• Smart Structures and Systems
• /
• v.31 no.4
• /
• pp.383-392
• /
• 2023

Post-earthquake building condition assessment is crucial for subsequent rescue and remediation and can be automated by emerging computer vision and deep learning technologies. This study is based on an endeavour for the 2nd International Competition of Structural Health Monitoring (IC-SHM 2021). The task package includes five image segmentation objectives - defects (crack/spall/rebar exposure), structural component, and damage state. The structural component and damage state tasks are identified as the priority that can form actionable decisions. A multi-task Convolutional Neural Network (CNN) is proposed to conduct the two major tasks simultaneously. The rest 3 sub-tasks (spall/crack/rebar exposure) were incorporated as auxiliary tasks. By synchronously learning defect information (spall/crack/rebar exposure), the multi-task CNN model outperforms the counterpart single-task models in recognizing structural components and estimating damage states. Particularly, the pixel-level damage state estimation witnesses a mIoU (mean intersection over union) improvement from 0.5855 to 0.6374. For the defect detection tasks, rebar exposure is omitted due to the extremely biased sample distribution. The segmentations of crack and spall are automated by single-task U-Net but with extra efforts to resample the provided data. The segmentation of small objects (spall and crack) benefits from the resampling method, with a substantial IoU increment of nearly 10%.

• Proceedings of the IEEK Conference
• /
• 2000.07a
• /
• pp.450-453
• /
• 2000

In this paper, an automated segmentation algorithm is proposed for MR brain images using T1-weighted, T2-weighted, and PD images complementarily. The proposed segmentation algorithm is composed of 3 steps. In the first step, cerebrum images are extracted by putting a cerebrum mask upon the three input images. In the second step, outstanding clusters that represent inner tissues of the cerebrum are chosen among 3-dimensional (3D) clusters. 3D clusters are determined by intersecting densely distributed parts of 2D histogram in the 3D space formed with three optimal scale images. Optimal scale image best describes the shape of densely distributed parts of pixels in 2D histogram. In the final step, cerebrum images are segmented using FCM algorithm with it’s initial centroid value as the outstanding cluster’s centroid value. The proposed segmentation algorithm complements the defect of FCM algorithm, being influenced upon initial centroid, by calculating cluster’s centroid accurately And also can get better segmentation results from the proposed segmentation algorithm with multi spectral analysis than the results of single spectral analysis.

• MALSORI
• /
• no.58
• /
• pp.67-81
• /
• 2006

For a large corpus of time-aligned data, HMM based approaches are most widely used for automatic segmentation, providing a consistent and accurate phone labeling scheme. There are two methods for training in HMM. Flat starting method has a property that human interference is minimized but it has low accuracy. Bootstrap method has a high accuracy, but it has a defect that manual segmentation is required In this paper, a new algorithm is proposed to minimize manual work and to improve the performance of automatic segmentation. At first phase, voiced, unvoiced and silence classification is performed for each speech data frame. At second phase, the phoneme sequence is aligned dynamically to the voiced/unvoiced/silence sequence according to the acoustic phonetic rules. Finally, using these segmented speech data as a bootstrap, phoneme model parameters based on HMM are trained. For the performance test, hand labeled ETRI speech DB was used. The experiment results showed that our algorithm achieved 10% improvement of segmentation accuracy within 20 ms tolerable error range. Especially for the unvoiced consonants, it showed 30% improvement.

• Journal of Information Processing Systems
• /
• v.16 no.5
• /
• pp.1008-1014
• /
• 2020

Aiming at the defect detection quality of denim fabric, this paper designs an improved algorithm based on the optimized Gabor filter. Firstly, we propose an improved defect detection algorithm of jean fabric based on the maximum two-dimensional image entropy and the loss evaluation function. Secondly, 24 Gabor filter banks with 4 scales and 6 directions are created and the optimal filter is selected from the filter banks by the one-dimensional image entropy algorithm and the two-dimensional image entropy algorithm respectively. Thirdly, these two optimized Gabor filters are compared to realize the common defect detection of denim fabric, such as normal texture, miss of weft, hole and oil stain. The results show that the improved algorithm has better detection effect on common defects of denim fabrics and the average detection rate is more than 91.25%.

