• Journal of Biosystems Engineering
• /
• 제42권4호
• /
• pp.350-357
• /
• 2017

Purpose: In-situ monitoring of water quality is fundamental to most environmental applications. The high cost and long delays of conventional laboratory methods used to determine water quality, including on-site sampling and chemical analysis, have limited their use in efficiently managing water sources while preventing environmental pollution. The objective of this study was to develop an on-site water monitoring system consisting mainly of an Arduino board and a sensor array of multiple ion selective electrodes (ISEs) to measure the concentration of $NO_3$ ions. Methods: The developed system includes a combination of three ISEs, double-junction reference electrode, solution container, sampling system consisting of three pumps and solenoid valves, signal processing circuit, and an Arduino board for data acquisition and system control. Prior to each sample measurement, a two-point normalization method was applied for a sensitivity adjustment followed by an offset adjustment to minimize the potential drift that could occur during continuous measurement and standardize the response of multiple electrodes. To investigate its utility in on-site nitrate monitoring, the prototype was tested in a facility where drinking water was collected from a water supply source. Results: Differences in the electric potentials of the $NO_3$ ISEs between 10 and $100mg{\cdot}L^{-1}$ $NO_3$ concentration levels were nearly constant with negative sensitivities of 58 to 62 mV during the period of sample measurement, which is representative of a stable electrode response. The $NO_3$ concentrations determined by the ISEs were almost comparable to those obtained with standard instruments within 15% relative errors. Conclusions: The use of the developed on-site nitrate monitoring system based on automatic sampling and two-point normalization was feasible for detecting abrupt changes in nitrate concentration at various water supply sites, showing a maximum difference of $4.2mg{\cdot}L^{-1}$ from an actual concentration of $14mg{\cdot}L^{-1}$.

• 한국정보통신학회:학술대회논문집
• /
• 한국정보통신학회 2015년도 춘계학술대회
• /
• pp.376-380
• /
• 2015

본 논문은 안개 낀 날 차선을 인식하는 방법을 제안한다. 이것은 주행 중 안개구간이 나타나면 시야확보가 어려운 운전자의 안전을 도모하고 또한 자동차 자율 주행을 가능하게 하기 위한 것이다. 제안한 방법은 먼저 입력 영상에서 화소 수 분포와 시작점으로 안개 구간인지를 판단한다. 만약 안개구간이면 Retinex 알고리즘에서 미디언 필터를 입력영상의 범위만큼 한 후 히스토그램 평활화와 정규화를 수행한다. 실험 결과 기존 연구보다 차선 검출이 정확하고 먼 거리까지 인식할 수 있었다.

• 대한임베디드공학회논문지
• /
• 제14권1호
• /
• pp.1-9
• /
• 2019

In this paper, we propose a method of generating a 2D gray image from LiDAR 3D reflection intensity. The proposed method uses the Fully Convolutional Network (FCN) to generate the gray image from 2D reflection intensity which is projected from LiDAR 3D intensity. Both encoder and decoder of FCN are configured with several convolution blocks in the symmetric fashion. Each convolution block consists of a convolution layer with $3{\times}3$ filter, batch normalization layer and activation function. The performance of the proposed method architecture is empirically evaluated by varying depths of convolution blocks. The well-known KITTI data set for various scenarios is used for training and performance evaluation. The simulation results show that the proposed method produces the improvements of 8.56 dB in peak signal-to-noise ratio and 0.33 in structural similarity index measure compared with conventional interpolation methods such as inverse distance weighted and nearest neighbor. The proposed method can be possibly used as an assistance tool in the night-time driving system for autonomous vehicles.

