• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.11 no.8
• /
• pp.3965-3983
• /
• 2017

We propose a novel eye tracking method that can estimate 3D world coordinates using an infrared (IR) stereo camera for indoor and outdoor environments. This method first detects dark evidences such as eyes, eyebrows and mouths by fast multi-level thresholding. Among these evidences, eye pair evidences are detected by evidential reasoning and geometrical rules. For robust accuracy, two classifiers based on multiple layer perceptron (MLP) using gradient local binary patterns (GLBPs) verify whether the detected evidences are real eye pairs or not. Finally, the 3D world coordinates of detected eyes are calculated by region-based stereo matching. Compared with other eye detection methods, the proposed method can detect the eyes of people wearing sunglasses due to the use of the IR spectrum. Especially, when people are in dark environments such as driving at nighttime, driving in an indoor carpark, or passing through a tunnel, human eyes can be robustly detected because we use active IR illuminators. In the experimental results, it is shown that the proposed method can detect eye pairs with high performance in real-time under variable illumination conditions. Therefore, the proposed method can contribute to human-computer interactions (HCIs) and intelligent transportation systems (ITSs) applications such as gaze tracking, windshield head-up display and drowsiness detection.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.16 no.5
• /
• pp.556-561
• /
• 2006

It is very important to classify the DNA Chip image pattern in order to acquire useful information about genetic disease of people. In this paper, we developed the novel pattern classification method of DNA Chip image using MLP based back-propagation and Self organizing Map learning algorithm. And then we compared and analyzed these classified pattern results. Also we carried out experiment in the MV2440 board using CPU Cote for S3C2440(ARM 920T) and PC environment, and displayed its results in order to give the genetic information to user mote easily in various environment.

• Korean Journal of Remote Sensing
• /
• v.22 no.5
• /
• pp.367-372
• /
• 2006

Cloud detection algorithm is being developed as primary one of the 16 baseline products of CMDPS (COMS Meteorological Data Processing System), which is under development for the real-time application of data will be observed from COMS Meteorological Imager. For cloud detection from satellite data, we studied two different algorithms. One is threshold technique based algorithm, which is traditionally used, and another is artificial neural network model. MPEF scene analysis algorithm is the basic idea of threshold cloud detection algorithm, and some modifications are conducted for COMS. For the neural network, we selected MLP with back-propagation algorithm. Prototype software of each algorithm was completed and evaluated by using the MTSAT-IR and GOES-9 data. Currently the software codes are standardized using Fortran90 language. For the preparation as an operational algorithm, we will setup the validation strategy and tune up the algorithm continuously. This paper shows the outline of the two cloud detection algorithms and preliminary test results of both algorithms.

• The Journal of Korea Robotics Society
• /
• v.14 no.3
• /
• pp.228-235
• /
• 2019

This paper presents a multiple DoFs (degrees-of-freedom) prosthetic forearm and sEMG (surface electromyogram) pattern recognition and motion intent classification of forearm amputee. The developed prosthetic forearm has 9 DoFs hand and single-DoF wrist, and the socket is designed considering wearability. In addition, the pattern recognition based on sEMG is proposed for prosthetic control. Several experiments were conducted to substantiate the performance of the prosthetic forearm. First, the developed prosthetic forearm could perform various motions required for activity of daily living of forearm amputee. It was able to control according to shape and size of the object. Additionally, the amputee was able to perform 'tying up shoe' using the prosthetic forearm. Secondly, pattern recognition and classification experiments using the sEMG signals were performed to find out whether it could classify the motions according to the user's intents. For this purpose, sEMG signals were applied to the multilayer perceptron (MLP) for training and testing. As a result, overall classification accuracy arrived at 99.6% for all participants, and all the postures showed more than 97% accuracy.

• Nuclear Engineering and Technology
• /
• v.50 no.7
• /
• pp.1154-1159
• /
• 2018

Precise prediction of the radiation interaction position in scintillators plays an important role in medical and industrial imaging systems. In this research, the incident position of the gamma rays was predicted precisely in a plastic rod scintillator by using attenuation technique and multilayer perceptron (MLP) neural network, for the first time. Also, this procedure was performed using nonlinear regression (NLR) method. The experimental setup is comprised of a plastic rod scintillator (BC400) coupled with two PMTs at two sides, a $^{60}Co$ gamma source and two counters that record count rates. Using two proposed techniques (ANN and NLR), the radiation interaction position was predicted in a plastic rod scintillator with a mean relative error percentage less than 4.6% and 14.6%, respectively. The mean absolute error was measured less than 2.5 and 5.5. The correlation coefficient was calculated 0.998 and 0.984, respectively. Also, the ANN technique was confirmed by leave-one-out (LOO) method with 1% error. These results presented the superiority of the ANN method in comparison with NLR and the other methods. The technique and set up used are simpler and faster than other the previous position sensitive detectors. Thus, the time, cost and shielding and electronics requirements are minimized and optimized.

