• KIISE Transactions on Computing Practices
• /
• v.23 no.5
• /
• pp.299-303
• /
• 2017

By piling up hidden layers in artificial neural networks, deep learning is delivering outstanding performances for high-level abstraction problems such as object/speech recognition and natural language processing. Alternatively, deep-learning users often struggle with the tremendous amounts of time and resources that are required to train deep neural networks. To alleviate this computational challenge, many approaches have been proposed in a diversity of areas. In this work, two of the existing Apache Spark-based acceleration frameworks for deep learning (SparkNet and DeepSpark) are compared and analyzed in terms of the training accuracy and the time demands. In the authors' experiments with the CIFAR-10 and CIFAR-100 benchmark datasets, SparkNet showed a more stable convergence behavior than DeepSpark; but in terms of the training accuracy, DeepSpark delivered a higher classification accuracy of approximately 15%. For some of the cases, DeepSpark also outperformed the sequential implementation running on a single machine in terms of both the accuracy and the running time.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.19 no.6
• /
• pp.227-233
• /
• 2019

Recently, along with the recent development of deep learning technique, neural networks are achieving success in computer vision filed. Convolutional neural network have shown outstanding performance in not only for a simple image classification task, but also for tasks with high difficulty such as object segmentation and detection. However many such deep learning models are based on supervised-learning, which requires more annotation labels than image-level label. Especially image semantic segmentation model requires pixel-level annotations for training, which is very. To solve these problems, this paper proposes a weakly-supervised semantic segmentation method which requires only image level label to train network. Existing weakly-supervised learning methods have limitations in detecting only specific area of object. In this paper, on the other hand, we use multi-classifier deep learning architecture so that our model recognizes more different parts of objects. The proposed method is evaluated using VOC 2012 validation dataset.

• Journal of Practical Engineering Education
• /
• v.6 no.1
• /
• pp.51-55
• /
• 2014

A case study on engineering clinic operation based on industry needs is introduced. Engineering clinic is a course that students and professor solve bottleneck techniques provided from an industry. The industry presents the bottleneck techniques to the professor and the professor plans a course that the students learn how to solve them, and the students train field adaptability by solving them. From the course evaluation of the engineering clinic, the students give high scores to the awareness of the course objectives, the performance period, the smooth communication, the application and understanding of major, the problem solving skill, the cooperation ability, the opportunity of carrier choice, and the course recommendation. Two semesters give higher satisfaction to the students than one semester because two semesters are long enough to solve the bottleneck techniques provided from the industry. It gives good opportunity that the students get a job through completing the course.

• KIPS Transactions on Software and Data Engineering
• /
• v.9 no.10
• /
• pp.309-316
• /
• 2020

We predict the mortality of the elderly patients visiting the emergency department who are over 65 years old using Feed Forward Neural Network (FFNN) and Convolutional Neural Network (CNN) respectively. Medical data consist of 99 features including basic information such as sex, age, temperature, and heart rate as well as past history, various blood tests and culture tests, and etc. Among these, we used random forest to select features by measuring the importance of features in the prediction of mortality. As a result, using the top 80 features with high importance is best in the mortality prediction. The performance of the FFNN and CNN is compared by using the selected features for training each neural network. To train CNN with images, we convert medical data to fixed size images. We acquire better results with CNN than with FFNN. With CNN for mortality prediction, F1 score and the AUC for test data are 56.9 and 92.1 respectively.

• Journal of the Korean Society for Railway
• /
• v.12 no.5
• /
• pp.641-648
• /
• 2009

Concrete structures become more common in railway systems with an advancement of high speed train technologies. As the service life of concrete structures increases, structural strengthening for concrete structures may be necessary. There are several typical strengthening techniques using steel plate and fiber reinforced polymer (FRP) materials, which have their own inherent shortcomings. In order to enhance greater durability and resistance to fire and other environmental attacks, basalt fiber material attracts engineer's attention due to its characteristics. This study investigates bonding performance of basalt fiber sheet as a structural strengthening material. Experimental variables include bond width, length and number of layer. From the bonding tests, there were three different types of bonding failure modes: debonding, rupture and rip-off. Among the variables, bond width indicated more significant effect on bonding characteristics. In addition the bond length did not contribute to bond strength in proportion to the bond length. Hence this study evaluated effective bond length and effective bond strength. The effective bond strength was compared to those suggested by other researches which used different types of FRP strengthening materials such as carbon FRP.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.26 no.4
• /
• pp.259-266
• /
• 2016

Recently, CCTV cameras are emplaced actively to reinforce security and intelligent surveillance systems have been under development for detecting and monitoring of the objects in the video. In this study, we propose a method for detection of upper body in intelligent surveillance system using FCM-based RBFNN classifier realized with the aid of HOG features. Firstly, HOG features that have been originally proposed to detect the pedestrian are adopted to train the unique gradient features about upper body. However, HOG features typically exhibit a very high dimension of which is proportional to the size of the input image, it is necessary to reduce the dimension of inputs of the RBFNN classifier. Thus the well-known PCA algorithm is applied prior to the RBFNN classification step. In the computer simulation experiments, the RBFNN classifier was trained using pre-classified upper body images and non-person images and then the performance of the proposed classifier for upper body detection is evaluated by using test images and video sequences.

