• Journal of Multimedia Information System
• /
• v.6 no.4
• /
• pp.225-234
• /
• 2019

Deep learning has become increasingly popular in both academic and industrial areas nowadays. Various domains including pattern recognition, Computer vision have witnessed the great power of deep neural networks. However, current studies on deep learning mainly focus on quality data sets with balanced class labels, while training on bad and imbalanced data set have been providing great challenges for classification tasks. We propose in this paper a method of data analysis-based data reduction techniques for selecting good and diversity data samples from a large dataset for a deep learning model. Furthermore, data sampling techniques could be applied to decrease the large size of raw data by retrieving its useful knowledge as representatives. Therefore, instead of dealing with large size of raw data, we can use some data reduction techniques to sample data without losing important information. We group PCB characters in classes and train deep learning on the ResNet56 v2 and SENet model in order to improve the classification performance of optical character recognition (OCR) character classifier.

• 제어로봇시스템학회:학술대회논문집
• /
• 1988.10a
• /
• pp.353-359
• /
• 1988

In this paper, the design and implementation of a multi-processor based die bonder machine for the semiconductor will be described. This is a final research results carried out for two years from June, 1986 to July, 1988. The mechanical system consists of three subsystems such as bonding head module, wafer feeding module, and lead frame feeding module. The overall control system consists of the following three subsystems each of which employs a 16 bit microprocessor MC 68000 : (i) supervisory control system, (ii) visual recognition / inspection system and (iii) the display system. Specifically, the supervisory control system supervises the whole sequence of die bonder machine, performs a self-diagnostics while it controls the bonding head module according to the prespecified bonding cycle. The vision system recognizes the die to inspect the die quality and deviation / orientation of a die with respect to a reference position, while it controls the wafer feeding module. Finally, the display system performs a character display, image display ans various error messages to communicate with operator. Lead frame feeding module is controlled by this subsystem. It is reported that the proposed control system were applied to an engineering sample and tested in real-time, and the results are sucessful as an engineering sample phase.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.28 no.7
• /
• pp.1047-1054
• /
• 2004

Internal defects in the glass, like-as micro-voids, micro-cracks, or inclusions, easily cause the failure when the glass is exposed to the shock or the thermal variation. In order to produce the highly reliable glass product, the precision inspection of the defect in the glass is required. For this purpose, this paper proposes a machine vision technique based on the auto-focusing method, which searches the defect and measures the location under the fact that the edge image of defect must be the most clear when the focal plane of CCD camera is coincided with the defect. As for the search index, the gradient indicator is presented. The basic principles are verified through the simulations for the computer-generated defect images, where the affects of defect shape, gray level of background, and the brightness of the defect image are also analyzed. Finally, experimental results for actual glass specimens are shown to confirm the applicability of this method to the actual field.

• Journal of the Korean Society for Precision Engineering
• /
• v.31 no.9
• /
• pp.831-838
• /
• 2014

Chip on glass (COG) bonding using anisotropic conductive film (ACF) is a key technology to assemble a driver IC onto a LCD glass panel. In this paper, an experimental investigation was conducted to investigate the correlation between contact resistance and characteristics of image taken by machine vision based inspection system. The results show that the contact resistance was strongly influenced by the contrast ratio of conductive particle rather than the number of conductive particles. Also, number of conductive particles whose contrast ratio is below 0.75 is crucial for determining the quality of the assembled samples. On the other hand, in the result of high temperature high humidity storage test, the contrast ratio of samples was increased. However, in the case of open-circuit samples after temperature humidity storage test, the number of conductive particles whose contrast ratio is above 0.75 was more than that of the closed-circuit samples.

• Korean Journal of Remote Sensing
• /
• v.17 no.4
• /
• pp.319-334
• /
• 2001

Extracting primitives from imagery plays an important task in visual information processing since the primitives provide useful information about characteristics of the objects and patterns. The human visual system utilizes features without difficulty for image interpretation, scene analysis and object recognition. However, to extract and to analyze feature are difficult processing. The ultimate goal of digital image processing is to extract information and reconstruct objects automatically. The objective of this study is to develop robust method to achieve the goal of the image processing. In this study, an adaptive strategy was developed by implementing Gabor filters in order to extract feature information and to segment images. The Gabor filters are conceived as hypothetical structures of the retinal receptive fields in human vision system. Therefore, to develop a method which resembles the performance of human visual perception is possible using the Gabor filters. A method to compute appropriate parameters of the Gabor filters without human visual inspection is proposed. The entire framework is based on the theory of human visual perception. Digital images were used to evaluate the performance of the proposed strategy. The results show that the proposed adaptive approach improves performance of the Gabor filters for feature extraction and segmentation.

