• IEMEK Journal of Embedded Systems and Applications
• /
• v.17 no.2
• /
• pp.77-83
• /
• 2022

In this paper, the two-stage object detection approach is proposed to implement a deep learning-based label inspection system on edge computing environments. Since the label printed on the products during the production process contains important information related to the product, it is significantly to check the label information is correct. The proposed system uses the lightweight deep learning model that able to employ in the low-performance edge computing devices, and the two-stage object detection approach is applied to compensate for the low accuracy relatively. The proposed Two-Stage object detection approach consists of two object detection networks, Label Area Detection Network and Character Detection Network. Label Area Detection Network finds the label area in the product image, and Character Detection Network detects the words in the label area. Using this approach, we can detect characters precise even with a lightweight deep learning models. The SF-YOLO model applied in the proposed system is the YOLO-based lightweight object detection network designed for edge computing devices. This model showed up to 2 times faster processing time and a considerable improvement in accuracy, compared to other YOLO-based lightweight models such as YOLOv3-tiny and YOLOv4-tiny. Also since the amount of computation is low, it can be easily applied in edge computing environments.

• Journal of the Korea Society of Computer and Information
• /
• v.29 no.2
• /
• pp.31-41
• /
• 2024

Deep learning models show excellent performance in tasks such as image classification and object detection in the field of computer vision, and are used in various ways in actual industrial sites. Recently, research on improving robustness has been actively conducted, along with pointing out that this deep learning model is vulnerable to hostile examples. A hostile example is an image in which small noise is added to induce misclassification, and can pose a significant threat when applying a deep learning model to a real environment. In this paper, we tried to confirm the robustness of the edge-learning classification model and the performance of the adversarial example detection model using it for adversarial examples of various algorithms. As a result of robustness experiments, the basic classification model showed about 17% accuracy for the FGSM algorithm, while the edge-learning models maintained accuracy in the 60-70% range, and the basic classification model showed accuracy in the 0-1% range for the PGD/DeepFool/CW algorithm, while the edge-learning models maintained accuracy in 80-90%. As a result of the adversarial example detection experiment, a high detection rate of 91-95% was confirmed for all algorithms of FGSM/PGD/DeepFool/CW. By presenting the possibility of defending against various hostile algorithms through this study, it is expected to improve the safety and reliability of deep learning models in various industries using computer vision.

• Journal of the Institute of Convergence Signal Processing
• /
• v.4 no.2
• /
• pp.1-6
• /
• 2003

The study is to automatically or semi-automatically detect the accurate contour of tumors or lesions using active contour models (Snakes) in the MRI images of the brain. In the study we have improved the energy-minimization problem of snakes using dynamic programming and have utilized the values of the canny edge detector by the image force to make the snake less sensitive in noises. For the extracted boundary, the inside area, the perimeter and its center coordinates could be calculated. In addition, the multiple 2D slices with the contour of the lesion wore combined to visualized the shape of the lesion in 3D. We expect that the proposed method in this paper will be useful to make a treatment plan as well as to evaluate the treatments.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.9 no.3
• /
• pp.369-376
• /
• 2014

Recent climate change and ensure each country's marine fisheries resources due to the sharp decline to address the eco-friendly farming has emerged. Marine aquaculture operations to provide ease of use for the fishing ship cranes. In this paper, we use the default handler for the shipping cranes and improve the working environment of the received video equipment in the work area through the monitoring and analysis of visual information to optimize the image, convenient and high visibility for workers to have real-time video guide system is proposed.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.5 no.4
• /
• pp.856-873
• /
• 2011

Reversible watermarking is an open problem in information hiding field, with embedding the encoded bit '1' or '0' into some sensitive images, such as the law enforcement, medical records and military images. The technique can retrieve the original image without distortion, after the embedded message has been extracted. Histogram-based scheme is a remarkable breakthrough in reversible watermarking schemes, in terms of high embedding capacity and low distortion. This scheme is lack of capacity control due to the requirement for embedding large-scale data, because the largest hidden capacity is decided by the amount of pixels with the peak point. In this paper, we propose a reversible watermarking scheme to enlarge the number of pixels with the peak point as large as possible. This algorithm is based on an adaptive edge-guided interpolation, furthermore, hides messages by interpolation-error, i.e. the difference between the original and interpolated image value. Simulation results compared with other state-of-the-art reversible watermarking schemes in this paper demonstrate the validity of the proposed algorithm.

