• Journal of Biomedical Engineering Research
• /
• v.25 no.2
• /
• pp.89-95
• /
• 2004

Intensity-modulated radiation therapy (IMRT) often uses small beam segments. The heterogeneity effect is well known for relatively large field sizes used in the conventional radiation treatments. However, this effect is not known in small fields such as the beamlets used in IMRT. There are many factors that can cause errors in the small field i.e. electronic disequilibrium and multiple electron scattering. This study prepared geometrically regular heterogeneous phantoms, and compared the measurements with the calculations using the Convolution/Superposition algorithm and Monte Carlo method for small beams. This study used the BEAM00/EGS4 code to simulate the head of a Varian 2300C/D. The commissioning of a 6MV photon beam were performed from two points of view, the beam profiles and depth doses. The calculated voxel size was 1${\times}$1${\times}$2$\textrm{cm}^2$ with field sizes of 1${\times}$1$\textrm{cm}^2$, 2${\times}$2$\textrm{cm}^2$, and 5${\times}$5$\textrm{cm}^2$. The XiOTM TPS (Treatment Planning System) was used for the calculation using the Convolution/Superposition algorithm. The 6MV photon beam was irradiated to homogeneous (water equivalent) and heterogeneous phantoms (water equivalent ＋ air cavity, water equivalent ＋ bone equivalent). The beam profiles were well matched within :t1 mm and the depth doses were within ${\pm}$2%. In conclusion, the dose calculations of the Convolution/Superposition and Monte Carlo simulations showed good agreement with the film measurements in the small field.

• Proceedings of the Korean Institute of Communication Sciences Conference
• /
• 1984.10a
• /
• pp.82-87
• /
• 1984

A 2-D digital filter with rational frequency response can be expanded into an infinite sequence of filterins operations. Each filtering operation can be implemented by convolution with a Low-order 20D finite-extent impulse response. If a convergence constraint is satisfied, the sequence of estimates will approach the desired output signal. In practice, as the number of iterations is finite, the frequency response implemented by iterative computations is an approximation to the desired rational frequency response.

• Korean Journal of Remote Sensing
• /
• v.35 no.5_1
• /
• pp.681-692
• /
• 2019

In this paper, a hybrid deep learning model, called 2D convolution with bidirectional long short-term memory (2DCBLSTM), is presented that can effectively combine both spatial and temporal features for crop classification. In the proposed model, 2D convolution operators are first applied to extract spatial features of crops and the extracted spatial features are then used as inputs for a bidirectional LSTM model that can effectively process temporal features. To evaluate the classification performance of the proposed model, a case study of crop classification was carried out using multi-temporal unmanned aerial vehicle images acquired in Anbandegi, Korea. For comparison purposes, we applied conventional deep learning models including two-dimensional convolutional neural network (CNN) using spatial features, LSTM using temporal features, and three-dimensional CNN using spatio-temporal features. Through the impact analysis of hyper-parameters on the classification performance, the use of both spatial and temporal features greatly reduced misclassification patterns of crops and the proposed hybrid model showed the best classification accuracy, compared to the conventional deep learning models that considered either spatial features or temporal features. Therefore, it is expected that the proposed model can be effectively applied to crop classification owing to its ability to consider spatio-temporal features of crops.

• Journal of the Korea Society of Computer and Information
• /
• v.13 no.6
• /
• pp.163-171
• /
• 2008

In this paper, we present an adaptive parametric cubic convolution technique in order to enlarge the low resolution image to the high resolution image. The proposed method consists of two steps. During the first interpolation step, we acquire adaptive parameters in introducing a new cost-function to reflect frequency properties. And, the second interpolation step performs cubic convolution by applying the parameters obtained from the first step. The enhanced interpolation kernel using adaptive parameters produces output image better than the conventional one using a fixed parameter. Experimental results show that the proposed method can not only provides the performances of $0.5{\sim}4dB$ improvements in terms of PSNR, but also exhibit better edge preservation ability and original image similarity than conventional methods in the enlarged images.

• 한국정보통신설비학회:학술대회논문집
• /
• 2007.08a
• /
• pp.386-389
• /
• 2007

TV technology started from black and white TV. Color TV invented and users request more realistic TV technology. The next technology is 3DTV. For 3DTV, 3D display technology, 3D coding technology, digital mux/demux technology in broadcast and 3D video acquisition are needed. Moreover, Almost every contents now exist are 2D contents. It causes necessity to convert from 2D to 3D. This article describes 2D/3D conversion algorithm and H/W platform on FPGA board. Time difference makes 3D effect and convolution filter increased the effect. Distorted image and original image give 3D effect. The algorithm is shown on 3D display. The display device shows 3D effect by parallax barrier method and has FPGA board.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.11
• /
• pp.1627-1634
• /
• 2021

This paper introduces a low-power compact ADC circuit for analog Convolutional filter for low-power neural network accelerator SOC. While convolutional neural network accelerators can speed up the learning and inference process, they have drawback of consuming excessive power and occupying large chip area due to large number of multiply-and-accumulate operators when implemented in complex digital circuits. To overcome these drawbacks, we implemented an analog convolutional filter that consists of an analog multiply-and-accumulate arithmetic circuit along with an ADC. This paper is focused on the design optimization of a low-power 8bit SAR ADC for the analog convolutional filter accelerator We demonstrate how to minimize the capacitor-array DAC, an important component of SAR ADC, which is three times smaller than the conventional circuit. The proposed ADC has been fabricated in CMOS 65nm process. It achieves an overall size of 1355.7㎛², power consumption of 2.6㎼ at a frequency of 100MHz, SNDR of 44.19 dB, and ENOB of 7.04bit.