• International Journal of High-Rise Buildings
• /
• v.9 no.4
• /
• pp.351-360
• /
• 2020

Monitoring civil structures periodically is necessary for ensuring the fitness of the structures. Cracks on inner and outer surfaces of the building plays a vital role in indicating the health of the building. Conventionally, human visual inspection techniques were carried up to human reachable altitudes. Monitoring of high rise infrastructures cannot be done using this primitive method. Also, there is a necessity for more accurate prediction of cracks on building surfaces for ensuring the health and safety of the building. The proposed research focused on developing an efficient crack classification model using Transfer Learning enabled EfficientNet (TL-EN) architecture. Though many other pre-trained models were available for crack classification, they rely on more number of training parameters for better accuracy. The TL-EN model attained an accuracy of 0.99 with less number of parameters on large dataset. A bench marked METU dataset with 40000 images were used to test and validate the proposed model. The surfaces of high rise buildings were investigated using vision enabled Unmanned Arial Vehicles (UAV). These UAV is fabricated with TL-EN model schema for capturing and analyzing the real time streaming video of building surfaces.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2010.04a
• /
• pp.369-372
• /
• 2010

평판 디스플레이(flat panel display)의 크기가 커짐에 따라 다양한 기판을 이용한 제조 방법이 개발되고 있다. 디스플레이 제조 공정 중 기판의 결함을 찾아서 분류하는 검사 시스템은 최종 제품의 품질을 결정하는 매우 중요한 부분이다. 본 연구는 머신비전 기술을 이용하여 불투명하고 반사율이 높은 기판 표면의 결함을 찾아내고, 이 결함을 스크래치(scratch), 흑결함(dark defect), 백결함(white defect)으로 분류하는 장치를 구현하는데 목적이 있다. 이를 구현하기 위해 본 논문에서는 정밀 스테이지(stage)와 라인 카메라(line CCD camera)을 이용한 광학계를 활용하여 검사 시스템을 구현하였다. 구축된 시스템을 이용하여 취득한 이미지를 12 개의 영역으로 등분하여 각각의 국부 영역에 대한 문턱값 연산(thresholding)을 적용함으로써 조명의 불균일을 의한 검출 에러율을 획기적으로 낮추었다. 간단한 컴퓨터비전 알고리듬의 채용으로도 검사 시스템의 구현이 가능함을 보였다.

• Journal of Internet Computing and Services
• /
• v.16 no.6
• /
• pp.97-107
• /
• 2015

Up until recently, Korea has ranked the first place in UN e-Government Survey for three consecutive years. In keeping with such accomplishment, the size of budget execution has been consistently growing in accordance with Korea's Government 3.0 policy and vision, leading to increase in big-sized informatization projects in the business. Especially in mission critical public sector's infrastructure where it affects many people, growing demand for establishing high-quality information system with new technologies being brought to attention in order to meet the complex needs of citizens. National defense information system, being one of representative domains examples in the concerned area, established high military competency by applying breakthrough technology. Network-oriented national defense knowledge informatization was set as the vision in order to implement core roles in making efficient national defense management; and effort has been made to materialize the vision by making advancement in national defense's information system and its informatization implementation system. This research studies new quality index relevant to test and evaluation (T&E)of informatization business in national defense which is the representative example of mission critical public sector's infrastructure. We studied international standards and guidelines, analyzed actual T&E cases, and applied them to the inspection items that are currently in use, complying with the e-government law (Act No. 12346, Official Announcement Date 2014. 1.28., Enforcement Date 2014. 7.29.) As a result of productivity analysis, based on hypothesis in which suggested model was applied to T&E of the national defense informatization business, we confirmed the possibility of enhancement in the T&E productivity by assessing reliability, expertise, and safety as evaluation factors.

• Journal of the Korean Wood Science and Technology
• /
• v.37 no.1
• /
• pp.37-47
• /
• 2009

For the inspection of wood, machine vision is the most common automated inspection method used at present. It is required to sort wood products by grade and to locate surface defects prior to cut-up. Many different sensing methods have been applied to inspection of wood including optical, ultrasonic, X-ray sensing in the wood industry. Nowadays the scanning system mainly employs CCD line-scan camera to meet the needs of accurate detection of lumber defects and real-time image processing. But this system needs exact feeding system and low deviation of lumber thickness. In this study low cost CCD area sensor was used for the development of image processing system for lumber being fed. When domestic red pine being fed on the conveyer belt, lumber images of irregular term of captured area were acquired because belt conveyor slipped between belt and roller. To overcome incorrect image merging by the unstable feeding speed of belt conveyor, it was applied template matching algorithm which was a measure of the similarity between the pattern of current image and the next one. Feeding the lumber over 13.8 m/min, general area sensor generates unreadable image pattern by the motion blur. The red channel of RGB filter showed a good performance for removing background of the green conveyor belt from merged image. Threshold value reduction method that was a image-based thresholding algorithm performed well for knot detection.

• Journal of the Korea institute for structural maintenance and inspection
• /
• v.27 no.5
• /
• pp.113-119
• /
• 2023

Most infrastructure structures were completed during periods of economic growth. The number of infrastructure structures reaching their lifespan is increasing, and the proportion of old structures is gradually increasing. The functions and performance of these structures at the time of design may deteriorate and may even lead to safety accidents. To prevent this repercussion, accurate inspection and appropriate repair are requisite. To this end, demand is increasing for computer vision and deep learning technology to accurately detect even minute cracks. However, deep learning algorithms require a large number of training data. In particular, label images indicating the location of cracks in the image are required. To secure a large number of those label images, a lot of labor and time are consumed. To reduce these costs as well as increase detection accuracy, this study proposed a learning structure based on mean teacher method. This learning structure was trained on a dataset of 900 labeled image dataset and 3000 unlabeled image dataset. The crack detection network model was evaluated on over 300 labeled image dataset, and the detection accuracy recorded a mean intersection over union of 89.23% and an F1 score of 89.12%. Through this experiment, it was confirmed that detection performance was improved compared to supervised learning. It is expected that this proposed method will be used in the future to reduce the cost required to secure label images.