【百家大讲堂】第288期： 光谱解混与端元可变性研究 

　　讲座题目：光谱解混与端元可变性研究 

　　报  告 人：Jocelyn Chanussot 

　　时  　 间：2019年11月29日 下午15:00-17:00

　　地　   点：中关村校区10号教学楼205

　　主办单位：研究生院、 信息与电子学院

　　报名方式：登录北京理工大学微信企业号---第二课堂---课程报名中选择“【百家大讲堂】第288期：光谱解混与端元可变性研究  ”

 

【主讲人简介】

 Jocelyn Chanussot，法国格勒诺布尔理工学院教授。 长期从事于图像分析，数据融合，机器学习以及人工智能在遥感领域应用等研究。现任IEEE地球科学与遥感学会副主席，负责协会会议组织相关工作。担任IEEE T-GRS杂志与IEEE T-IP杂志副主编，从2011年到2015年，曾任 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING 杂志主编。 发表国际期刊论文160余篇， 多次获得相关国际学术奖励。 2012年当选美国IEEE会士， 2018、2019年两次入选汤森路透社高被引科学家。 

 Jocelyn Chanussot is currently a Professor of signal and image processing at the Grenoble Institute of Technology, France. His research interests include image analysis, data fusion, machine learning and artificial intelligence in remote sensing. Dr. Chanussot is the Vice President of the IEEE Geoscience and Remote Sensing Society, in charge of meetings and symposia. He is an Associate Editor of the IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING and the IEEE TRANSACTIONS ON IMAGE PROCESSING. He was the Editor-in-Chief of the IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING from 2011 to 2015. He is the co-author of over 165 papers in international journals and has received several scientific awards and recognitions.  He is a Fellow of the IEEE (2012) and a Highly Cited Researcher (Clarivate Analytics/Thomson Reuters, 2018, 2019).

 

【讲座信息】

光谱解混用于复原图像中物质的纯净光谱，是高光谱成像中一项重要的逆问题。线性解混模型通常应用于现有光谱解混研究，并假设物质与光谱存在一一对应关系。然而，在实际应用中，此类假设会产生严重的光谱类间变异性问题。因此，需要在光谱解混中允许光谱端元存在变化以达到更加鲁棒的解混效果。本次讲座回顾现有针对端元变异问题的研究，并对其分类，且在数据集进行测试分析，以验证端元变异问题对光谱解混的影响。此项工作由Lucas Drumetz在其博士期间研究完成。

Spectral Unmixing is an inverse problem in hyperspectral imaging which aims at recovering the spectra of the pure constituents of an image (called endmembers), as well as at estimating the proportions of said materials in each pixel (called abundances). A linear mixing model is typically used for this purpose, but this approach implicitly assumes that one spectrum can completely characterize each material, while in practice they are always subject to intra-class variability. Taking this phenomenon into account within an image amounts to allowing the endmembers to vary on a per-pixel basis. In this talk, we review and categorize the recent methods addressing this endmember variability and compare their results on a real dataset, thus showing the benefits of incorporating it in the unmixing chain. The work was conducted by Lucas Drumetz during his PhD.