Cross Scene Hyperspectral Image Classification Combined with Fusional Spatial-Spectral Attention

One of the main challenges in hyperspectral image classification is the small amount of labeled data and the spectral shift problem between the source and target domains in cross scene hyperspectral image classification, which leads to low accuracy and poor robustness of the training model. This paper proposed a feature extraction method combined with Fusional Spatial Spectral Attention Net (FSSAN), and multi-scale and cluster domain adaptation were used to reduce spectral shift and improve model training accuracy. Experiments and studies on two publicly available cross scene datasets, Pavia and Indiana, have shown that the proposed classification algorithm has superior performance. Compared to other cross scene hyperspectral image classification models, the model proposed in this paper has higher accuracy in land feature classification, which will greatly promote the practical application of cross scene hyperspectral intelligent perception technology.