• Journal of Broadcast Engineering
• /
• v.26 no.4
• /
• pp.419-428
• /
• 2021

Shadow removal plays a key role for the pre-processing of image processing techniques such as object tracking and detection. With the advances of image recognition based on deep convolution neural networks, researches for shadow removal have been actively conducted. In this paper, we propose a novel method for shadow removal, which utilizes self attention distillation to extract semantic features. The proposed method gradually refines results of shadow detection, which are extracted from each layer of the proposed network, via top-down distillation. Specifically, the training procedure can be efficiently performed by learning the contextual information for shadow removal without shadow masks. Experimental results on various datasets show the effectiveness of the proposed method for shadow removal under real world environments.

• Genes and Genomics
• /
• v.40 no.12
• /
• pp.1363-1371
• /
• 2018

One of the main concerns in biology is extracting sophisticated features from DNA sequence for gene interaction determination, receiving a great deal of researchers' attention. The epigenetic modifications along with their patterns have been intensely recognized as dominant features affecting on gene expression. However, studying sequenced-based features highly correlated to this key element has remained limited. The main objective in this research was to propose a new feature highly correlated to epigenetic modifications capable of classification of genes. In this paper, classification of 34 genes in PPAR signaling pathway associated with muscle fat tissue in human was performed. Using different statistical outlier detection methods, we proposed that 5-mers highly correlated to epigenetic modifications can correctly categorize the genes involved in the same biological pathway or process. Thirty-four genes in PPAR signaling pathway were classified via applying a proposed feature, 5-mers strongly associated to 17 different epigenetic modifications. For this, diverse statistical outlier detection methods were applied to specify the group of thoroughly correlated genes. The results indicated that these 5-mers can appropriately identify correlated genes. In addition, our results corresponded to GeneMania interaction information, leading to support the suggested method. The appealing findings imply that not only epigenetic modifications but also their highly correlated 5-mers can be applied for reconstructing gene regulatory networks as supplementary data as well as other applications like physical interaction, genes prioritization, indicating some sort of data fusion in this analysis.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.1
• /
• pp.63-68
• /
• 2021

Recently there is sharply increasing interest in watching and creating video contents such as YouTube. However, creating such video contents without privacy protection technique can expose other people in the background in public, which is consequently violating their privacy rights. This paper seeks to remedy these problems and proposes a technique that identifies faces and protecting portrait rights by blurring the face. The key contribution of this paper lies on our deep-learning technique with low detection error and high computation that allow to protect portrait rights in real-time videos. To reduce errors, an efficient tracking algorithm was used in this system with face detection and face recognition algorithm. This paper compares the performance of the proposed system with and without the tracking algorithm. We believe this system can be used wherever the video is used.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.20 no.6
• /
• pp.331-345
• /
• 2021

Virtual testing is considered a major requirement for the safety verification of autonomous driving functions. For virtual testing, both the autonomous vehicle and the driving environment should be modeled appropriately. In particular, a realistic modeling of the perception sensor system such as the one having a camera and radar is important. However, research on modeling to consistently generate realistic perception results is lacking. Therefore, this paper presents a sensor modeling method to provide realistic object detection results in a MILS (Model in the Loop Simulation) environment. First, the key parameters for modeling are defined, and the object detection characteristics of actual cameras and radar sensors are analyzed. Then, the detection characteristics of a sensor modeled in a simulation environment, based on the analysis results, are validated through a correlation coefficient analysis that considers an actual sensor.

• Journal of the Korean Society of Visualization
• /
• v.20 no.3
• /
• pp.136-145
• /
• 2022

Later flow immunoassay (LFIA) is a protein analytical method based on immunoreaction. On the LFIA based protein analytical method, bioreceptor molecule plays a key role, and so a system that evaluates and manages the binding affinity of bioreceptor is needed to secure detection reliability. In this study, Lateral Flow Immunoassay based rapid Bioreceptor Screening Method (rBSM) is presented that provide a simple and quick evaluating method for the binding affinity to the target protein of the antibody as model bioreceptor. To verify this evaluation method, Virus-like particles (VLP) and anti-VLP antibodies are selected as a model norovirus, which is target protein, and the candidate bioreceptors respectively. Among the 5 different candidate antibodies, appropriate antibody could be sorted out within 30 minutes through rBSM. In addition, selected antibodies were applied to two representative LFIA based techniques, sandwich assay and competitive assay. Among these methods, sandwich assay showed more effective VLP detection method. Through applying selected antibodies and techniques to the commercialized mass production lines, an VLP detecting LFIA kit was developed with a detection limit of 10¹² copies/g of VLPs in real samples. Since this proposed method in this study could be easily transformable into other combinations with bioreceptors, it is expected that this technique would be applied to LFIA kit development system and bioreceptor quality management.

• Journal of Veterinary Science
• /
• v.23 no.3
• /
• pp.50.1-50.12
• /
• 2022

Background: There is an urgent need to find reliable and rapid bovine tuberculosis (bTB) diagnostics in response to the rising prevalence of bTB worldwide. Toll-like receptor 2 (TLR2) recognizes components of bTB and initiates antigen-presenting cells to mediate humoral immunity. Evaluating the affinity of antigens with TLR2 can form the basis of a new method for the diagnosis of bTB based on humoral immunity. Objectives: To develop a reliable and rapid strategy to improve diagnostic tools for bTB. Methods: In this study, we expressed and purified the sixteen bTB-specific recombinant proteins in Escherichia coli. The two antigenic proteins, MPT70 and MPT83, which were most valuable for serological diagnosis of bTB were screened. Molecular docking technology was used to analyze the affinity of MPT70, MPT83, dominant epitope peptide of MPT70 (M1), and dominant epitope peptide MPT83 (M2) with TLR2, combined with the detection results of enzyme-linked immunosorbent assay to evaluate the molecular docking effect. Results: The results showed that interaction surface Cα-atom root mean square deviation of proteins (M1, M2, MPT70, MPT83)-TLR2 protein are less than 2.5 A, showing a high affinity. It is verified by clinical serum samples that MPT70, MPT83, MPT70-MPT83 showed good diagnostic potential for the detection of anti-bTB IgG and M1, M2 can replace the whole protein as the detection antigen. Conclusions: Molecular docking to evaluate the affinity of bTB protein and TLR2 combined with ELISA provides new insights for the diagnosis of bTB.

