Acta Geodaetica et Cartographica Sinica ›› 2023, Vol. 52 ›› Issue (10): 1749-1759.doi: 10.11947/j.AGCS.2023.20220637

• Photogrammetry and Remote Sensing • Previous Articles     Next Articles

Robust hyperspectral image clustering integrating total Bregman divergence and bipartite graph

LIU Han1, WU Chengmao2, LI Changxing3   

  1. 1. School of Communication and Information Engineering, Xi'an University of Posts and Telecommunications, Xi'an 710121, China;
    2. School of Electronic Engineering, Xi'an University of Posts and Telecommunications, Xi'an 710121, China;
    3. School of Science, Xi'an University of Posts and Telecommunications, Xi'an 710121, China
  • Received:2022-11-17 Revised:2023-09-26 Published:2023-10-31
  • Supported by:
    The National Natural Science Foundation of China(No. 62071378)

Abstract: In view of the high computational complexity and low clustering accuracy of traditional graph based spectral clustering algorithms, which are difficult to apply to large-scale data clustering, this paper proposes a graph based clustering algorithm to deal with hyperspectral image classification problems by using the similarity measurement between anchor points and data points, which is called robust hyperspectral image clustering integrating total Bregman divergence and bipartite graph (RTBBG). Firstly, the spatial information of hyperspectral image is added in the construction of bipartite graph, which makes full use of the rich spatial information of hyperspectral image. Secondly, the total Bregman divergence is used to optimize the traditional Euclidean distance as a similarity measure between data points and anchors, which makes the constructed bipartite graph more stable and enhances the robustness of the algorithm. Finally, K-means algorithm is used to directly cluster the spectra to obtain the final clustering results. The effectiveness of the algorithm is verified by testing on three large-scale hyperspectral datasets.

Key words: hyperspectral image, bipartite graph, total Bregman divergence, similarity measurement, spatial information

CLC Number: