Eye Fatigue State Recognition of Gabor Wavelet Optimization HMM Algorithm

Yu Xiang 보안공학연구지원센터 2016 International Journal of Hybrid Information Techno Vol.9 No.9

It is easy for the internet learners to generate learning fatigue because of the long-term lack of emotional interaction in the learning process, and learning fatigue often manifests through the eye condition, in order to do effective monitoring for remote intelligent tutoring system, the learning fatigue eye state recognition algorithm is put forward based on Gabor wavelet and HMM. The algorithm has certain distinguishing characteristics aiming at the degree of eye openness of network learner under 3 learning states: normal learning, fatigue and confusion, first, it does gray difference disposal for eye image by Laplace operator in YCbCr color space, then, it selects two-dimension Gabor kernel function to build 48 optimal filters, obtain 48 characteristic values, these 48 characteristic values generate 48 eigenvectors, at last, it use a set of observation sequence O formed by eigenvector of HMM for eye state image to do eye state recognition. Experimental results show that the recognition rate of this algorithm for network learning reaches 95.68%, and this algorithm has a good robustness.

An Efficient Approach of Marking Packet at Source Side for IP Traceback

Yu Xiang,Choong Seon Hong 한국통신학회 2004 한국통신학회 학술대회논문집 Vol.2004 No.7

Hounsfield Units as an Independent Predictor of Failed Percutaneous Drainage of Spinal Tuberculosis Paraspinal Abscess Under Computed Tomography Guidance

Yu Xiang,Jinyue He,Ruonan Bai,Huorong Gou,Fei Luo,Xuequan Huang,Zehua Zhang 대한척추신경외과학회 2023 Neurospine Vol.20 No.4

Objective: To investigate the value of Hounsfield units (HUs) as an independent predictor of failed percutaneous drainage of spinal tuberculosis paraspinal abscess under computed tomography (CT) guidance. Methods: A retrospective analysis was conducted on 61 patients who underwent CT-guided percutaneous drainage for spinal tuberculosis paraspinal abscess between October 2017 and October 2020. Preoperative CT scans were used to measure the HUs of the abscess. Patients were categorized into successful drainage (n = 49) and failed drainage (n = 12) groups. Statistical analysis involved independent sample t-tests and chi-square tests to compare between the 2 groups. Binary logistic regression was performed to identify independent predictive factors for drainage failure. Receiver operating characteristic (ROC) curves were employed to ascertain risk factor thresholds and diagnostic performance. Results: Among the patients, 49 experienced successful drainage while 12 faced drainage failure. The mean HUs of abscesses in the failed drainage group were significantly higher than those in the successful drainage group (p < 0.001). ROC analysis revealed an area under the curve of 0.897 (95% confidence interval, 0.808–0.986) for predicting drainage failure based on HUs. The optimal HU cutoff value for predicting drainage failure was 22.3, with a sensitivity of 91.7% and specificity of 69.4%. Conclusion: HUs are an independent predictor of failed percutaneous drainage of spinal tuberculosis paraspinal abscess under CT guidance. The HU value of 22.3 can be used as an initial screening threshold for predicting the success or failure of drainage.

The Image Sparse Denoising of Redundant Dictionary Based on Filtering Guidance

Yu Xiang 보안공학연구지원센터 2016 International Journal of Signal Processing, Image Vol.9 No.9

This paper conducts a research on information loss of local feature existing in the image denoising process and puts forward the method of image sparse denoising of redundant dictionary based on filtering guidance. This method utilizes bias noise (additional noise and image errors after denoised image and the corresponding additional noise deviation) for image sparse expression, and extracts the feature information of bias noise to improve the effectiveness of de-noising. In the first place, based on filtering guidance, the method carries out aftertreatment to bias noise still existing after denoise the image. And then, the method, in the basis of this bias noise, designs a new dictionary training method, and obtains redundant dictionary for image processing through self-adaption. Finally, the method extracts featured texture from bias noise image based on the dictionary mentioned above. And it takes advantage of filtering guidance in combination with featured texture extracting information and denoising image to realize image restoration. According to emulated data, the performance of proposed algorithm should be better than the selected comparing algorithm and be equipped with a better visual recovery effect.

Local Binary Haar Feathers Kadane and Multi-Threshold AdaBoost for Facial Classification and Recognition

Yu Xiang 보안공학연구지원센터 2016 International Journal of Signal Processing, Image Vol.9 No.10

A face recognition algorithm based on local binary Haar feathers which represented as Kadane optimizing multi-threshold AdaBoost was proposed according to the problems of texture shape feature representation and classification algorithm accuracy in the process of facial classifying detection and recognition, First, improve the traditional expression by using image Local binary pattern of Haar features , improve image model of texture and shape feature expression ability ; Secondly, for single threshold weak learning algorithm we can not make full use of local binary Haar feature information, resulting in a lower classification accuracy problem proposed Kadane optimizing multi-threshold AdaBoost classifier, to achieve local binary Haar feature representation of facial high accuracy recognition; Finally, through the experiments show, efficient face recognition rate can reach more than 90% by the algorithm,which is superior to the selected comparison algorithm.