• 농업과학연구
• /
• 제39권3호
• /
• pp.413-420
• /
• 2012

Spectroscopic measurement method based on visible and near-infrared wavelengths was prominent technology for rapid and non-destructive evaluation of internal quality of fruits. Reflectance measurement was performed to evaluate firmness, soluble solid content, and acid content of truss tomatoes by hyperspectral reflectance imaging system. The Vis/NIR reflectance spectra was acquired from truss tomatoes sorted by 6 ripening stages. The multivariable analysis based on partial least square (PLS) was used to develop regression models with several preporcessing methods, such as smoothing, normalization, multiplicative scatter correction (MSC), and standard normal variate (SNV). The best model was selected in terms of coefficient of determination of calibration ($R_c^2$) and full cross validation ($R_{cv}^2$), and root mean standard error of calibration (RMSEC) and full cross validation (RMSECV). The results of selected models were 0.8976 ($R_p^2$), 6.0207 kgf (RMSEP) with gaussian filter of smoothing, 0.8379 ($R_p^2$), $0.2674^{\circ}Bx$ (RMSEP) with the mean of normalization, and 0.7779 ($R_p^2$), 0.1033% (RMSEP) with median filter of smoothing for firmness, soluble solid content (SSC), and acid content, respectively. Results show that Vis / NIR hyperspectral reflectance imaging technique has good potential for the measurement of internal quality of truss tomato.

• 대한전자공학회논문지SP
• /
• 제41권3호
• /
• pp.163-174
• /
• 2004

치과 분야에서는 치아교정이나 수술 시뮬레이션을 위해서 각 치아를 개별적으로 조작할 수 있는 3차원 치아모델이 필요하다. 치아 CT 영상으로부터 이러한 치아모델의 재구성을 위해서는 각 치아를 이웃한 치아나 치조골로부터 정확하게 분리할 수 있어야 한다. 본 연구에서는 치아 영역을 효과적으로 분리하기 위한 영상정규화 방안과 최적임계화방안을 제안한다. 제안된 방법은 연속적인 CT 영상 슬라이스들에서 치아영역의 형태와 밝기는 점진적으로 변한다는 사실을 근거로 이전 슬라이스에서 추정된 임계치를 이용하여 현 슬라이스의 임시치아경계를 결정하고 이것을 바탕으로 보다 정확한 임계치를 계산한다.

• 대한전자공학회논문지SP
• /
• 제44권3호
• /
• pp.80-87
• /
• 2007

국부 통계치에 근거한 서양금석문 영상향상 기법을 고안한다. 영상데이터는 배경과 정보의 두 영역으로 구분한다. 통계 및 함수적 분석을 통해 서양금석문 영상 대부분이 가우스 회색도분포임을 규명하고 분포 및 영역특성을 고려한 모델을 구현한다. 모델을 대상으로 국부정규화처리 알고리즘을 수식화하고 파라미터를 추출하며 이동창에서의 기능과 특성을 논의한다. 파라미터와 이동창의 크기를 조정하여 화소 회색도의 공간 분포를 변형하고 영역을 선별한다. 이 때 국부통계치는 알고리즘을 실현하는 유용한 근거로 활용된다. 알고리즘 적용에 의해 영역의 잡음과 불규칙한 분포 상태가 평활되는 동시에 영역 사이의 회색도 격차를 증대시켜 영상을 향상한다. 실험결과는 제안된 방식이 기존의 영상향상 기법보다 우수한 효과가 있음을 보여준다.

• Fisheries and Aquatic Sciences
• /
• 제19권4호
• /
• pp.21.1-21.11
• /
• 2016

Background: The evaluation of suitable reference genes as normalization controls is a prerequisite requirement for launching quantitative reverse transcription-PCR (RT-qPCR)-based expression study. In order to select the stable reference genes in abalone Haliotis discus hannai tissues (gill and hepatopancreas) under heavy metal exposure conditions (Cu, Zn, and Cd), 12 potential candidate housekeeping genes were subjected to expression stability based on the comprehensive ranking while integrating four different statistical algorithms (geNorm, NormFinder, BestKeeper, and ${\Delta}CT$ method). Results: Expression stability in the gill subset was determined as RPL7 > RPL8 > ACTB > RPL3 > PPIB > RPL7A > EF1A > RPL4 > GAPDH > RPL5 > UBE2 > B-TU. On the other hand, the ranking in the subset for hepatopancreas was RPL7 > RPL3 > RPL8 > ACTB > RPL4 > EF1A > RPL5 > RPL7A > B-TU > UBE2 > PPIB > GAPDH. The pairwise variation assessed by the geNorm program indicates that two reference genes could be sufficient for accurate normalization in both gill and hepatopancreas subsets. Overall, both gill and hepatopancreas subsets recommended ribosomal protein genes (particularly RPL7) as stable references, whereas traditional housekeepers such as ${\beta}-tubulin$ (B-TU) and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) genes were ranked as unstable genes. The validation of reference gene selection was confirmed with the quantitative assay of MT transcripts. Conclusions: The present analysis showed the importance of validating reference genes with multiple algorithmic approaches to select genes that are truly stable. Our results indicate that expression stability of a given reference gene could not always have consensus across tissue types. The data from this study could be a good guide for the future design of RT-qPCR studies with respect to metal regulation/detoxification and other related physiologies in this abalone species.

