Terrain classification is an important research direction in the field of remote sensing. Hyperspectral remote sensing image data contain a large amount of rich ground object information. However, such data have the characteristics of high spatial dimensions of features, strong data correlation, high data redundancy, and long operation time, which lead to difficulty in image data classification. A data dimensionality reduction algorithm can transform the data into low-dimensional data with strong features and then classify the dimensionally reduced data. However, most classification methods cannot effectively extract dimensionality-reduced data features. In this paper, different dimensionality reduction and machine learning supervised classification algorithms are explored to determine a suitable combination method of dimensionality reduction and classification for hyperspectral images. Soft and hard classification methods are adopted to achieve the classification of pixels according to diversity. The results show that the data after dimensionality reduction retain the data features with high overall feature correlation, and the data dimension is drastically reduced. The dimensionality reduction method of unified manifold approximation and projection and the classification method of support vector machine achieve the best terrain classification with 99.57% classification accuracy. High-precision fitting of neural networks for soft classification of hyperspectral images with a model fitting correlation coefficient (R2) of up to 0.979 solves the problem of mixed pixel decomposition.

Feature selection is an important way to achieve high precision and efficient classification of hyperspectral remote sensing images. However, most existing FS methods use a fixed scale to extract features and the relationship between spatial and spectral dimensions is ignored. In fact, this correlation is useful for classification. In this paper, a multi-scale feature fusion network (MSFGW) is proposed in which a global weighting mechanism is explored to catch spatial-spectral information at multiple scales. First, the multi-scale feature extraction module composed of group convolution and dilated convolution is utilized to extract the multiscale features. With the increase of the dilation rate, the module takes the spatial differences at varying scales. Secondly, a 3D weighting mechanism is used to combine the spatial and spectral correlated information for reducing the interference of homologous and heterologous and boosting the feature discrimination ability. Then, multi-scale weighted features are fused to integrate the internal information of all bands at different scales. Finally, the band reconstruction network is used to select representative bands according to their entropy. The experimental results with the state-of-the-art feature selection algorithms on three widely hyperspectral datasets demonstrate that the features selected by MSFGW have obvious advantages in classification with only a few training samples.