报告题目：Sparse Subspace Clustering for Hyperspectral Imagery
报告人： 杜谦 教授
Dr. Du is Bobby Shackouls Professor with the Department of Electrical and Computer Engineering, Mississippi State University, USA. Her research interests include hyperspectral image analysis and applications, pattern recognition, and machine learning. Dr. Du is a Fellow of IEEE and SPIE—International Society for Optics and Photonics. She served as Co-Chair for the Data Fusion Technical Committee of IEEE Geoscience and Remote Sensing Society (GRSS) in 2009–2013, and Chair for Remote Sensing and Mapping Technical Committee of International Association for Pattern Recognition (IAPR) in 2010–2014. She served as Associate Editor for IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (2011–2015), IEEE Signal Processing Letters (2012–2015), and Journal of Applied Remote Sensing (2014–2015). Currently, Dr. Du is the Editor-in-Chief of IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (JSTARS). She was the Guest Editor of several special issues published in IEEE Transactions on Geoscience and Remote Sensing, IEEE JSTARS, Journal of Applied Remote Sensing, Pattern Recognition Letters, Remote Sensing, Sensors. Dr. Du is the General Chair of the 4th IEEE GRSS Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS) in 2012, and the General Chair of the 7th and 8th IAPR Workshop on Pattern Recognition in Remote Sensing in 2012 and 2014, respectively.
Hyperspectral imaging has been of great interest in pattern recognition and Earth observations due to the fact that its high spectral resolution offers powerful discriminant capability in separating objects or materials with subtle spectral discrepancy. However, the resulting high spectral dimensionality may bring out difficulty in data processing and analysis. Traditional clustering techniques, such as k-means clustering, may not work well for high-dimensional data because distance measurement becomes less accurate in such a case. Sparse subspace clustering is more suitable for high-dimensional data clustering which is to cluster data points that lie in a union of low-dimensional subspaces. The key idea is to find a sparse representation of a data point (in terms of other points), which corresponds to selecting a few points from the same subspace. The solution is used in a spectral clustering framework to infer the clustering of the data into subspaces. In this talk, the original sparse subspace clustering, low-rank subspace clustering, low-rank and sparse subspace clustering, and its multiview versions are introduced. Moreover, their kernel extension and approximate kernel extension are also discussed. Comparison with state-of-the-arts in terms of unsupervised classification accuracy is presented to demonstrate the superiority of multiview versions. Existing challenges will also be discussed.