• Journal of Korean Association for Spatial Structures
• /
• v.21 no.2
• /
• pp.99-110
• /
• 2021

This study presents the estimation of crack depth by analyzing temperatures extracted from thermal images and environmental parameters such as air temperature, air humidity, illumination. The statistics of all acquired features and the correlation coefficient among thermal images and environmental parameters are presented. The concrete crack depths were predicted by four different machine learning models: Multi-Layer Perceptron (MLP), Random Forest (RF), Gradient Boosting (GB), and AdaBoost (AB). The machine learning algorithms are validated by the coefficient of determination, accuracy, and Mean Absolute Percentage Error (MAPE). The AB model had a great performance among the four models due to the non-linearity of features and weak learner aggregation with weights on misclassified data. The maximum depth 11 of the base estimator in the AB model is efficient with high performance with 97.6% of accuracy and 0.07% of MAPE. Feature importances, permutation importance, and partial dependence are analyzed in the AB model. The results show that the marginal effect of air humidity, crack depth, and crack temperature in order is higher than that of the others.

• International Journal of Computer Science & Network Security
• /
• v.22 no.10
• /
• pp.406-412
• /
• 2022

Scene classification of very high-resolution (VHR) imagery can attribute semantics to land cover in a variety of domains. Real-world application requirements have not been addressed by conventional techniques for remote sensing image classification. Recent research has demonstrated that deep convolutional neural networks (CNNs) are effective at extracting features due to their strong feature extraction capabilities. In order to improve classification performance, these approaches rely primarily on semantic information. Since the abstract and global semantic information makes it difficult for the network to correctly classify scene images with similar structures and high interclass similarity, it achieves a low classification accuracy. We propose a VHR remote sensing image classification model that uses extracts the global feature from the original VHR image using an EfficientNet-V2L CNN pre-trained to detect similar classes. The image is then classified using a multilayer perceptron (MLP). This method was evaluated using two benchmark remote sensing datasets: the 21-class UC Merced, and the 38-class PatternNet. As compared to other state-of-the-art models, the proposed model significantly improves performance.

• Journal of Korea Multimedia Society
• /
• v.25 no.10
• /
• pp.1448-1456
• /
• 2022

This paper studies a model to diagnose efficiency reduction of inverter using Multilayer Perceptron(MLP). In this study, two inverter data which started operation at different day was used. A Multilayer Perceptron model was made to predict photovoltaic power data of the latest inverter. As a result of the model's performance test, the Mean Absolute Percentage Error(MAPE) was 4.1034. The verified model was applied to one-year-old and two-year-old data after old inverter starting operation. The predictive power of one-year-old inverter was larger than the observed power by 724.9243 on average. And two-year-old inverter's predictive value was larger than the observed power by 836.4616 on average. The prediction error of two-year-old inverter rose 111.5572 on a year. This error is 0.4% of the total capacity. It was proved that the error is meaningful difference by t-test. The error is predicted value minus actual value. Which means that PV system actually generated less than prediction. Therefore, increasing error is decreasing conversion efficiency of inverter. Finally, conversion efficiency of the inverter decreased by 0.4% over a year using this model.

• Journal of Positioning, Navigation, and Timing
• /
• v.12 no.1
• /
• pp.1-9
• /
• 2023

The problem of classifying a non-line-of-sight (NLOS) signal in a multipath channel is important to improve global navigation satellite system (GNSS) positioning accuracy in urban areas. Conventional deep learning-based NLOS signal classifiers use GNSS satellite measurements such as the carrier-to-noise-density ratio (CN_0), pseudorange, and elevation angle as inputs. However, there is a computational inefficiency with use of these measurements and the NLOS signal features expressed by the measurements are limited. In this paper, we propose a Convolutional Neural Network (CNN)-based NLOS signal classifier that receives successive Auto-correlation function (ACF) outputs according to a time-series, which is the most primitive output of GNSS signal processing. We compared the proposed classifier to other DL-based NLOS signal classifiers such as a multi-layer perceptron (MLP) and Gated Recurrent Unit (GRU) to show the superiority of the proposed classifier. The results show the proposed classifier does not require the navigation data extraction stage to classify the NLOS signals, and it has been verified that it has the best detection performance among all compared classifiers, with an accuracy of up to 97%.

• Journal of Korea Society of Digital Industry and Information Management
• /
• v.18 no.4
• /
• pp.67-75
• /
• 2022

Recently, e-learning has been attracting significant attention due to COVID-19. However, while e-learning has many advantages, it has disadvantages as well. One of the main disadvantages of e-learning is that it is difficult for teachers to continuously and systematically monitor learners. Although services such as personalized e-learning are provided to compensate for the shortcoming, systematic monitoring of learners' concentration is insufficient. This study suggests a method to evaluate the learner's concentration by applying machine learning techniques. In this study, emotion and gaze data were extracted from 184 videos of 92 participants. First, the learners' concentration was labeled by experts. Then, statistical-based status indicators were preprocessed from the data. Random Forests (RF), Support Vector Machines (SVMs), Multilayer Perceptron (MLP), and an ensemble model have been used in the experiment. Long Short-Term Memory (LSTM) has also been used for comparison. As a result, it was possible to predict e-learners' concentration with an accuracy of 90.54%. This study is expected to improve learners' immersion by providing a customized educational curriculum according to the learner's concentration level.