Hyperspectral images have the characteristics of multiple bands and strong correlation among bands, which leads to information redundancy of hyperspectral images, resulting in dimensionality disaster and difficulty in classification. Therefore, a dimensionality reduction classification method of hyperspectral images based on LDA and extreme learning machine is proposed. In this method, hyperspectral data are firstly processed by LDA for dimensionality reduction, so as to overcome the problem of hyperspectral image information redundancy and keep the image feature information as far as possible. After reducing the spectral image dimension, ELM is adopted to classify and identify hyperspectral remote sensing images. The method proposed is applied to Pavia University and Salinas hyperspectral image processing, and the classification accuracy reaches 98. 78% and 99. 94% respectively, which effectively improves the feature classification performance of hyperspectral images and has strong practicability.