• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.1
• /
• pp.245-265
• /
• 2022

In response to problems such as insufficient extraction information, low detection accuracy, and frequent misdetection in the field of Thangka image defects, this paper proposes a YOLOv5 prediction algorithm fused with the attention mechanism. Firstly, the Backbone network is used for feature extraction, and the attention mechanism is fused to represent different features, so that the network can fully extract the texture and semantic features of the defect area. The extracted features are then weighted and fused, so as to reduce the loss of information. Next, the weighted fused features are transferred to the Neck network, the semantic features and texture features of different layers are fused by FPN, and the defect target is located more accurately by PAN. In the detection network, the CIOU loss function is used to replace the GIOU loss function to locate the image defect area quickly and accurately, generate the bounding box, and predict the defect category. The results show that compared with the original network, YOLOv5-SE and YOLOv5-CBAM achieve an improvement of 8.95% and 12.87% in detection accuracy respectively. The improved networks can identify the location and category of defects more accurately, and greatly improve the accuracy of defect detection of Thangka images.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.17 no.3
• /
• pp.129-138
• /
• 2022

Collecting and labeling sufficient training data, which is essential to deep learning-based visual inspection, is difficult for manufacturers to perform because it is very expensive. This paper presents a steel plate surface defect detection system with industrial-grade detection performance by training a small amount of steel plate surface images consisting of labeled and non-labeled data. To overcome the problem of lack of training data, we propose two data augmentation techniques: program-based augmentation, which generates defect images in a geometric way, and generative model-based augmentation, which learns the distribution of labeled data. We also propose a 4-step semi-supervised learning using pseudo labels and consistency training with fixed-size augmentation in order to utilize unlabeled data for training. The proposed technique obtained about 99% defect detection performance for four defect types by using 100 real images including labeled and unlabeled data.

• JSTS:Journal of Semiconductor Technology and Science
• /
• v.17 no.1
• /
• pp.86-93
• /
• 2017

In this paper, we propose a novel defect detection method using component tree representations of scanning electron microscopy (SEM) images. The component tree contains rich information about the topological structure of images such as the stiffness of intensity changes, area, and volume of the lobes. This information can be used effectively in detecting suspicious defect areas. A quasi-linear algorithm is available for constructing the component tree and computing these attributes. In this paper, we modify the original component tree algorithm to be suitable for our defect detection application. First, we exclude pixels that are near the ground level during the initial stage of component tree construction. Next, we detect significant lobes based on multiple attributes and edge information. Our experiments performed with actual SEM wafer images show promising results. For a $1000{\times}1000$ image, the proposed algorithm performed the whole process in 1.36 seconds.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.17 no.9
• /
• pp.2145-2152
• /
• 2013

Paper manufacturing industries have huge facilities with automatic equipments. Especially, in order to improve the efficiency of the paper manufacturing processes, it is necessary to detect the paper cutting defect effectively and to classify the causes correctly. In this paper, we review the problems of web monitoring system and web inspection system that have been traditionally used in industries for defect detection. Then we propose a novel paper cutting defect detection method based on the local binary pattern analysis and its implementation to mitigate the practical problems in industry environment. The proposed algorithm classifies the defects into edge-type and region-type and then it is shown that the proposed system works stably on the real paper cutting defect detection system.

• Proceedings of the KIEE Conference
• /
• 1999.07b
• /
• pp.668-670
• /
• 1999

We have developed a automatic surface inspection system for cold Rolled strips in steel making process for several years. We have experienced the various kinds of surface inspection systems, including linear CCD camera type and the laser type inspection system which was installed in cold rolled strips production lines. But, we did not satisfied with these inspection systems owing to insufficient detection and classification rate, real time processing performance and limited line speed of real production lines. In order to increase detection and computing power, we have used the Dark Field illumination with Infra_Red LED, Bright Field illumination with Xenon Lamp, Parallel Computing Processor with Area typed CCD camera and full software based image processing technique for the ease up_grading and maintenance. In this paper, we introduced the automatic inspection system and real time image processing technique using the Object Detection, Defect Detection, Classification algorithms. As a result of experiment, under the situation of the high speed processed line(max 1000 meter per minute) defect detection is above 90% for all occurred defects in real line, defect name classification rate is about 80% for most frequently occurred 8 defect, and defect grade classification rate is 84% for name classified defect.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.16 no.6
• /
• pp.369-377
• /
• 2023