• Animal Bioscience
• /
• v.35 no.4
• /
• pp.511-516
• /
• 2022

Objective: A genomic region associated with a particular phenotype is called quantitative trait loci (QTL). To detect the optimal F₂ population size associated with QTLs in native chicken, we performed a simulation study on F₂ population derived from crosses between two different breeds. Methods: A total of 15 males and 150 females were randomly selected from the last generation of each F₁ population which was composed of different breed to create two different F₂ populations. The progenies produced from these selected individuals were simulated for six more generations. Their marker genotypes were simulated with a density of 50K at three different heritability levels for the traits such as 0.1, 0.3, and 0.5. Our study compared 100, 500, 1,000 reference population (RP) groups to each other with three different heritability levels. And a total of 35 QTLs were used, and their locations were randomly created. Results: With a RP size of 100, no QTL was detected to satisfy Bonferroni value at three different heritability levels. In a RP size of 500, two QTLs were detected when the heritability was 0.5. With a RP size of 1,000, 0.1 heritability was detected only one QTL, and 0.5 heritability detected five QTLs. To sum up, RP size and heritability play a key role in detecting QTLs in a QTL study. The larger RP size and greater heritability value, the higher the probability of detection of QTLs. Conclusion: Our study suggests that the use of a large RP and heritability can improve QTL detection in an F₂ chicken population.

• Smart Structures and Systems
• /
• v.29 no.1
• /
• pp.195-206
• /
• 2022

Crack detection is essential for inspection of existing structures and crack segmentation based on deep learning is a significant solution. However, datasets are usually one of the key issues. When building a new dataset for deep learning, laborious and time-consuming annotation of a large number of crack images is an obstacle. The aim of this study is to develop an approach that can automatically select a small portion of the most informative crack images from a large pool in order to annotate them, not to label all crack images. An active learning method with difficulty learning mechanism for crack segmentation tasks is proposed. Experiments are carried out on a crack image dataset of a steel box girder, which contains 500 images of 320×320 size for training, 100 for validation, and 190 for testing. In active learning experiments, the 500 images for training are acted as unlabeled image. The acquisition function in our method is compared with traditional acquisition functions, i.e., Query-By-Committee (QBC), Entropy, and Core-set. Further, comparisons are made on four common segmentation networks: U-Net, DeepLabV3, Feature Pyramid Network (FPN), and PSPNet. The results show that when training occurs with 200 (40%) of the most informative crack images that are selected by our method, the four segmentation networks can achieve 92%-95% of the obtained performance when training takes place with 500 (100%) crack images. The acquisition function in our method shows more accurate measurements of informativeness for unlabeled crack images compared to the four traditional acquisition functions at most active learning stages. Our method can select the most informative images for annotation from many unlabeled crack images automatically and accurately. Additionally, the dataset built after selecting 40% of all crack images can support crack segmentation networks that perform more than 92% when all the images are used.

• Smart Media Journal
• /
• v.11 no.11
• /
• pp.92-98
• /
• 2022

Identifying defects in textiles is a key procedure for quality control. This study attempted to create a model that detects defects by analyzing the images of the fabrics. The models used in the study were deep learning-based VGGNet and ResNet, and the defect detection performance of the two models was compared and evaluated. The accuracy of the VGGNet and the ResNet model was 0.859 and 0.893, respectively, which showed the higher accuracy of the ResNet. In addition, the region of attention of the model was derived by using the Grad-CAM algorithm, an eXplainable Artificial Intelligence (XAI) technique, to find out the location of the region that the deep learning model recognized as a defect in the fabric image. As a result, it was confirmed that the region recognized by the deep learning model as a defect in the fabric was actually defective even with the naked eyes. The results of this study are expected to reduce the time and cost incurred in the fabric production process by utilizing deep learning-based artificial intelligence in the defect detection of the textile industry.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.7
• /
• pp.1841-1857
• /
• 2023

CenterNet, a novel object detection algorithm without anchor based on key points, regards the object as a single center point for prediction and directly regresses the object's height and width. However, because the objects have different sizes, directly regressing their height and width will make the model difficult to converge and lose the intrinsic relationship between object's width and height, thereby reducing the stability of the model and the consistency of prediction accuracy. For this problem, we proposed an algorithm based on the regression of the diagonal half-length and the center angle, which significantly compresses the solution space of the regression components and enhances the intrinsic relationship between the decoded components. First, encode the object's width and height into the diagonal half-length and the center angle, where the center angle is the angle between the diagonal and the vertical centreline. Secondly, the predicted diagonal half-length and center angle are decoded into two length components. Finally, the position of the object bounding box can be accurately obtained by combining the corresponding center point coordinates. Experiments show that, when using CenterNet as the improved baseline and resnet50 as the Backbone, the improved model achieved 81.6% and 79.7% mAP on the VOC 2007 and 2012 test sets, respectively. When using Hourglass-104 as the Backbone, the improved model achieved 43.3% mAP on the COCO 2017 test sets. Compared with CenterNet, the improved model has a faster convergence rate and significantly improved the stability and prediction accuracy.