A Novel Classification Framework for Hyperspectral Image Classification Based on Multiscale Spectral-Spatial Convolutional Network

Multiscale spectral-spatial classification has been widely applied to hyperspectral image (HSI). Convolution neural networks (CNN) with multiscale spectral-spatial features have been introduced for hyperspectral image classification (HSIC) in recent years. However, most of current methods mainly use patches as input, which may cause a lot of redundancy in the testing phase and reduce processing efficiency. In this paper, we design a multiscale spectral-spatial CNN for HSIs (HyMSCN) based on a novel image-based classification framework. This network integrates multiple receptive fields fused features with multiscale spatial features at different levels. Experimental results from two real hyperspectral images demonstrate the efficiency of the proposed method.