 MSRPNet is a new approach to hyperspectral image classification that combines superpixels with two-dimensional singular spectrum analysis and random patch convolution. It takes advantage of both global and local spectral knowledge, as well as noise reduction and spatial feature extraction. The resulting features are then fused together and classified using support vector machines. Experimental results show that this approach outperforms existing methods. This article was authored by Hua Yuchun, Ting Ting Wang, Tao Chen, and others.