• Korea Science and Art Forum
• /
• v.37 no.3
• /
• pp.125-134
• /
• 2019

This study was conducted to organize a program to classify and analyze the characteristics of images for the automation of background music selection in the video content production process. The results and contents of the study are as follows: video characteristics are selected as subject category, emotion, pixel motion speed, color, and character material. Subject categories and feelings were extracted using Microsoft's Azure Video Indexer, Pixel Movement Speed was an Optional flow, Color was an Image Histogram for Image, and character materials was CNN(Convolutional Neural Network). The results of this study are significant in that video analysis was conducted to match background music in the recent content production process of 'Internet One-person Broadcasting Creators'.

• The Journal of Korea Robotics Society
• /
• v.14 no.2
• /
• pp.87-93
• /
• 2019

This paper presents a 6-DOF relocalization using a 3D laser scanner and a monocular camera. A relocalization problem in robotics is to estimate pose of sensor when a robot revisits the area. A deep convolutional neural network (CNN) is designed to regress 6-DOF sensor pose and trained using both RGB image and 3D point cloud information in end-to-end manner. We generate the new input that consists of RGB and range information. After training step, the relocalization system results in the pose of the sensor corresponding to each input when a new input is received. However, most of cases, mobile robot navigation system has successive sensor measurements. In order to improve the localization performance, the output of CNN is used for measurements of the particle filter that smooth the trajectory. We evaluate our relocalization method on real world datasets using a mobile robot platform.

• Journal of the Society of Naval Architects of Korea
• /
• v.59 no.4
• /
• pp.243-250
• /
• 2022

In the design stage of hull forms, a fast prediction method of resistance performance is needed. In these days, large test matrix of candidate hull forms is tested using Computational Fluid Dynamics (CFD) in order to choose the best hull form before the model test. This process requires large computing times and resources. If there is a fast and reliable prediction method for hull form performance, it can be used as the first filter before applying CFD. In this paper, we suggest the offset-based performance prediction method. The hull form geometry information is applied in the form of 2D offset (non-dimensionalized by breadth and draft), and it is studied using Convolutional Neural Network (CNN) and adapted to the model test results (Residual Resistance Coefficient; C_R). Some additional variables which are not included in the offset data such as main dimensions are merged with the offset data in the process. The present model shows better performance comparing with the simple regression models.

• Journal of information and communication convergence engineering
• /
• v.22 no.2
• /
• pp.109-120
• /
• 2024

Trash or garbage is one of the most dangerous health and environmental problems that affect pollution. Pollution affects nature, human life, and wildlife. In this paper, we propose modern solutions for cleaning the environment of trash pollution by enforcing strict action against people who dump trash inappropriately on streets, outside the home, and in unnecessary places. Artificial Intelligence (AI), especially Deep Learning (DL), has been used to automate and solve issues in the world. We availed this as an excellent opportunity to develop a system that identifies trash using a deep convolutional neural network (CNN). This paper proposes a real-time garbage identification system based on a deep CNN architecture with eight distinct classes for the training dataset. After identifying the garbage, the CCTV camera captures a video of the individual placing the trash in the incorrect location and sends an alert notice to the relevant authority.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2017.11a
• /
• pp.11-12
• /
• 2017

본 논문에서는 부류가 정해진 훈련 집합이 불필요한 온라인 학습 기반 추적 기법을 제안한다. 추적기의 학습을 위해 합성곱 신경망(convolutional neural network: CNN)을 이용하였다. 추적영상으로부터 직접 훈련 샘플을 수집함으로써 분류기 학습을 위한 비용을 감소시킬 수 있었고, 목표 영상에 적응적인 객체 모델을 생성할 수 있다. 실험 결과를 통해 제안하는 방법이 우수한 성능을 보임을 입증하였다.

