Hyperspectral Image Band Selection and Fuzzy Clustering Integrated Method Based on Multiobjective Evolutionary Optimization

Hyperspectral remote sensing image (HSI) can record complete and continuous spectral curves of land features while obtaining the image information, which can achieve precise classification. However, HSIs have complex and nonlinear structures due to the high dimensionality and uncertainty of signals, resulting in redundant band information and difficulty in sample labeling. Although a series of band selection and unsupervised classification methods have been developed, they currently follow the research framework of “band selection first and then clustering”, resulting in transmission and accumulation errors. In this paper, based on the multiobjective evolutionary algorithm, an integrated method for HSI band selection and fuzzy clustering (IBSC) was proposed. The proposed IBSC method can achieve multitask encoding and modeling for band selection and fuzzy clustering, and the encoded individual solutions can generate new candidate solutions through information exchange between band selection and clustering. The effectiveness of IBSC was proved by experiments with three commonly used HSI clustering datasets, and the overall clustering accuracy is improved by about 1%. Thus, this integrated method can reduce the cumulative error between band selection and fuzzy clustering independent models.