• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.22 no.5
• /
• pp.537-542
• /
• 2021

This paper improves the possibility of recognizing fast-moving objects through the YOLO (You Only Look Once) deep learning recognition model in an application environment for object recognition in images. The purpose was to study the method of collecting semantic data through processing. In the recognition model, the moving object recognition error was identified as unrecognized because of the difference between the frame rate of the camera and the moving speed of the object and a misrecognition due to the existence of a similar object in an environment adjacent to the object. To minimize the recognition errors by compensating for errors, such as unrecognized and misrecognized objects through the proposed data collection method, and applying vision processing technology for the causes of errors that may occur in images acquired for sports (tennis games) that can represent real similar environments. The effectiveness of effective secondary data collection was improved by research on methods and processing structures. Therefore, by applying the data collection method proposed in this study, ordinary people can collect and manage data to improve their health and athletic performance in the sports and health industry through the simple shooting of a smart-phone camera.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.15 no.4
• /
• pp.257-270
• /
• 2022

In this paper, we design a smart factory risk improvement system and risk improvement method using AI edge devices. The smart factory risk improvement system collects, analyzes, prevents, and promptly responds to the worker's work performance process in the smart factory using AI edge devices, and can reduce the risk that may occur during work with improving the defect rate when workers perfom jobs. In particular, based on worker image information, worker biometric information, equipment operation information, and quality information of manufactured products, it is possible to set an abnormal risk condition, and it is possible to improve the risk so that the work is efficient and for the accurate performance. In addition, all data collected from cameras and IoT sensors inside the smart factory are processed by the AI edge device instead of all data being sent to the cloud, and only necessary data can be transmitted to the cloud, so the processing speed is fast and it has the advantage that security problems are low. Additionally, the use of AI edge devices has the advantage of reducing of data communication costs and the costs of data transmission bandwidth acquisition due to decrease of the amount of data transmission to the cloud.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.10
• /
• pp.29-41
• /
• 2022

Recently, with the development of deep learning technology, a variety of huge models with excellent performance have been devised by pre-training massive amounts of text data. However, in order for such a model to be applied to real-life services, the inference speed must be fast and the amount of computation must be low, so the technology for model compression is attracting attention. Knowledge distillation, a representative model compression, is attracting attention as it can be used in a variety of ways as a method of transferring the knowledge already learned by the teacher model to a relatively small-sized student model. However, knowledge distillation has a limitation in that it is difficult to solve problems with low similarity to previously learned data because only knowledge necessary for solving a given problem is learned in a teacher model and knowledge distillation to a student model is performed from the same point of view. Therefore, we propose a heterogeneous knowledge distillation method in which the teacher model learns a higher-level concept rather than the knowledge required for the task that the student model needs to solve, and the teacher model distills this knowledge to the student model. In addition, through classification experiments on about 18,000 documents, we confirmed that the heterogeneous knowledge distillation method showed superior performance in all aspects of learning efficiency and accuracy compared to the traditional knowledge distillation.

• Proceedings of the Korean Society of Near Infrared Spectroscopy Conference
• /
• 2001.06a
• /
• pp.1251-1251
• /
• 2001

The routine analysis of milk chemical components is of major importance both for the management of animals in dairy farms and for quality control in dairy industries. NIRS technology is an analytical technique which greatly simplifies this routine. One of the most critical aspects in NIRS analysis of milk is sample preparation and analysis modes which should be fast and straightforward. An important difficulty when obtaining NIR spectra of milk is the high water content (80 to 90%) of this product, since water absorbs most of the infrared radiation, and, therefore, limits the accuracy of calibrating for other constituents. To avoid this problem, the DESIR system was set up. Other ways of radiation-sample interaction adapted for liquids or semi-liquids exist, which are practically instantaneous and with limited or null necessity of sample preparation: Transmission and Folded Transmission or Transflectance. The objective of the present work is to compare the precision and accuracy of milk calibration equations in two analysis modes: Reflectance (dry milk) and Folded Transmission (liquid milk). A FOSS-NIR Systems 6500 I spectrophotometer (400-2500 nm) provided with a spinning module was used. Two NIR spectroscopic methods for milk analysis were compared: a) folded transmission: liquid milk samples in a 0.1 pathlength sample cell (ref. IH-0345) and b) reflectance: dried milk samples in glass fibre filters placed in a standard ring cell. A set of 101 milk samples was used to develop the calibration equations, for the two NIR analysis modes, to predict casein, protein, fat and dry matter contents, and 48 milk samples to predict Somatic Cell Count (SCC). The calibrations obtained for protein, fat and dry matter have an excellent quantitative prediction power, since they present $r^2$ values higher than 0.9. The $r^2$ values are slightly lower for casein and SCC (0.88 and 0.89 respectively), but they still are sufficiently high. The accuracy of casein, protein and SCC equations is not affected by the analysis modes, since their ETVC values are very similar in reflectance and folded transmission (0.19% vs 0.21%; 0.16% vs 0.19% and 55.57% vs 53.11% respectively), Lower SECV values were obtained for the prediction of fat and dry matter with the folded transmission equations (0.14% and 0.25% respectively) compared to the results with the reflectance ones (0.43% and 0.34% respectively). In terms of accuracy and speed of analytical response, NIRS analysis of liquid milk is recommended (folded transmission), since the drying procedure takes 24 hours. However, both analysis modes offer satisfactory results.

