学术讲座-Tensor Factorization and Completion Methods for Image Processing and Hyperspectal Data Analysis

19.03.2019  14:40
Tensor Factorization and Completion Methods for Image Processing   and Hyperspectal Data Analysis
数理学院

谢亚君正高龙建辉
自然科学全校师生2019-03-20 15:00
北区至真2号楼203室
谢亚君,博士,福建江夏学院教授。主要从事数值代数、最优化理论等方面研究。专业领域涉及互补问题、压缩感知、广义Sylvester矩阵方程、鞍点问题及PageRank等方面研究,并取得一系列成果。在《Applied   Mathematical Modelling》、 《Applied  Mathematics  and  Computation》、 《Numerical   Algorithm》、《Computers and Mathematics with Applications》、《Journal of   Inequalities and Applications》、《Computational and Applied Mathematics》、《Chinese   Quarterly Journal  of Mathematics》、《Chinese Quarterly Journal  of   Mathematics》、《计算数学》等国内外重要专业学术期刊上发表30多篇论文,其中权威期刊17篇,被SCI收录14篇,其中一篇入选ESI高被引论文其被引用次数已归入其学术领域中最优秀的1%之列。先后主持省级课题2项,厅级课题1项,并参与国家自然科学基金2项、省自然科学基金项目3项及青年基金项目1项,参编教材2部。曾获2012-2013年度福建江夏学院“邮储金雁奖”科研新秀;并于2015年入选福建省高校“杰出青年人才培养计划”;2017年获福建省教育厅高等学校优秀学科(专业)带头人赴英国University   of Liverpool研修1年。
In this talk, efficient nonnegative matrix factorization approach,   an multilevel accelerated regularization nonnegative matrix factorization method   with smoothness items is proposed to solve some image reconstruction problems.   Furthermore, the method is extended skillfully to nonnegative tensor   factorization with association of acceleration alternating least square   nonnegative matrix factorization, and application to hyperspectral data   analysis. The novel approaches show the efficiency and stability when be applied   to some popular and classical methods, such as multiplicative update nonnegative   matrix factorization, gradient decent nonnegative matrix factorization ,   alternating least square nonnegative tensor factorization and their   regularization versions etc. Therefore, the proposed techniques tend to be   treated as the generalization and improvement for these current popular methods.   A large number of numerical experiments demonstrate the presented methods are   thoroughly attractive and valid for some real-world   applications.