• Korean Journal of Remote Sensing
• /
• v.35 no.6_1
• /
• pp.959-971
• /
• 2019

CNN (Convolutional Neural Network) is one representative deep learning algorithm, which can extract high-level spatial and spectral features, and has been applied for hyperspectral image classification. However, one significant drawback behind the application of CNNs in hyperspectral images is the high dimensionality of the data, which increases the training time and processing complexity. To address this problem, several CNN based hyperspectral image classification studies have exploited PCA (Principal Component Analysis) for dimensionality reduction. One limitation to this is that the spectral information of the original image can be lost through PCA. Although it is clear that the use of PCA affects the accuracy and the CNN training time, the impact of PCA for CNN based hyperspectral image classification has been understudied. The purpose of this study is to analyze the quantitative effect of PCA in CNN for hyperspectral image classification. The hyperspectral images were first transformed through PCA and applied into the CNN model by varying the size of the reduced dimensionality. In addition, 2D-CNN and 3D-CNN frameworks were applied to analyze the sensitivity of the PCA with respect to the convolution kernel in the model. Experimental results were evaluated based on classification accuracy, learning time, variance ratio, and training process. The size of the reduced dimensionality was the most efficient when the explained variance ratio recorded 99.7%~99.8%. Since the 3D kernel had higher classification accuracy in the original-CNN than the PCA-CNN in comparison to the 2D-CNN, the results revealed that the dimensionality reduction was relatively less effective in 3D kernel.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2023.01a
• /
• pp.369-372
• /
• 2023

본 논문에서는 삼각형 구조로 구성된 3차원 메쉬(Mesh)에서 합성곱 신경망(Convolution Neural Network, CNN)을 응용하여 정확도가 높은 새로운 학습 표현 기법을 제시한다. 우리는 메쉬를 구성하고 있는 폴리곤의 edge와 face의 로컬 특징을 기반으로 학습을 진행한다. 일반적으로 딥러닝은 인공신경망을 수많은 계층 형태로 연결한 기법을 말하며, 주요 처리 대상은 1, 2차원 데이터 형태인 오디오 파일과 이미지였다. 인공지능에 대한 연구가 지속되면서 3차원 딥러닝이 도입되었지만, 기존의 학습과는 달리 3차원 딥러닝은 데이터의 확보가 쉽지 않다. 혼합현실과 메타버스 시장의 확대로 인해 3차원 모델링 시장이 증가하고, 기술의 발전으로 데이터를 획득할 수 있는 방법이 생겼지만, 3차원 데이터를 직접적으로 학습에 이용하는 방식으로 적용하는 것은 쉽지 않다. 그렇게 때문에 본 논문에서는 산업 현장에서 이용되는 데이터인 메쉬 구조를 폴리곤의 최소 단위인 삼각형 형태로 구성하여 학습 데이터를 구성해 기존의 방법보다 정확도가 높은 학습 기법을 제안한다.

• Electronics and Telecommunications Trends
• /
• v.34 no.4
• /
• pp.15-22
• /
• 2019

Recent technological advances in three-dimensional (3D) sensing devices and machine learning such as deep leaning has enabled data-driven 3D applications. Research on artificial intelligence has developed for the past few years and 3D deep learning has been introduced. This is the result of the availability of high-quality big data, increases in computing power, and development of new algorithms; before the introduction of 3D deep leaning, the main targets for deep learning were one-dimensional (1D) audio files and two-dimensional (2D) images. The research field of deep leaning has extended from discriminative models such as classification/segmentation/reconstruction models to generative models such as those including style transfer and generation of non-existing data. Unlike 2D learning, it is not easy to acquire 3D learning data. Although low-cost 3D data acquisition sensors have become increasingly popular owing to advances in 3D vision technology, the generation/acquisition of 3D data is still very difficult. Even if 3D data can be acquired, post-processing remains a significant problem. Moreover, it is not easy to directly apply existing network models such as convolution networks owing to the various ways in which 3D data is represented. In this paper, we summarize technological trends in AI-based 3D content generation.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.52 no.1
• /
• pp.89-101
• /
• 2015

This paper proposes a statistical model of 3-D positions when tracking moving targets using the uncooled infrared (IR) stereo camera system. The proposed model is derived from two errors. One is the position error which is caused by the sampling pixels in the digital image. The other is the timing jitter which results from the irregular capture-timing in the infrared cameras. The capture-timing in the IR camera is measured using the jitter meter designed in this paper, and the observed jitters are statistically modeled as Gaussian distribution. This paper derives an integrated probability distribution by combining jitter error with pixel position error. The combined error is modeled as the convolution of two error distributions. To verify the proposed statistical position error model, this paper has some experiments in tracking moving objects with IR stereo camera. The 3-D positions of object are accurately measured by the trajectory scanner, and 3-D positions are also estimated by stereo matching from IR stereo camera system. According to the experiments, the positions of moving object are estimated within the statistically reliable range which is derived by convolution of two probability models of pixel position error and timing jitter respectively. It is expected that the proposed statistical model can be applied to estimate the uncertain 3-D positions of moving objects in the diverse fields.

