improve the accuracy of the algorithm, the nonlinear up-sampling of the matching loss in the parallax dimension is carried out by using the method of multi-category output, and the training model is combined with two kinds of loss functions. Compared with the benchmark algorithm, the proposed algorithm not only improves the accuracy but also shortens the running time by about 30%.

• Journal of Korea Multimedia Society
• /
• v.21 no.12
• /
• pp.1459-1466
• /
• 2018

Robots which detect human and recognize action are important factors for human interaction, and many researches have been conducted. Recently, deep learning technology has developed and learning based robot's technology is a major research area. These studies require a database to learn and evaluate for intelligent human perception. In this paper, we propose a multi-modal sensor-based image database condition considering the security task by analyzing the image database to detect the person in the outdoor environment and to recognize the behavior during the running of the robot.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2022.10a
• /
• pp.168-169
• /
• 2022

As the electric vehicle market has grown rapidly, the battery market has grown exponentially. Due to the gap between the generation speed of quality control technology and battery mass production speed for batteries mounted on electric vehicles, many durability problems have arisen for batteries. Most accidents are caused by electrical factors, but there is no technology to quickly inspect them. In this paper, we are going to propose a quick analysis of battery cell defects using the GAN model.

• Nuclear Engineering and Technology
• /
• v.56 no.10
• /
• pp.4463-4469
• /
• 2024

This paper details a computational framework to produce automated, graphical workflows, and how this framework can be deployed to support complex modeling problems like those in nuclear engineering. Key benefits of the framework include: automating previously manual workflows; intuitive construction and communication of workflows through a graphical interface; and automated file transfer and handling for workflows deployed across heterogeneous computing resources. This paper demonstrates the framework's application to probabilistic post-closure performance assessment of systems for deep geologic disposal of nuclear waste. However, the framework is a general capability that can help users running a variety of computational studies.

• Journal of Biomedical Engineering Research
• /
• v.39 no.6
• /
• pp.229-236
• /
• 2018

Liver cancer is one of the highest incidents in the world, and the mortality rate is the second most common disease after lung cancer. The purpose of this study is to evaluate the diagnostic ability of deep learning in the classification of malignant and benign tumors in CT images of patients with liver tumors. We also tried to identify the best data processing methods and deep learning models for classifying malignant and benign tumors in the liver. In this study, CT data were collected from 92 patients (benign liver tumors: 44, malignant liver tumors: 48) at the Gil Medical Center. The CT data of each patient were used for cross-sectional images of 3,024 liver tumors. In AlexNet and VggNet, the average of the overall accuracy at each image size was calculated: the average of the overall accuracy of the $200{\times}200$ image size is 69.58% (AlexNet), 69.4% (VggNet), $150{\times}150$ image size is 71.54%, 67%, $100{\times}100$ image size is 68.79%, 66.2%. In conclusion, the overall accuracy of each does not exceed 80%, so it does not have a high level of accuracy. In addition, the average accuracy in benign was 90.3% and the accuracy in malignant was 46.2%, which is a significant difference between benign and malignant. Also, the time it takes for AlexNet to learn is about 1.6 times faster than VggNet but statistically no different (p > 0.05). Since both models are less than 90% of the overall accuracy, more research and development are needed, such as learning the liver tumor data using a new model, or the process of pre-processing the data images in other methods. In the future, it will be useful to use specialists for image reading using deep learning.

• Journal of the Institute of Convergence Signal Processing
• /
• v.22 no.4
• /
• pp.149-155
• /
• 2021

In this study, we intend to develop a defective road surface object recognition model that automatically detects road surface defects that restrict the movement of the transportation handicapped using electric mobile devices with deep learning. For this purpose, road surface information was collected from the pedestrian and running routes where the electric mobility aid device is expected to move in five areas within the city of Busan. For data, images were collected by dividing the road surface and surroundings into objects constituting the surroundings. A series of recognition items such as the detection of breakage levels of sidewalk blocks were defined by classifying according to the degree of impeding the movement of the transportation handicapped in traffic from the collected data. A road surface object recognition deep learning model was implemented. In the final stage of the study, the performance verification process of a deep learning model that automatically detects defective road surface objects through model learning and validation after processing, refining, and annotation of image data separated and collected in units of objects through actual driving. proceeded.