• 한국어정보학회:학술대회논문집
• /
• 2017.10a
• /
• pp.56-59
• /
• 2017

CNN(Convolutional Neural Network)을 이용하여 발화 주제 다중 분류 task를 multi-labeling 방법과, cluster 방법을 이용하여 수행하고, 각 방법론에 MSE(Mean Square Error), softmax cross-entropy, sigmoid cross-entropy를 적용하여 성능을 평가하였다. Network는 음절 단위로 tokenize하고, 품사정보를 각 token의 추가한 sequence와, Naver DB를 통하여 얻은 named entity 정보를 입력으로 사용한다. 실험결과 cluster 방법으로 문제를 변형하고, sigmoid를 output layer의 activation function으로 사용하고 cross entropy cost function을 이용하여 network를 학습시켰을 때 F1 0.9873으로 가장 좋은 성능을 보였다.

• Journal of Korea Multimedia Society
• /
• v.21 no.10
• /
• pp.1195-1202
• /
• 2018

Artificial neural network is widely used for its excellent performance and implementability. However, traditional neural network needs to learn the system from scratch, with the addition of new input data, the variation of the observation environment, or the change in the form of input/output data. To resolve such a problem, the technique of transfer learning has been proposed. Transfer learning constructs a newly developed target system partially updating existing system and hence provides much more efficient learning process. Until now, transfer learning is mainly studied in the field of image processing and is not yet widely employed in acoustic data processing. In this paper, focusing on the scalability of transfer learning, we apply the concept of transfer learning to the problem of guitar chord classification and evaluate its performance. For this purpose, we build a target system of convolutional neutral network (CNN) based 48 guitar chords classification system by applying the concept of transfer learning to a source system of CNN based 24 guitar chords classification system. We show that the system with transfer learning has performance similar to that of conventional system, but it requires only half the learning time.

• Journal of Korea Multimedia Society
• /
• v.22 no.1
• /
• pp.1-9
• /
• 2019

Single image rain removal is a typical inverse problem which decomposes the image into a background scene and a rain streak. Recent works have witnessed a substantial progress on the task due to the development of convolutional neural network (CNN). However, existing CNN-based approaches train the network with synthetically generated training examples. These data tend to make the network bias to the synthetic scenes. In this paper, we present an unsupervised framework for removing rain streaks from real-world rainy images. We focus on the natural phenomena that static rainy scenes capture a common background but different rain streak. From this observation, we train siamese network with the real rain image pairs, which outputs identical backgrounds from the pairs. To train our network, a real rainy dataset is constructed via web-crawling. We show that our unsupervised framework outperforms the recent CNN-based approaches, which are trained by supervised manner. Experimental results demonstrate that the effectiveness of our framework on both synthetic and real-world datasets, showing improved performance over previous approaches.

• The Journal of the Acoustical Society of Korea
• /
• v.40 no.2
• /
• pp.130-138
• /
• 2021

This paper presents an effective structure by applying various normalization to Convolutional Neural Networks (CNN) for seismic event classification. Normalization techniques can not only improve the learning speed of neural networks, but also show robustness to noise. In this paper, we analyze the effect of input data normalization and hidden layer normalization on the deep learning model for seismic event classification. In addition an effective model is derived through various experiments according to the structure of the applied hidden layer. As a result of various experiments, the model that applied input data normalization and weight normalization to the first hidden layer showed the most stable performance improvement.

• Journal of the Korea Computer Graphics Society
• /
• v.26 no.2
• /
• pp.11-19
• /
• 2020

When a character's pose changes, its center of mass(COM) also changes. The change of COM has distinctive patterns corresponding to various motion types like walking, running or sitting. Thus the motion type can be predicted by using COM movement. We propose a motion generator that uses character's center of mass information. This generator can generate various motions without annotated action type labels. Thus dataset for training and running can be generated full-automatically. Our neural network model takes the motion history of the character and its center of mass information as inputs and generates a full-body pose for the current frame, and is trained using simple Convolutional Neural Network(CNN) that performs 1D convolution to deal with time-series motion data.