• Journal of the Korea Convergence Society
• /
• v.6 no.1
• /
• pp.1-5
• /
• 2015

In this paper, we describe the design of a fusion labeling system which is controlled by the Wi-Fi signals. The Current labeling system which is used in the industry is designed to work independently on the production line not connected with internet network services. For such reasons, it is very inconvenient for the labeling system to transfer such labeling data of the production line to the server computer. We propose a labeling system connected to the Wi-Fi service being able to send real-time transmission of labeling data. This system can supply the labeling data of production line to the server computer in realtime and improve the production quality than the existing system.

• Database Research
• /
• v.34 no.3
• /
• pp.137-147
• /
• 2018

A key to a well-performing human activity recognition (HAR) system through machine learning technique is the availability of a substantial amount of labeled data. Collecting sufficient labeled data is an expensive and time-consuming task. To build a HAR system in a new environment (i.e., the target domain) with very limited labeled data, it is unfavorable to naively exploit the data or trained classifier model from the existing environment (i.e., the source domain) as it is due to the domain difference. While traditional machine learning approaches are unable to address such distribution mismatch, transfer learning approach leverages the utilization of knowledge from existing well-established source domains that help to build an accurate classifier in the target domain. In this work, we propose a transfer learning approach to create an accurate HAR classifier with very limited data through the multitask neural network. The classifier loss function minimization for source and target domain are treated as two different tasks. The knowledge transfer is performed by simultaneously minimizing the loss function of both tasks using a single neural network model. Furthermore, we utilize the unlabeled data in an unsupervised manner to help the model training. The experiment result shows that the proposed work consistently outperforms existing approaches.

• Journal of Advanced Navigation Technology
• /
• v.22 no.1
• /
• pp.20-26
• /
• 2018

In this paper, we propose a moving object detection and tracking algorithm based on multi-frame feature point tracking information to reduce false positives. However, there are problems of detection error and tracking speed in existing studies. In order to compensate for this, we first calculate the corner feature points and the optical flow of multiple frames for camera movement compensation and object tracking. Next, the tracking error of the optical flow is reduced by the multi-frame forward-backward tracking, and the traced feature points are divided into the background and the moving object candidate based on homography and RANSAC algorithm for camera movement compensation. Among the transformed corner feature points, the outlier points removed by the RANSAC are clustered and the outlier cluster of a certain size is classified as the moving object candidate. Objects classified as moving object candidates are tracked according to label tracking based data association analysis. In this paper, we prove that the proposed algorithm improves both precision and recall compared with existing algorithms by using quadrotor image - based detection and tracking performance experiments.

• Journal of Nutrition and Health
• /
• v.50 no.4
• /
• pp.383-390
• /
• 2017

Purpose: This study investigated the importance and performance of food/nutrition labeling. The aim was to determine how important students consider food nutritional labeling, utilization of nutrition labels in daily life, and consumer satisfaction of current nutritional labeling. Methods: This study was conducted using a primary survey of students at one high school in Seoul. A total of 300 of 382 questionnaires were analyzed. Results: Regarding difference analysis of the importance-performance of food/nutrition labeling, importance showed higher scores than performance in all 10 attributes. According to the results of Importance and Performance Analysis (IPA), 'health, weight control and maintenance, proper dietary habits, and personal satisfaction' displayed both high importance and performance in the first quadrant. Importance of two factors (health and nutritional factor and effects of media and education) of the 10 attributes positively influenced overall satisfaction in the multiple regression analysis. Conclusion: To develop healthier food choices, it is necessary to educate adolescents about food/nutrition labeling and improve the food/nutrition labeling system.

• Journal of Korea Society of Industrial Information Systems
• /
• v.21 no.1
• /
• pp.27-34
• /
• 2016

In this paper, we propose how to recognize a multi-touch in the virtual touch type. Virtual touch has an advantage that it is installed only simple depth camera compared to the physical touch manners and it can be implemented with low cost for extracting an object exactly from only the difference of the depth values between the object and background. However, the accuracy for implementing the multi-touch has lowered. This paper presents a method to increase the accuracy of the multi-touch through the algorithms of binarization, labelling, and object tracking for multi-object recognition. Simulation results show that the proposed method can provide a variety of multi-touch events.