• 동의신경정신과학회지
• /
• 제17권2호
• /
• pp.133-145
• /
• 2006

Objective : The purpose of this experimental-controlled study was to investigate the characteristics of the autonomic nervous system in students with Keongke by using HRV(Heart rate variability) and PSV(Pupil size variability). Method : The study group was consisted of 11 students with self recognition as the experimental group, and 25 normal students as the control group. Informations on gender and age were obtained by medical charts and personal interviews. By using heart rate variability and pupil size variability, we measured the value of HRT(Heart rate), SDNN(Standard deviation of NN intervals), LFnorm(Low frequency normalization), HFnorm(High frequency normalization), LF/HF ratio, Pupil area, B.S.(Basic size), C.R.(Max Constriction Rate) and 1s.d.(1sec Dilation Rate). I compared the degrees of the sympathetic and parasympathetic activity. Result : 1. In the result of heart rate variability between experimental and control group, none of the parameters of experimental group were significantly different from control group. And even though there were no statistical significance, there were some numerical differences in SDNN, LF norm, HF norm. 2. In pupil size variability, C.R. and 1s.d. of the experimental group were increased compared to control group. Conclusion : The study results suggest that the group with Keongke has differences of autonomic nervous system as compared to those in normal state. Measurement value of PSV is a new technical approach to estimate the autonomic nervous system.

• 인터넷정보학회논문지
• /
• 제21권2호
• /
• pp.49-55
• /
• 2020

랜덤 심볼열을 기반으로 한 정보이론적 학습법 (ITL)은 특정 확률분포를 갖도록 랜덤하게 발생시킨 심볼열을 타겟 데이터로 활용하고, 입력 데이터 사이의 확률분포 거리 최소화를 비용함수로 하여 설계된다. 이 방식의 단점으로, 고정상수를 알고리듬 갱신의 스텝사이즈로 사용하므로 입력 전력의 통계적 추이를 활용할 수 없다. 정보포텐셜 출력(information potential output, IPO)와 연관된 기울기에서는 정보포텐셜 입력(information potential input, IPI)이, 정보포텐셜 오차(information potential error, IPE)와 관련된 기울기에서는 입력자체가 입력으로 작용함을 이 연구에서 밝혀내고, 입력의 전력 추이를 따로 계산하여 스텝사이즈 (step size)를 정규화하도록 제안하였다. 제안된 알고리듬은 충격성잡음과 다중경로 페이딩 환경의 통신시스템 실험에서 기존 방식보다 약 4dB 정도 더 낮은 정상상태 오차 전력, 약 2배 이상 빠른 수렴속도를 나타냈다.

• 한국멀티미디어학회논문지
• /
• 제23권4호
• /
• pp.525-538
• /
• 2020

A practical animal face classification system that classifies animals in image and video data is considered as a pivotal topic in machine learning. In this research, we are proposing a novel method of fully connected dual Deep Convolutional Neural Network (DCNN), which extracts and analyzes image features on a large scale. With the inclusion of the state of the art Batch Normalization layer and Exponential Linear Unit (ELU) layer, our proposed DCNN has gained the capability of analyzing a large amount of dataset as well as extracting more features than before. For this research, we have built our dataset containing ten thousand animal faces of ten animal classes and a dual DCNN. The significance of our network is that it has four sets of convolutional functions that work laterally with each other. We used a relatively small amount of batch size and a large number of iteration to mitigate overfitting during the training session. We have also used image augmentation to vary the shapes of the training images for the better learning process. The results demonstrate that, with an accuracy rate of 92.0%, the proposed DCNN outruns its counterparts while causing less computing costs.