• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.14 no.3
• /
• pp.201-207
• /
• 2021

Deep learning, a type of machine learning, performs learning while changing the weights as it progresses through each learning process. Tensor Flow and Keras provide the results of the end of the learning in graph form. Thus, If an error occurs, the result must be discarded. Consequently, existing technologies provide a function to roll back learning results, but the rollback function is limited to results up to five times. Moreover, they applied the concept of MLOps to track the deep learning process, but no rollback capability is provided. In this paper, we construct a system that manages the intermediate value of the learning process by blockchain to record the intermediate learning process and can rollback in the event of an error. To perform the functions of blockchain, the deep learning process and the rollback of learning results are designed to work by writing Smart Contracts. Performance evaluation shows that, when evaluating the rollback function of the existing deep learning method, the proposed method has a 100% recovery rate, compared to the existing technique, which reduces the recovery rate after 6 times, down to 10% when 50 times. In addition, when using Smart Contract in Ethereum blockchain, it is confirmed that 1.57 million won is continuously consumed per block creation.

• Journal of the Korea Society of Computer and Information
• /
• v.24 no.3
• /
• pp.87-93
• /
• 2019

In this paper, we propose a real-time drone-based violent protest detection system. Our proposed system uses drones to detect scenes of violent protest in real-time. The important problem is that the victims and violent actions have to be manually searched in videos when the evidence has been collected. Firstly, we focused to solve the limitations of existing collecting evidence devices by using drone to collect evidence live and upload in AWS(Amazon Web Service)[1]. Secondly, we built a Deep Learning based violence detection model from the videos using Yolov3 Feature Pyramid Network for human activity recognition, in order to detect three types of violent action. The built model classifies people with possession of gun, swinging pipe, and violent activity with the accuracy of 92, 91 and 80.5% respectively. This system is expected to significantly save time and human resource of the existing collecting evidence.

• Journal of Korea Multimedia Society
• /
• v.25 no.10
• /
• pp.1433-1447
• /
• 2022

Recently, as the demand for physical security technology to prevent leakage of technical and business information of companies and public institutions increases, the high tech companies are operating X-ray security checkpoints at building entrances to protect their intellectual property and technology. X-ray security checkpoints are operated to detect cameras and storage media that may store or leak important technologies in the bags of people entering and leaving the building. In this study, we propose an X-ray security checkpoint system that automatically detects a storage medium in an X-ray image using a deep learning based object detection method. The proposed system consists of an edge computing unit and a cloud-computing unit. We employ the RetinaNet for automatic storage media detection in the X-ray security checkpoint images. The proposed approach achieved mAP of 95.92% on private dataset.

• Journal of Korea Multimedia Society
• /
• v.25 no.8
• /
• pp.1049-1058
• /
• 2022

HAR(Human Action Recognition) such as anomaly and object detection has become a trend in research field(s) that focus on utilizing Artificial Intelligence (AI) methods to analyze patterns of human action in crime-ridden area(s), media services, and industrial facilities. Especially, in real-time system(s) using video streaming data, HAR has become a more important AI-based research field in application development and many different research fields using HAR have currently been developed and improved. In this paper, we propose and analyze a deep-learning-based HAR that provides more efficient scheme(s) using an intelligent AI models, such system can be applied to media services using RGB video streaming data usage without feature extraction pre-processing. For the method, we adopt Slowfast based on the Deep Neural Network(DNN) model under an open dataset(HMDB-51 or UCF101) for improvement in prediction accuracy.

• Smart Structures and Systems
• /
• v.29 no.6
• /
• pp.777-783
• /
• 2022

Even though several deep learning-based methods have been applied in the field of acoustic source localization, the previous works have only been conducted using the two-dimensional representation of the beamforming maps, particularly with the planar array system. While the acoustic sources are more required to be localized in a spherical microphone array system considering that we live and hear in the 3D world, the conventional 2D equirectangular map of the spherical beamforming map is highly vulnerable to the distortion that occurs when the 3D map is projected to the 2D space. In this study, a 3D deep learning approach is proposed to fulfill accurate source localization via distortion-free 3D representation. A target function is first proposed to obtain 3D source distribution maps that can represent multiple sources' positional and strength information. While the proposed target map expands the source localization task into a point-wise prediction task, a PointNet-based deep neural network is developed to precisely estimate the multiple sources' positions and strength information. While the proposed model's localization performance is evaluated, it is shown that the proposed method can achieve improved localization results from both quantitative and qualitative perspectives.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.8
• /
• pp.61-68
• /
• 2022