• Proceedings of the Korean Society of Medical Physics Conference
• /
• 2002.09a
• /
• pp.119-120
• /
• 2002

The aim is to urge the need of elaborate commissioning of 3D RTP system from the firsthand experience. A 3D RTP system requires so much data such as beam data and patient data. Most data of radiation beam are directly transferred from a 3D dose scanning system, and some other data are input by editing. In the process inputting parameters and/or data, no error should occur. For RTP system using algorithm-bas ed-on beam-modeling, careless beam-data processing could also cause the treatment error. Beam data of 3 different qualities of photon from two linear accelerators, patient data and calculated results were commissioned. For PDD, the doses by Clarkson, convolution, superposition and fast superposition methods at 10 cm for 10${\times}$10 cm field, 100 cm SSD were compared with the measured. An error in the SCD for one quality was input by the service engineer. Whole SCD defined by a physicist is SAD plus d$\sub$max/, the value was just SAD. That resulted in increase of MU by 100${\times}$((1_d$\sub$max//SAD)$^2$-1)%. For 10${\times}$10 cm open field, 1 m SSD and at 10 cm depth in uniform medium of relative electron density (RED) 1, PDDs for 4 algorithms of dose calculation, Clarkson, convolution, superposition and fast-superposition, were compared with the measured. The calculated PDD were similar to the measured. For 10${\times}$10 cm open field, 1 m SSD and at 10 cm depth with 5 cm thick inhomogeneity of RED 0.2 under 2 cm thick RED 1 medium, PDDs for 4 algorithms were compared. PDDs ranged from 72.2% to 77.0% for 4 MV X-ray and from 90.9% to 95.6% for 6 MV X-ray. PDDs were of maximum for convolution and of minimum for superposition. For 15${\times}$15 cm symmetric wedged field, wedge factor was not constant for calculation mode, even though same geometry. The reason is that their wedge factor is considering beam hardness and ray path. Their definition requires their users to change the concept of wedge factor. RTP user should elaborately review beam data and calculation algorithm in commissioning.

• International Journal of Computer Science & Network Security
• /
• v.23 no.3
• /
• pp.177-186
• /
• 2023

Hyperspectral imaging technology is one of the most efficient and fast-growing technologies in recent years. Hyperspectral image (HSI) comprises contiguous spectral bands for every pixel that is used to detect the object with significant accuracy and details. HSI contains high dimensionality of spectral information which is not easy to classify every pixel. To confront the problem, we propose a novel RGB channel Assimilation for classification methods. The color features are extracted by using chromaticity computation. Additionally, this work discusses the classification of hyperspectral image based on Domain Transform Interpolated Convolution Filter (DTICF) and 3D-CNN with Bi-directional-Long Short Term Memory (Bi-LSTM). There are three steps for the proposed techniques: First, HSI data is converted to RGB images with spatial features. Before using the DTICF, the RGB images of HSI and patch of the input image from raw HSI are integrated. Afterward, the pair features of spectral and spatial are excerpted using DTICF from integrated HSI. Those obtained spatial and spectral features are finally given into the designed 3D-CNN with Bi-LSTM framework. In the second step, the excerpted color features are classified by 2D-CNN. The probabilistic classification map of 3D-CNN-Bi-LSTM, and 2D-CNN are fused. In the last step, additionally, Markov Random Field (MRF) is utilized for improving the fused probabilistic classification map efficiently. Based on the experimental results, two different hyperspectral images prove that novel RGB channel assimilation of DTICF-3D-CNN-Bi-LSTM approach is more important and provides good classification results compared to other classification approaches.

• The Journal of the Convergence on Culture Technology
• /
• v.8 no.6
• /
• pp.867-871
• /
• 2022

Amid the recent rapid trend change, the change in design has a great impact on the sales of fashion companies, so it is inevitable to be careful in choosing new designs. With the recent development of the artificial intelligence field, various machine learning is being used a lot in the fashion market to increase consumers' preferences. To contribute to increasing reliability in the development of new products by quantifying abstract concepts such as preferences, we generate new images that do not exist through three adversarial generative neural networks (GANs) and numerically compare abstract concepts of preferences using pre-trained convolution neural networks (CNNs). Deep convolutional generative adversarial networks (DCGAN), Progressive growing adversarial networks (PGGAN), and Dual Discriminator generative adversarial networks (DANs), which were trained to produce comparative, high-level, and high-level images. The degree of similarity measured was considered as a preference, and the experimental results showed that D2GAN showed a relatively high similarity compared to DCGAN and PGGAN.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2006.11a
• /
• pp.89-92
• /
• 2006

본 논문은 SIFT(Scale Invariant Feature Transform)알고리즘의 실시간처리 응용프로그램 작성기법을 기술하고 있는데, 단일 프로세서에서 병렬처리 기능을 지원하도록 설계된 SSE 명령어 집합을 사용하여 가우시안 convolution을 구현하고 있다. SIFT알고리즘의 Scale-space를 생성하는 과정에 수행되는 가우시안 Convolution은 연산시간이 과도하게 요구된다.[1] 2D의 가우시안 필터가 영상을 구성하는 모든 셀과 1:1로 연산을 수행하므로 이 연산의 소요시간은 영상의 가로, 세로 길이 그리고 필터의 크기에 비례하여 결정된다. 이 논문에서 제안하는 방법은 연산을 위해 CPU 내부로 한번 읽어 들인 픽셀자료에 대해 가능한 모든 연산을 SSE 명령어 집합을 사용하여 수행함으로써 병렬 연산에 의한 연산시간 절감과 메모리 접근 최소화를 통한 입출력시간 절감을 통해 전체 연산시간을 단축 하였다.