Features reduction collaborative fuzzy clustering for hyperspectral remote sensing images analysis

Article type: Research Article

Authors: Dang, Trong Hop^a | Do, Viet Duc^b | Mai, Dinh Sinh^{b; *} | Ngo, Long Thanh^b | Trinh, Le Hung^b

Affiliations: [a] Hanoi University of Industry, Hanoi, Vietnam | [b] Le Quy Don Technical University, Hanoi, Vietnam

Correspondence: [*] Corresponding author. Dinh Sinh Mai. E-mail: [email protected].

Abstract:  In image processing, segmentation is a fundamental problem but an important step for advanced image processing problems. When dealing with hyperspectral image data, the task becomes much more challenging due to the large number of features (dimension), higher nonlinearity, and greater capacity of the data. This paper proposes a solution of features reduction collaborative fuzzy c-means clustering (FR-CFCM) for hyperspectral remote sensing image analysis using random projection. The dimensional reduction technique is based on the Johnson Lindenstrauss lemma algorithm, preserving the relative distance between data samples. This can make clustering easier without affecting the clustering results. Moreover, by reducing dimensionality and sharing information among sub-data in collaborative clustering, it is possible to improve the performance and accuracy of hyperspectral remote sensing image analysis results. The experiments conducted on two hyperspectral image data sets with five validity indexes show that the proposed methods perform better compared with the other methods.

Keywords: hyperspectral image, fuzzy clustering, collaborative clustering, feature reduction

DOI: 10.3233/JIFS-230511

Journal: Journal of Intelligent & Fuzzy Systems, vol. 45, no. 5, pp. 7739-7752, 2023