• International Journal of Naval Architecture and Ocean Engineering
• /
• v.5 no.1
• /
• pp.161-177
• /
• 2013

High-speed vessels require good resistance and seakeeping performance for safe operations in rough seas. The resistance and seakeeping performance of high-speed vessels varies significantly depending on their hull forms. In this study, three planing hulls that have almost the same displacement and principal dimension are designed and the hydrodynamic characteristics of those hulls are estimated by high-speed model tests. All model ships are deep-V type planing hulls. The bows of no.2 and no.3 model ships are designed to be advantageous for wave-piercing in rough water. No.2 and no.3 model ships have concave and straight forebody cross-sections, respectively. And length-to-beam ratios of no.2 and no.3 models are larger than that of no.1 model. In calm water tests, running attitude and resistance of model ships are measured at various speeds. And motion tests in regular waves are performed to measure the heave and pitch motion responses of the model ships. The required power of no.1 (VPS) model is smallest, but its vertical motion amplitudes in waves are the largest. No.2 (VWC) model shows the smallest motion amplitudes in waves, but needs the greatest power at high speed. The resistance and seakeeping performance of no.3 (VWS) model ship are the middle of three model ships, respectively. And in regular waves, no.1 model ship experiences 'fly over' phenomena around its resonant frequency. Vertical accelerations at specific locations such as F.P., center of gravity of model ships are measured at their resonant frequency. It is necessary to measure accelerations by accelerometers or other devices in model tests for the accurate prediction of vertical accelerations in real ships.

• Journal of Digital Convergence
• /
• v.15 no.3
• /
• pp.221-228
• /
• 2017

With the recent development of IoT technology, medical services using IoT technology are increasing in many medical institutions providing health care services. However, as the number of IoT sensors attached to the user body increases, the healthcare information transmitted to the server becomes complicated, thereby increasing the time required for analyzing the user's healthcare information in the server. In this paper, we propose a deep learning based health care information management method to collect and process healthcare information in a server for a large amount of healthcare information delivered through a user - attached IoT device. The proposed scheme constructs a subnet according to the attribute value by assigning an attribute value to the healthcare information transmitted to the server, and extracts the association information between the subnets as a seed and groups them into a hierarchical structure. The server extracts optimized information that can improve the observation speed and accuracy of user's treatment and prescription by using deep running of grouped healthcare information. As a result of the performance evaluation, the proposed method shows that the processing speed of the medical service operated in the healthcare service model is improved by 14.1% on average and the server overhead is 6.7% lower than the conventional technique. The accuracy of healthcare information extraction was 10.1% higher than the conventional method.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.39 no.3
• /
• pp.141-148
• /
• 2021

This paper presents a method to detect zebra-crossing using deep learning which combines SegNet and ResNet. For the blind, a safe crossing system is important to know exactly where the zebra-crossings are. Zebra-crossing detection by deep learning can be a good solution to this problem and robotic vision-based assistive technologies sprung up over the past few years, which focused on specific scene objects using monocular detectors. These traditional methods have achieved significant results with relatively long processing times, and enhanced the zebra-crossing perception to a large extent. However, running all detectors jointly incurs a long latency and becomes computationally prohibitive on wearable embedded systems. In this paper, we propose a model for fast and stable segmentation of zebra-crossing from captured images. The model is improved based on a combination of SegNet and ResNet and consists of three steps. First, the input image is subsampled to extract image features and the convolutional neural network of ResNet is modified to make it the new encoder. Second, through the SegNet original up-sampling network, the abstract features are restored to the original image size. Finally, the method classifies all pixels and calculates the accuracy of each pixel. The experimental results prove the efficiency of the modified semantic segmentation algorithm with a relatively high computing speed.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2022.10a
• /
• pp.360-363
• /
• 2022

According to the 2018 Kaggle ML & DS Survey, among the proportions of frameworks for machine learning and data science, TensorFlow and Keras each account for 41.82%. It was found to be 34.09%, and in the case of development programming, it is confirmed that about 82% use Python. A significant number of machine learning and deep learning structures utilize the Keras framework and Python, but in the case of Python, distribution and execution are limited to the Python script environment due to the script language, so it is judged that it is difficult to operate in various environments. This paper implemented a machine learning and deep learning system using C# and Keras running in Visual Studio 2019. Using the Mnist dataset, 100 tests were performed in Python 3.8,2 and C# .NET 5.0 environments, and the minimum time for Python was 1.86 seconds, the maximum time was 2.38 seconds, and the average time was 1.98 seconds. Time 1.78 seconds, maximum time 2.11 seconds, average time 1.85 seconds, total time 37.02 seconds. As a result of the experiment, the performance of C# improved by about 6% compared to Python, and it is expected that the utilization will be high because executable files can be extracted.