Spectral–Spatial Exploration for Hyperspectral Image Classification via the Fusion of Fully Convolutional Networks

Due to its remarkable feature representation capability and high performance, convolutional neural networks (CNN) have emerged as a popular choice for hyperspectral image (HSI) analysis. However, the performances of traditional CNN-based patch-wise classification methods are limited by insufficient training samples, and the evaluation strategies tend to provide overoptimistic results due to training-test information leakage. To address these concerns, we propose a novel spectral-spatial 3-D fully convolutional network (SS3FCN) to jointly explore the spectral-spatial information and the semantic information. SS3FCN takes small patches of original HSI as inputs and produces the corresponding sized outputs, which enhances the utilization rate of the scarce labeled images and boosts the classification accuracy. In addition, to avoid the potential information leakage and make a fair comparison, we introduce a new principle to generate classification benchmarks. Experimental results on four popular benchmark datasets, including Salinas Valley, Pavia University, Indian Pines, and Houston University, demonstrate that the SS3FCN outperforms state-of-the-art methods and can be served as a baseline for future research on HSI classification.