• International conference on construction engineering and project management
• /
• 2022.06a
• /
• pp.1243-1244
• /
• 2022

In recent years, the growing interest in off-site construction has led to factories scaling up their manufacturing and production processes in the construction sector. Consequently, continuous large-scale site monitoring in low-variability environments, such as prefabricated components production plants (precast concrete production), has gained increasing importance. Although many studies on computer vision-based site monitoring have been conducted, challenges for deploying this technology for large-scale field applications still remain. One of the issues is collecting and transmitting vast amounts of video data. Continuous site monitoring systems are based on real-time video data collection and analysis, which requires excessive computational resources and network traffic. In addition, it is difficult to integrate various object information with different sizes and scales into a single scene. Various sizes and types of objects (e.g., workers, heavy equipment, and materials) exist in a plant production environment, and these objects should be detected simultaneously for effective site monitoring. However, with the existing object detection algorithms, it is difficult to simultaneously detect objects with significant differences in size because collecting and training massive amounts of object image data with various scales is necessary. This study thus developed a large-scale site monitoring system using edge computing and a small-object detection system to solve these problems. Edge computing is a distributed information technology architecture wherein the image or video data is processed near the originating source, not on a centralized server or cloud. By inferring information from the AI computing module equipped with CCTVs and communicating only the processed information with the server, it is possible to reduce excessive network traffic. Small-object detection is an innovative method to detect different-sized objects by cropping the raw image and setting the appropriate number of rows and columns for image splitting based on the target object size. This enables the detection of small objects from cropped and magnified images. The detected small objects can then be expressed in the original image. In the inference process, this study used the YOLO-v5 algorithm, known for its fast processing speed and widely used for real-time object detection. This method could effectively detect large and even small objects that were difficult to detect with the existing object detection algorithms. When the large-scale site monitoring system was tested, it performed well in detecting small objects, such as workers in a large-scale view of construction sites, which were inaccurately detected by the existing algorithms. Our next goal is to incorporate various safety monitoring and risk analysis algorithms into this system, such as collision risk estimation, based on the time-to-collision concept, enabling the optimization of safety routes by accumulating workers' paths and inferring the risky areas based on workers' trajectory patterns. Through such developments, this continuous large-scale site monitoring system can guide a construction plant's safety management system more effectively.

• Journal of the Korea Society of Computer and Information
• /
• v.14 no.9
• /
• pp.95-104
• /
• 2009

This paper proposes the noise robust algorithm to detect the starting point of music. Detection of starting point of music is necessary to solve computational-waste problem and retrieval-comparison problem with inconsistent input data in music content based retrieval system. In particular, such detection is even more necessary in time sequential retrieval method that compares data in the sequential order of time in contents based music retrieval system. Whereas it has the long point that the retrieval is fast since it executes simple comparison in the order of time, time sequential retrieval method has the short point that data starting time to be compared should be the same. However, digitalized music cannot guarantee the equity of starting time by bit rate conversion. Therefore, this paper ensured that recognition rate shall not decrease even while executing high speed retrieval by applying time sequential retrieval method through detection of music starting point in the pre-processing stage of retrieval. Starting point detection used minimum wave model that can detect effective sound, and for strength against noise, the noises existing in mute sound were swapped. The proposed algorithm was confirmed to produce about 38% more excellent performance than the results to which starting point detection was not applied, and was verified for the strength against noise.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.40 no.1
• /
• pp.15-22
• /
• 2022