In this paper, we propose a system to predict the GPS trajectory of a pedestrian based on a deep learning model. Pedestrian trajectory prediction is a study that can prevent pedestrian danger and collision situations through notifications, and has an impact on business such as various marketing. In addition, it can be used not only for pedestrians but also for path prediction of unmanned transportation, which is receiving a lot of spotlight. Among various trajectory prediction methods, this paper is a study of trajectory prediction using GPS data. It is a deep learning model-based study that predicts the next route by learning the GPS trajectory of pedestrians, which is time series data. In this paper, we presented a data set construction method that allows the deep learning model to learn the GPS route of pedestrians, and proposes a trajectory prediction deep learning model that does not have large restrictions on the prediction range. The parameters suitable for the trajectory prediction deep learning model of this study are presented, and the model's test performance are presented.

• Journal of KIISE
• /
• v.42 no.11
• /
• pp.1452-1458
• /
• 2015

Korean Morphological Analysis is a difficult process. Because Korean is an agglutinative language, one of the most important processes in Morphological Analysis is Morpheme Recovery. There are some methods using Heuristic rules and Pre-Analyzed Partial Words that were examined for this process. These methods have performance limits as a result of not using contextual information. In this study, we built a Korean morpheme recovery system using deep learning, and this system used word embedding for the utilization of contextual information. In '들/VV' and '듣/VV' morpheme recovery, the system showed 97.97% accuracy, a better performance than with SVM(Support Vector Machine) which showed 96.22% accuracy.

• Smart Media Journal
• /
• v.9 no.1
• /
• pp.45-50
• /
• 2020

With the development of deep learning, studies using artificial neural networks based on deep learning in recommendation systems are being actively conducted. Especially, the recommendation system based on RNN (Recurrent Neural Network) shows good performance because it considers the sequential characteristics of data. This study proposes a travel route recommendation system using GRU(Gated Recurrent Unit) and Session-based Parallel Mini-batch which are RNN-based algorithm. This study improved the recommendation performance through an ensemble of top1 and bpr(Bayesian personalized ranking) error functions. In addition, it was confirmed that the RNN-based recommendation system considering the sequential characteristics in the data makes a recommendation reflecting the meaning of the travel destination inherent in the travel route.

• International journal of advanced smart convergence
• /
• v.10 no.4
• /
• pp.110-116
• /
• 2021

This paper proposed a real-time earlobe detection system using deep learning on the web. Existing deep learning-based detection methods often find independent objects such as cars, mugs, cats, and people. We proposed a way to receive an image through the camera of the user device in a web environment and detect the earlobe on the server. First, we took a picture of the user's face with the user's device camera on the web so that the user's ears were visible. After that, we sent the photographed user's face to the server to find the earlobe. Based on the detected results, we printed an earring model on the user's earlobe on the web. We trained an existing YOLO v5 model using a dataset of about 200 that created a bounding box on the earlobe. We estimated the position of the earlobe through a trained deep learning model. Through this process, we proposed a real-time earlobe detection system on the web. The proposed method showed the performance of detecting earlobes in real-time and loading 3D models from the web in real-time.

• Korean Journal of Artificial Intelligence
• /
• v.11 no.4
• /
• pp.21-27
• /
• 2023

This paper discusses a deep learning-based road surface analysis system that collects data by installing vibration sensors on the 4-axis wheel bearings of a vehicle, analyzes the data, and appropriately classifies the characteristics of the current driving road surface for use in the vehicle's control system. The data used for road surface analysis is real-time large-capacity data, with 48K samples per second, and the A2B protocol, which is used for large-capacity real-time data communication in modern vehicles, was used to collect the data. CAN and CAN-FD commonly used in vehicle communication, are unable to perform real-time road surface analysis due to bandwidth limitations. By using A2B communication, data was collected at a maximum bandwidth for real-time analysis, requiring a minimum of 24K samples/sec for evaluation. Based on the data collected for real-time analysis, performance was assessed using deep learning models such as LSTM, GRU, and RNN. The results showed similar road surface classification performance across all models. It was also observed that the quality of data used during the training process had an impact on the performance of each model.