• Proceedings of the KSRS Conference
• /
• 2007.03a
• /
• pp.167-170
• /
• 2007

연안 김 양식장의 효과적 관리를 위해서는 실제 시설량의 조사가 펼요하며 인공위성을 이용한 방법이 가장 효과적이다. 본 연구에서는 스펙트로미터에 의한 해수 및 김 양식장 시설에 대한 광 측정을 통하여 파장별 특성을 조사하였다 10m의 해상도를 갖고 있는 SPOT-5 다중분광영상을 사용하였으며，김 양식장의 자동탐지알고리듬의 개발을 위하여 경기도 화성시 제부도 남방해역에 대한 2005년도 영상을 사용하였다. 김 양식장을 추출하기 위하여 우선 3밴드 영상의 분광특성을 이용한 밴드차(Band difference) 영상을 작성하여，두 가지 방법(형태학적 처리기법 및 Canny 에지 탐지기법)으로 처리를 한 후，두 결과를 합성하여 라벨령함으로써 탐지율을 극대화하였다 양식장 시설 현황 조사 결과는，정부에서 전체 생산량을 조절할 수 있게 하며，양식업자가 좋은 수확을 달성하는데 도움이 될 수 있을 것이다.

• Journal of Korea Multimedia Society
• /
• v.16 no.5
• /
• pp.549-557
• /
• 2013

This paper proposes a method using the multiple threshold regions in the image preprocessing procedure for container identifier recognition system. The multiple threshold regions are set by considering the container image characteristics and used as the candidates for the final one, The image is transformed to black and white images using these threshold regions, then labeling, panelling and panels merging are executed for each candidate, respectively. Finally the best threshold region is selected through this procedure and the character region can be extracted. Applying the similar method the noises are removed and the characters of identifier are segmented from the extracted region. In the experiments with 162 different images the success rates for extracting of the character region and segmenting the characters are 99.04% and 98.09%, respectively.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.25 no.5
• /
• pp.841-849
• /
• 2001

The pixel information of the object was obtained sequentially and pixels were clustered to a label by the line labeling method. Feature points were determined by finding the slope for edge pixels after selecting the fixed number of edge pixels. The slope was estimated by the least square method to reduce the detection error. Once a matching point was determined by comparing the feature information of the object and the pattern, the parameters for translation, scaling and rotation were obtained by selecting the longer line of the two which passed through the matching point from left and right sides. Finally, modified Hausdorff Distance has been used to identify the similarity between the object and the given pattern. The multi-label method was developed for recognizing the patterns with more than one label, which performs the modified Hausdorff Distance twice. Experiments have been performed to verify the performance of the proposed algorithm and method for simple target image, complex target image, simple pattern, and complex pattern as well as the partially hidden object. It was proved via experiments that the proposed image matching algorithm for recognizing the object had a good performance of matching.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2022.05a
• /
• pp.154-157
• /
• 2022

In the existing Tensorflow CNN environment, the object detection method is a method of performing object labeling and detection by Tensorflow itself. However, with the advent of YOLO, the efficiency of image object detection has increased. As a result, more deep layers can be built than existing neural networks, and the image object recognition rate can be increased. Therefore, in this paper, the detection ability and speed were compared and analyzed by designing an object detection system based on Darknet and YOLO and performing multi-layer construction and learning based on the existing convolutional neural network. For this reason, in this paper, a neural network methodology that efficiently uses Darknet's learning is presented.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.18-22
• /
• 2019

FAQ 분류는 자주 묻는 질문을 범주화하고 사용자 질의에 대해 가장 유사한 클래스를 추론하는 방식으로 진행된다. FAQ 데이터셋은 클래스가 다수 존재하기 때문에 클래스 간 포함 및 연관 관계가 존재하고 특정 데이터가 서로 다른 클래스에 동시에 속할 수 있다는 특징이 있다. 그러나 최근 FAQ 분류는 다중 클래스 분류 방법론을 적용하는 데 그쳤고 FAQ 데이터셋의 특징을 모델에 반영하는 연구는 미미했다. 현 분류 방법론은 이러한 FAQ 데이터셋의 특징을 고려하지 못하기 때문에 정답으로 해석될 수 있는 예측도 오답으로 여기는 경우가 발생한다. 본 논문에서는 신뢰성이 부족한 FAQ 데이터셋에서도 분류를 잘 하기 위해 손실 함수를 조정하는 정규화 기법을 소개한다. 이 정규화 기법은 클래스 간 포함 및 연관 관계를 반영할 수 있도록 오답을 예측한 경우에도 예측 강도에 비례하여 손실을 줄인다. 이는 오답을 높은 확률로 예측할수록 데이터의 신뢰성이 낮을 가능성이 크다고 판단하여 학습을 강하게 하지 않게 하기 위함이다. 실험을 위해서는 다중 클래스 분류에서 가장 좋은 성능을 보이고 있는 모형인 BERT를 이용했으며, 비교 실험을 위한 정규화 방법으로는 통상적으로 사용되는 라벨 스무딩을 채택했다. 실험 결과, 본 연구에서 제안한 방법은 기존 방법보다 성능이 개선되고 보다 안정적으로 학습이 된다는 것을 확인했으며, 데이터의 신뢰성이 부족한 상황에서 효과적으로 분류를 수행함을 알 수 있었다.