In order to improve the accuracy of the deep learning object detection technique, the effect of magnification rate conditions and seasonal factors on detection accuracy in aerial photographs and drone images was analyzed through experiments. Among the deep learning object detection techniques, Mask R-CNN, which shows fast learning speed and high accuracy, was used to detect the vehicle to be detected in pixel units. Through Seoul's aerial photo service, learning images were captured at different screen magnifications, and the accuracy was analyzed by learning each. According to the experimental results, the higher the magnification level, the higher the mAP average to 60%, 67%, and 75%. When the magnification rates of train and test data of the data set were alternately arranged, low magnification data was arranged as train data, and high magnification data was arranged as test data, showing a difference of more than 20% compared to the opposite case. And in the case of drone images with a seasonal difference with a time difference of 4 months, the results of learning the image data at the same period showed high accuracy with an average of 93%, confirming that seasonal differences also affect learning.

• Asia-pacific Journal of Multimedia Services Convergent with Art, Humanities, and Sociology
• /
• v.8 no.11
• /
• pp.215-222
• /
• 2018

In this paper, the research on the energy storage system adapting super-capacitor has been performed. The most advanced features compared to the conventional lead-acid battery systems is that it can obtain high power capability due to the super capacitor power characteristics. The suggested system can attain high power in short times and achieve high power quality improvements. The application areas are power quality improvement system, motor start power which requires high power during transient times. The energy conversion system consists of bi-directional converter and inverter and advantages of high speed, high power charging and discharging performances. The design steps for the two loop controller of the bi-directional inverter are suggested and verified by the experiment and manufacturing. The two loop controller design starts from linearized transfer function which is calculated from the state averaging model including state decoupling method. The current controller requirements are 20% overshoot and settling time and voltage controller are no overshoot and settling time which is 10 times longer than current controller. The design is verified from the step input response. The designed controllers have unity power factor characteristics and thus can improve the power quality of the grid. It also has fast response time and zero steady state error.

• Journal of Marine Life Science
• /
• v.1 no.1
• /
• pp.25-30
• /
• 2016

We investigated the possibility of a bio-monitoring system for detecting hypoxic water in coastal area using shell valve movements of Pacific oyster (Crassostrea gigas), which showed most aquaculture production in Korea, with Hall element sensor. In filtrating water to confirm shell valve movement (SVM) under normal condition, it showed spikes which mean a relatively fast closing condition after opened condition of average 5~12 mm, and then the SVM showed back to opening condition slower than closing speed SVM numbers during light period were similar to that of dark period (p<0.05). When dissolved oxygen (DO) concentration was reduced from 7 mg l^-1 to 3 mg l^-1, SVM numbers were increasing with decreasing of DO, and showed abnormality SVMs as compare with normal condition. Moreover, in the condition of 2 mg l^-1, Distance between light and left shell showed gradually decreased, and then we could not detected SVMs due to closed condition. Thus, if we quickly detect abnormal environmental variations as hypoxia water using bio-monitoring of SVM, it may be contribute to increased productivity by dramatically reducing damages in aquaculture.

• Smart Media Journal
• /
• v.12 no.1
• /
• pp.9-16
• /
• 2023

Recently, a convolutional neural network (CNN) based system is being developed to overcome the limitations of human resources in the apple quality classification of farmhouse. However, since convolutional neural networks receive only images of the same size, preprocessing such as sampling may be required, and in the case of oversampling, information loss of the original image such as image quality degradation and blurring occurs. In this paper, in order to minimize the above problem, to generate a image patch based graph of an original image and propose a random walk-based positional encoding method to apply the graph transformer model. The above method continuously learns the position embedding information of patches which don't have a positional information based on the random walk algorithm, and finds the optimal graph structure by aggregating useful node information through the self-attention technique of graph transformer model. Therefore, it is robust and shows good performance even in a new graph structure of random node order and an arbitrary graph structure according to the location of an object in an image. As a result, when experimented with 5 apple quality datasets, the learning accuracy was higher than other GNN models by a minimum of 1.3% to a maximum of 4.7%, and the number of parameters was 3.59M, which was about 15% less than the 23.52M of the ResNet18 model. Therefore, it shows fast reasoning speed according to the reduction of the amount of computation and proves the effect.