Tobacco Retail License Recognition Based on Dual Attention Mechanism

Yuxiang Shan,Qin Ren,Cheng Wang,Xiuhui Wang 한국정보처리학회 2022 Journal of information processing systems Vol.18 No.4

Images of tobacco retail licenses have complex unstructured characteristics, which is an urgent technicalproblem in the robot process automation of tobacco marketing. In this paper, a novel recognition approachusing a double attention mechanism is presented to realize the automatic recognition and information extractionfrom such images. First, we utilized a DenseNet network to extract the license information from the inputtobacco retail license data. Second, bi-directional long short-term memory was used for coding and decodingusing a continuous decoder integrating dual attention to realize the recognition and information extraction oftobacco retail license images without segmentation. Finally, several performance experiments were conductedusing a largescale dataset of tobacco retail licenses. The experimental results show that the proposed approachachieves a correction accuracy of 98.36% on the ZY-LQ dataset, outperforming most existing methods.

Private Equity Valuation under Model Uncertainty

Yuxiang BIAN 한국유통과학회 2022 The Journal of Asian Finance, Economics and Busine Vol.9 No.1

The study incorporates model uncertainty into the private equity (PE) valuation model (SWY model) (Sorensen et al., 2014) to evaluate how model uncertainty distorts the leverage and valuations of PE funds. This study applies a continuous-time model to PE project valuation, modeling the LPs’ goal as multiplier preferences provided by Anderson et al. (2003), and assuming that LPs’ aversion to model uncertainty causes endogenous belief distortions with entropy as a measure of model discrepancies. Concerns regarding model uncertainty, according to the theoretical model, have an unclear effect on LPs’ risk attitude and GPs’ decision, which is based on the value of the PE asset. It also demonstrates that model uncertainty lowers the certainty-equivalent valuation of the LPs. Finally, we compare the outcomes of the Full-spanning risk model with the Non-spanned risk model, and they match the intuitive economic reasoning. The most important implication is that model uncertainty will have negative effects on the LPs’ certainty-equivalent valuation but has ambiguous effects on the portfolio allocation choice of liquid wealth. Our works contribute to two literature streams. The first is the literature that models the PE funds. The second is the literature introduces model uncertainty into standard finance models.

Entity Classification Based on Graph Convolutional Networks for Knowledge Graphs

Yuxiang Sun(순위샹),Hongzhu Duan(단홍주),Yongju Lee(이용주) 한국정보기술학회 2022 Proceedings of KIIT Conference Vol.2022 No.6

지식 그래프의 엔티티 분류는 엔터티를 연속적인 저차원 공간에 투영하고 엔티티의 벡터 값을 기반으로 클러스터링을 수행하는 것을 목표로 한다. 본 논문에서는 엔터티를 2차원 공간에 임베드하기 위한 그래프 컨볼루션 네트워크 모델을 제안하고, 밀도 기반 공간 클러스터링 알고리즘을 사용하여 다양한 임베딩 모델의 클러스터링 성능을 평가한다. 기존의 임베딩 기반 접근 방식과 비교하여 제안된 그래프 컨볼루션 네트워크 모델은 더 나은 클러스터링 성능을 보여주고 이의 우수한 엔티티 표현 능력도 증명된다. The entity classification of knowledge graphs aims to project entities into a continuous low-dimensional space and perform clustering based on the vector values of entities. In this paper, we proposed a graph convolutional network model to embed entities into a two-dimensional space. Afterward, a density-based spatial clustering algorithm is utilized to evaluate clustering performance of different embedding models. Compare to traditional embedding-based approaches, the proposed graph convolutional network model shows a better clustering performant, its excellent entity representation ability is also proved.

Research on Action Strategies and Simulations of DRL and MCTS-based Intelligent Round Game

Yuxiang Sun,Bo Yuan,Yongliang Zhang,Wanwen Zheng,Qingfeng Xia,Bojian Tang,Xianzhong Zhou 제어·로봇·시스템학회 2021 International Journal of Control, Automation, and Vol.19 No.9

The reinforcement learning problem of complex action control in multiplayer online battlefield games has brought considerable interest in the deep learning field. This problem involves more complex states and action spaces than traditional confrontation games, making it difficult to search for any strategy with human-level performance. This paper presents a deep reinforcement learning model to solve this problem from the perspective of game simulations and algorithm implementation. A reverse reinforcement-learning model based on high-level player training data is established to support downstream algorithms. With less training data, the proposed model is converged quicker, and more consistent with the action strategies of high-level players’ decision-making. Then an intelligent deduction algorithm based on DDQN is developed to achieve a better generalization ability under the guidance of a given reward function. At the game simulation level, this paper constructs Monte Carlo Tree SearchIntelligent Decision Model for turn-based antagonistic deduction games to generate next-step actions. Furthermore, a prototype game simulator that combines offline with online functions is implemented to verify the performance of proposed model and algorithm. The experiments show that our proposed approach not only has a better reference value to the antagonistic environment using incomplete information, but also accurate and effective in predicting the return value. Moreover, our work provides a theoretical validation platform and testbed for related research on game AI for deductive games.

Intelligent Auxiliary Diagnostic Architecture based on FHIR

Yuxiang Sun(순위샹),Jehyun Jung(정제현),Yongju Lee(이용주) 한국정보과학회 2020 한국정보과학회 학술발표논문집 Vol.2020 No.7