Hyperspectral Remote Sensing Images Terrain Classification Based on LDA and 2D-CNN

Hyperspectral remote sensing images (HRSIs) are characterized by high dimensionality and large data volume, but there is a large amount of redundant information between spectral dimensions, which not only affects the accuracy of terrain classification, but also increases the complexity of classification and recognition. In order to reduce the redundant information between spectral bands, reduce the computational complexity, and improve the efficiency of terrain classification of HRSIs, this paper firstly performs dimensionality reduction on HRSIs using linear discriminant analysis (LDA), projects the data to a lower dimensional feature subspace, extracts the most discriminative information, and then uses two-dimensional convolutional neural network (2D-CNN) for deep feature extraction and classification recognition. By classifying on three real HRSIs dataset, the experimental results show that the classification results of LDA-2D-CNN outperform those of the PCA-2D-CNN method using 2D-CNN combined with principal component analysis (PCA).