| Abstract: |
| Band selection is an important technique for eliminating spectral redundancy of hyperspectral imagery while preserving critical information. Recently, correlations among neighboring bands or pixels have been exploited in the form of graph regularizations to reduce the data dimensionality efficiently. However, the manipulation of graph regularizations is typically a computational bottleneck. In this presentation, we propose a fast and robust method for hyperspectral band selection based on spatial/spectral graph Laplacians and matrix CUR decomposition. The efficiency of the proposed method is shown on two real datasets by comparing with several other state-of-the-art band selection methods. |
|