In the industrial manufacturing sector, quality control is pivotal for minimizing defect rates; inadequate management can result in additional costs and production delays. This study underscores the significance of detecting texture defects in manufactured goods and proposes a more precise defect detection technique. While the DFR(Deep Feature Reconstruction) model adopted an approach based on feature map amalgamation and reconstruction, it had inherent limitations. Consequently, we incorporated a new loss function using statistical methodologies, integrated a skip connection structure, and conducted parameter tuning to overcome constraints. When this enhanced model was applied to the texture category of the MVTec-AD dataset, it recorded a 2.3% higher Defect Segmentation AUC compared to previous methods, and the overall defect detection performance was improved. These findings attest to the significant contribution of the proposed method in defect detection through the reconstruction of feature map combinations.

• Journal of the Semiconductor & Display Technology
• /
• v.22 no.1
• /
• pp.54-58
• /
• 2023

In this study, we compared the detection accuracy of the parameters of the scaffold failure detection model. A detection algorithm based on convolutional neural network was used to construct a failure detection model for scaffold. The parameter properties of the model were changed and the results were quantitatively verified. The detection accuracy of the model for each parameter was compared and the parameter with the highest accuracy was identified. We found that the activation function has a significant impact on the detection accuracy, which is 98% for softmax.

• Journal of Institute of Control, Robotics and Systems
• /
• v.19 no.12
• /
• pp.1094-1098
• /
• 2013

AVI (Automatic Vision Inspection) systems automatically detect defect features and measure their sizes via camera vision. Defect detection is not an easy process because of noises from various sources and optical distortion. In this paper the acquired images from a TFT panel are enhanced with the adoption of an HVS (Human Visual System). A human visual system is more sensitive on the defect area than the illumination components because it has greater sensitivity to variations of intensity. In this paper we modified an MTF (Modulation Transfer Function) in the Wavelet domain and utilized the characteristics of an HVS. The proposed algorithm flattens the inner illumination components while preserving the defect information intact.

• Journal of the Korean Society for Nondestructive Testing
• /
• v.30 no.3
• /
• pp.261-270
• /
• 2010

This paper presents an analytical investigation for a baseline-free imaging of a defect in plate-like structures using the time-reversal of Lamb waves. We first consider the flexural wave (A0 mode) propagation in a plate containing a defect, and reception and time reversal process of the output signal at the receiver. The received output signal is then composed of two parts: a directly propagated wave and a scattered wave from the defect. The time reversal of these waves recovers the original input signal, and produces two additional side bands that contain the time-of-flight information on the defect location. One of the side band signals is then extracted as a pure defect signal. A defect localization image is then constructed from a beamforming technique based on the time-frequency analysis of the side band signal for each transducer pair in a network of sensors. The simulation results show that the proposed scheme enables the accurate, baseline-free detection of a defect, so that experimental studies are needed to verify the proposed method and to be applied to real structure.

• Journal of the Korean Society for Precision Engineering
• /
• v.32 no.11
• /
• pp.989-995
• /
• 2015

In the literature, various stochastic anomaly detection methods, such as limit checking and PCA-based approaches, have been applied to weld defect detection. However, it is still a challenge to identify meaningful defect patterns from very limited sensor signals of laser welding, characterized by intermittent, discontinuous, very short, and non-stationary random signals. In order to effectively analyze the physical characteristics of laser weld signals: plasma intensity, weld pool temperature, and back reflection, we first transform the raw data of laser weld signals into the form of event logs. This is done by multidimensional discretization and event-codification, after which the event logs are decoded to extract weld defect patterns by $Na{\ddot{i}}ve$ Bayes classifier. The performance of the proposed method is examined in comparison with the commercial solution of PRECITEC's LWM$^{TM}$ and the most recent PCA-based detection method. The results show higher performance of the proposed method in terms of sensitivity (1.00) and specificity (0.98).