Hyperspectral Classification Using Heterologous Feature Alignment and Fusion

Despite multisource remote sensing data collaboration compensates for the limitations of hyperspectral (HS) sensors, it faces problems such as significant information differences and heterogeneity. In this paper, an alignment enhancement network (AENet) is designed for information propagation between HSI and auxiliary modalities, such as multispectral and synthetic aperture radar (SAR). Specifically, the auxiliary modalities achieve consistency projection with HS modality through spectral and spatial alignment. Therefore, feature alignment alleviates the problem of heterogeneity to a certain extent and improves fusion efficiency. Experimental results on multisource datasets demonstrate that the proposed AENet is able to provide competitive advantages.