• Kyungpook Mathematical Journal
• /
• v.48 no.3
• /
• pp.359-366
• /
• 2008

Let A(p) be the class of functions $f\;:\;z^p\;+\;\sum\limits_{j=1}^{\infty}a_jz^{p+j}$ analytic in the open unit disc E. Let, for any integer n > -p, $f_{n+p-1}(z)\;=\;z^p+\sum\limits_{j=1}^{\infty}(p+j)^{n+p-1}z^{p+j}$. We define $f_{n+p-1}^{(-1)}(z)$ by using convolution * as $f_{n+p-1}\;*\;f_{n+p-1}^{-1}=\frac{z^p}{(1-z)^{n+p}$. A function p, analytic in E with p(0) = 1, is in the class $P_k(\rho)$ if ${\int}_0^{2\pi}\|\frac{Re\;p(z)-\rho}{p-\rho}\|\;d\theta\;\leq\;k{\pi}$, where $z=re^{i\theta}$, $k\;\geq\;2$ and $0\;{\leq}\;\rho\;{\leq}\;p$. We use the class $P_k(\rho)$ to introduce a new class of multivalent analytic functions and define an integral operator $L_{n+p-1}(f)\;\;=\;f_{n+p-1}^{-1}\;*\;f$ for f(z) belonging to this class. We derive some interesting properties of this generalized integral operator which include inclusion results and radius problems.

• Journal of the Korean Mathematical Society
• /
• v.42 no.3
• /
• pp.573-597
• /
• 2005

A continuous linear operator T, on the space of entire functions in d variables, is PDE-preserving for a given set $\mathbb{P}\;\subseteq\;\mathbb{C}|\xi_{1},\ldots,\xi_{d}|$ of polynomials if it maps every kernel-set ker P(D), $P\;{\in}\;\mathbb{P}$, invariantly. It is clear that the set $\mathbb{O}({\mathbb{P}})$ of PDE-preserving operators for $\mathbb{P}$ forms an algebra under composition. We study and link properties and structures on the operator side $\mathbb{O}({\mathbb{P}})$ versus the corresponding family $\mathbb{P}$ of polynomials. For our purposes, we introduce notions such as the PDE-preserving hull and basic sets for a given set $\mathbb{P}$ which, roughly, is the largest, respectively a minimal, collection of polynomials that generate all the PDE-preserving operators for $\mathbb{P}$. We also describe PDE-preserving operators via a kernel theorem. We apply Hilbert's Nullstellensatz.

• Journal of Information Processing Systems
• /
• v.17 no.4
• /
• pp.787-800
• /
• 2021

Three-dimensional (3D) object classification tasks using point clouds are widely used in 3D modeling, face recognition, and robotic missions. However, processing raw point clouds directly is problematic for a traditional convolutional network due to the irregular data format of point clouds. This paper proposes a pointwise convolution neural network (CNN) structure that can process point cloud data directly without preprocessing. First, a 2D convolutional layer is introduced to percept coordinate information of each point. Then, multiple 2D convolutional layers and a global max pooling layer are applied to extract global features. Finally, based on the extracted features, fully connected layers predict the class labels of objects. We evaluated the proposed pointwise CNN structure on the ModelNet10 dataset. The proposed structure obtained higher accuracy compared to the existing methods. Experiments using the ModelNet10 dataset also prove that the difference in the point number of point clouds does not significantly influence on the proposed pointwise CNN structure.

• International Journal of Contents
• /
• v.15 no.4
• /
• pp.59-64
• /
• 2019

In recent years, emotion recognition has been an interesting and challenging topic. Compared to facial expressions and speech modality, gesture-based emotion recognition has not received much attention with only a few efforts using traditional hand-crafted methods. These approaches require major computational costs and do not offer many opportunities for improvement as most of the science community is conducting their research based on the deep learning technique. In this paper, we propose an end-to-end deep learning approach for classifying emotions based on bodily gestures. In particular, the informative keyframes are first extracted from raw videos as input for the 3D-CNN deep network. The 3D-CNN exploits the short-term spatiotemporal information of gesture features from selected keyframes, and the convolutional LSTM networks learn the long-term feature from the features results of 3D-CNN. The experimental results on the FABO dataset exceed most of the traditional methods results and achieve state-of-the-art results for the deep learning-based technique for gesture-based emotion recognition.

• Proceedings of the Korea Committee for Ocean Resources and Engineering Conference
• /
• 2001.10a
• /
• pp.164-168
• /
• 2001

The motion of a ship advancing in regular waves is analyzed in the time-domain using the convolution integral of the radiation forces. The memory effect functions and infinite frequency added masses are obtained from the solution of the three dimensional improved Green integral equation in the frequency domain by making use of the Fourier transformation. The ship motions in regular waves have been calculated by both the time and frequency domain methods. It has been shown that they agree very well with each other. The present time-domain method can be used to predict the time histories of unsteady motions in irregular waves. It can also be used to calculate the hydrostatic and Froude-Krylov forces over the instantaneous wetted surface of the ship hull to predict large ship motions, in a practical sense, advancing in large amplitude waves.

• Journal of Korean Ophthalmic Optics Society
• /
• v.6 no.1
• /
• pp.87-91
• /
• 2001

The a-C:H films have been grown on the glass substrate by PECVD method, where plasma was generated with a 60 Hz line power source. The carbonization is checked from peak intensities of D($sp^3$) and G($sp^2$) peaks in Raman spectra and is analyzed using the Gaussian convolution method of spectrum. Both the bonding strength of C-H and the ratio of $sp^3$ to $sp^2$ in bonding are found to be slightly dependent of partial pressure of $C_2H_2$.