Hyperspectral imaging classification based on LBP feature extraction and multimodel ensemble learning

Abstract

When aiming at hyperspectral classification, spectral characteristics are often considered, while spatial characteristics and information redundancy between spectra are neglected. We propose a method of feature extraction using a uniform local binary pattern and ensemble multiple models to classify features. This model uses a uniform local binary mode to extract the spatial features of hyperspectral images. The spatial features and the spectral features are fused to obtain high-dimensional fusion data, which are input into the sparse representation model for learning, and the residual of the test sample is obtained. At the same time, hyperspectral images have serious information redundancy, so we use the product moment correlation coefficient to reduce the interference between classes of sample information. Finally, through ensemble learning of different classification models, the hyperspectral classification can be accurately predicted. The experimental results on hyperspectral data show that this method can effectively improve the effect of hyperspectral image classification.