• Journal of Digital Convergence
• /
• v.16 no.12
• /
• pp.293-299
• /
• 2018

The weather has a lot of influence on the cultivation of crops. Weather information on agricultural crop cultivation areas is indispensable for efficient cultivation and management of agricultural crops. Despite the high demand for agricultural weather, research on this is in short supply. In this research, we deal with the production method of agricultural weather in Jeollanam-do, which is the main production area of onions through GloSea5 and deep learning. A deep neural network model using the sliding window method was used and utilized to train daily weather prediction for predicting the agricultural weather. RMSE and MAE are used for evaluating the accuracy of the model. The accuracy improves as the learning period increases, so we compare the prediction performance according to the learning period and the prediction period. As a result of the analysis, although the learning period and the prediction period are similar, there was a limit to reflect the trend according to the seasonal change. a modified deep layer neural network model was presented, that applying the difference between the predicted value and the observed value to the next day predicted value.

• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.41 no.4
• /
• pp.42-49
• /
• 2018

Pilot Aptitude Research Equipment (PARE) is a simulator developed to measure or research pilot aptitude and train for student pilots. Design of an ergonomic PARE operation console is required to operate the equipment effectively. This study carried out five steps : (S1) operator questionnaire survey, (S2) anthropometric design formula development, (S3) usability evaluation, (S4) improvement design, and (S5) validation considering both Physical User Interface (PUI) and Graphic User Interface (GUI) of PARE operation console. The operator questionnaire surveyed needs for each PUI and GUI part of the console from two PARE actual operators. In terms of PUI, the anthropometric design formula was developed by using design variables, body dimensions, target population characteristics, and reference posture related to the PARE console. In terms of GUI, the usability evaluation was conducted by three usability testing experts with a 7-point scale (1 : very low, 4 : neutral, 7 : very high) on GUI of the PARE operation console by seven usability criteria. The improved PARE operation console was designed to reflect the optimal values of design variables calculated from design formula, the results from usability testing, and the operator's needs. The improvement effect was observed by 20 people who had experience with the PARE operation console. As a result of the validation, monitor visibility and cockpit visibility for the improved PUI design and visibility and efficiency for the improved GUI design were significantly increased by more than 90% respectively. The improved design of the PARE operation console in this study can contribute to enhance operation performance of the PARE.

• Korean Journal of Plant Resources
• /
• v.35 no.2
• /
• pp.192-200
• /
• 2022

Nitrogen is the most essential macronutrient for the growth of fruit trees and is important factor determining the fruit yield. In order to produce high-quality fruits, it is necessary to supply the appropriate nitrogen fertilizer at the right time. For this, it is a prerequisite to accurately diagnose the nitrogen status of fruit trees. The fastest and most accurate way to determine the nitrogen deficiency of fruit trees is to measure the nitrogen concentration in leaves. However, it is not easy for citrus growers to measure nitrogen concentration through leaf analysis. In this study, several machine learning models were developed to classify the nitrogen deficiency based on the concentration measurement of mineral nutrients in the leaves of tangor Shiranuhi (Citrus unshiu × C. sinensis). The data analyzed from the leaves were increased to about 1,000 training dataset through the bootstrapping method and used to train the models. As a result of testing each model, gradient boosting model showed the best classification performance with an accuracy of 0.971.

• Journal of IKEEE
• /
• v.26 no.3
• /
• pp.341-348
• /
• 2022

Wi-Fi Sensing, which uses Wi-Fi technology to sense the surrounding environments, has strong potentials in a variety of sensing applications. Recently several advanced deep learning-based solutions using CSI (Channel State Information) data have achieved high performance, but it is still difficult to use in practice without explicit data collection, which requires expensive adaptation efforts for model retraining. In this study, we propose a Channel State Information Automatic Labeling System (CALS) that automatically collects and labels training CSI data for deep learning-based Wi-Fi sensing systems. The proposed system allows the CSI data collection process to efficiently collect labeled CSI for labeling for supervised learning using computer vision technologies such as object detection algorithms. We built a prototype of CALS to demonstrate its efficiency and collected data to train deep learning models for detecting the presence of a person in an indoor environment, showing to achieve an accuracy of over 90% with the auto-labeled data sets generated by CALS.