• Journal of Korea Society of Industrial Information Systems
• /
• v.9 no.2
• /
• pp.59-64
• /
• 2004

Most features of nature and phenomena we encounter in many branches of science are inherently very irregular and have fractal aspects. Thus the analysis of them with the traditional methods such as a differential operator may result in their ill-posed problems. To settle these problems, one may use several type of mean filters which smooth the input signal. However when a given function or data are complex in their nature, there may be loss of some original information during these process. In this paper, we utilized the tangent plane method instead of mean filters for the purpose of less loss of information and more smoothness. After then we attempt to take more accurate edges for the irregular image on the basis of the Otzu threshold. Finally we introduce the effective edge extracting method which use the fractal dimension representing the complexity of the given image.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2010.10a
• /
• pp.196-199
• /
• 2010

The human eyes are important facial landmarks for image normalization due to their relatively constant interocular distance. This paper introduces a novel approach for the eye detection task using optimal segmentation method for eye representation. The method consists of three steps: (1)edge extraction method that can be used to accurately extract eye region from the gray-scale face image, (2)extraction of eye region using labeling method, (3)eye localization based on intensity information. Experimental results show that a correct eye detection rate of 98.9% can be achieved on 2408 FERET images with variations in lighting condition and facial expressions.

• JSTS:Journal of Semiconductor Technology and Science
• /
• v.13 no.5
• /
• pp.511-515
• /
• 2013

A Monte Carlo (MC) simulation study has been done in order to investigate the effects of line-edge-roughness (LER) induced by either 1P1E (single-patterning and single-etching) or 2P2E (double-patterning and double-etching) on fully-depleted silicon-on-insulator (FDSOI) tri-gate metal-oxide-semiconductor field-effect transistors (MOSFETs). Three parameters for characterizing the LER profile [i.e., root-mean square deviation (${\sigma}$), correlation length (${\zeta}$), and fractal dimension (D)] are extracted from the image-processed scanning electron microscopy (SEM) image for each photolithography method. It is experimentally verified that two parameters (i.e., ${\sigma}$ and D) are almost the same in each case, but the correlation length in the 2P2E case is longer than that in the 1P1E case. The 2P2E-LER-induced $V_TH$ variation in FDSOI tri-gate MOSFETs is smaller than the 1P1E-LER-induced $V_TH$ variation. The total random variation in $V_TH$, however, is very dependent on the other major random variation sources, such as random dopant fluctuation (RDF) and work-function variation (WFV).

• Journal of the Korean Society for Precision Engineering
• /
• v.15 no.10
• /
• pp.34-40
• /
• 1998

The burr produced on piece part edges in machining operations must be removed for most parts to function effectively. Although considerable cost have been expended in improving deburring methods, little energy has been applied toward minimizing burrs. This study has been carried out to prevent the burrs produced on pure aluminium under various working condition and specimen shape in turning operations. The computer image processing system was used for measurement of size of burr, such as burr length, burr depth and burr area. The size of burr showed a decreasing tendency with the increase of rake angle and side cutting angle but it increased rapidly with the increase of depth of cut and the cutting speed has no effect on size of burrs. The size of burr rapidly decreased with the increase of edge angle and burrs are not occurred if edge angle is over 80$^{\circ}$.

• Journal of the Korean Society for Aviation and Aeronautics
• /
• v.11 no.2
• /
• pp.33-42
• /
• 2003

In this paper, we propose an enhancement of the ultrasonic image for non-destructive inspection of aircraft composite materials. The ultrasonic images are corrupted by a speckle noise which has the characteristic of granular pattern noise and is in all types of coherent imaging systems, the signal independent and multiplicative noise. In this paper, we derive a filter, called the AWLF(Adaptive Window Log Filter), from the nature of the speckle. The filter is made of the MEAN Filter in the edge region and Log Filter in the flat or noise region. To make a distinction between edge and flat region, we calculate the inclination around the local window instead of computing the local statistics of pixels such as local mean ${\bar{M}}$ and standard deviation ${\sigma}_s$. According to the obtained region, edge region is performed by the mean filter and flat region by the Log filter. Performance of the proposed filter is evaluated by the Enhanced Factor$(F_e)$ and the Speckle Index(SI).