• Journal of the Semiconductor & Display Technology
• /
• v.19 no.3
• /
• pp.12-18
• /
• 2020

In semiconductor manufacturing, defect detection is critical to maintain high yield. Currently, computer vision systems used in semiconductor photo lithography still have adopt to digital image processing algorithm, which often occur inspection faults due to sensitivity to external environment. Thus, we intend to handle this problem by means of using Mask R-CNN instead of digital image processing algorithm. Additionally, Mask R-CNN can be trained with image dataset pre-processed by means of the specific designed digital image filter to extract the enhanced feature map of Convolutional Neural Network (CNN). Our approach converged advantage of digital image processing and instance segmentation with deep learning yields more efficient semiconductor photo lithography inspection system than conventional system.

• Journal of the Institute of Convergence Signal Processing
• /
• v.4 no.2
• /
• pp.25-30
• /
• 2003

Recently, development of digital technology is occupying a large part of multimedia information like character, voice, image, video, etc. Research about video indexing and retrieval progresses especially in research relative to video. In this paper, we propose new algorithm(Frame Feature Comparison) for MPEG video segmentation. Shot, Scene Change detection is basic and important works that segment it in MPEG video sequence. Generally, the segmentation algorithm that uses much has defect that occurs an error detection according to a flash of camera, movement of camera and fast movement of an object, because of comparing former frames with present frames. Therefore, we distinguish a scene change one more time using a scene change point detected in the conventional algorithm through comparing its mean value with abutted frames. In the result, we could detect more corrective scene change than the conventional algorithm.

• Journal of Korea Multimedia Society
• /
• v.24 no.8
• /
• pp.1026-1034
• /
• 2021

Tire wear and defect are important factors for safe driving condition. These defects are generally inspected by some specialized experts or very expensive equipments such as stereo depth camera and depth gauge. In this paper, we propose tire safety vision inspector based on deep neural network (DNN). The status of tire wear is categorized into three: 'safety', 'warning', and 'danger' based on depth of tire tread. We propose an attention mechanism for emphasizing the feature of tread area. The attention-based feature is concatenated to output feature maps of the last convolution layer of ResNet-101 to extract more robust feature. Through experiments, the proposed tire wear classification model improves 1.8% of accuracy compared to the existing ResNet-101 model. For detecting the tire defections, the developed tire defect detection model shows up-to 91% of accuracy using the Mask R-CNN model. From these results, we can see that the suggested models are useful for checking on the safety condition of working tire in real environment.

• Journal of Sensor Science and Technology
• /
• v.15 no.5
• /
• pp.328-333
• /
• 2006

Magnetic Flux Leakage (MFL) methods are widely employed for the nondestructive evaluation (NDE) of gas pipelines. In the application of MFL pipeline inspection technology, corrosion anomalies are detected and identified via their leakage filed due to changes in wall thickness. The gas industry is keenly interested in automating the interpretation process, because a large amount of data to be analyzed is generated for in-line inspection. This paper presents a novel approach to the tasks of data segmentation, feature extraction and depth estimation from gas pipelines. Also, we will show that the proposed method successfully identifying artificial defects.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.41 no.4
• /
• pp.69-74
• /
• 2004

In this paper, we propose a novel method for improving defect-detection performance based on reconstruction of line-scan camera images using both the projection profiles and color space transform. The proposed method consists of RGB region segmentation, representative value reconstruction using the tracing system, and Y image reconstruction using color-space transformation. Through experiments it is demonstrated that the performance using the reconstructed image is better than that using aerial image for LCD surface inspection.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2015.10a
• /
• pp.687-690
• /
• 2015

Forward collision warning systems(FCWS) and lane change assist systems(LCAS) need regions of interest for detecting lanes and objects as road regions. Watershed segmentation is effective algorithm that classify the road. That algorithm is split results appear differently depending on Watershed line with local minimum in the early part of the seed. If not road regions or vehicles combined the road's seed, It segment road with the others. For compensate the that defect, It has to adaptive change by road environment. The method is that image segmentate the several of regions of interest. Then It is set in a straight line that is detected in regions of interest. If It was detected cars on seed, seed is adjusted the location. And If It wasn't include the line, seed is adjusted the length for final decision the seed. We can detect the road region using the final seed that selected according to the road environment.