This study introduces CLSMM, a contrastive pre-training framework that learns semantically invariant joint spatial–spectral representations by contrasting different spatial–spectral masked views of the same sample and incorporating cross-sample mixing. It reduces deep learning models’ reliance on large amounts of labeled data, offering a viable approach to land-cover mapping with limited annotations. (By Prof. Sen Jia)
Key words: Hyperspectral image classification; contrastive learning; spatial-spectral masking.
Link: doi.org/10.34133/remot…
Key words: Hyperspectral image classification; contrastive learning; spatial-spectral masking.
Link: doi.org/10.34133/remot…