面向高光谱遥感图像的MMRI-Boruta特征选择算法

打开文本图片集

中图分类号：TP391 文献标志码：A 文章编号：1671-6841（2026）01-0072-06

DOI：10.13705/j.issn.1671-6841.2024111

Abstract： The objective of hyperspectral remote sensing image feature selection was to choose the optimal subset of spectral features from a high-dimensional set，thereby eliminating redundancy and enhancing the efficiency and accuracy of image analysis. A hybrid feature selection algorithm named MMRI-Boruta was proposed. The filter-based MRI feature selection algorithm was initially enhanced by incorporating a new feature importance evaluation metric for MMRI-Boruta. Subsequently，the wrapper-based Boruta algorithm was employed to further optimize the feature subset.The strengths of both filter and wrapper algorithms were combined for the proposed feature selection algorithm，making it easier to obtain the optimal feature subset.To verify the efectiveness，two clasical hyperspectral remote sensing image datasets，Indian Pines and Salinas，were used for testing. Experimental results demonstrated that the proposed algorithm outperformed the comparison algorithms.

Key words：hyperspectral remote sensing image； feature selection；mutual information； relevance;entropy

0 引言

具有丰富的光谱和空间信息，可以精确区分不同的地表覆盖类型，因此被广泛应用于农业监测、林业保护、军事决策等领域[1]。（剩余9779字）