SEMI-SUPERVISED MULTI-SPECTRAL LAND COVER CLASSIFICATION WITH MULTI-ATTENTION AND ADAPTIVE KERNEL

Land cover classification is one vital and challenging task in remote sensing (RS) field. Tackling the limited labeled training data and for exploiting abundant information of multi-spectral bands, we proposed a novel semi-supervised multi-spectral classification network with multi-attention and adaptive kernel. We select 10 land cover related bands and combine them into different spectral groups. To further extract discriminating feature in spectral and spatial dimension, we design one multi-attention block. Furthermore, the adaptive receptive field mechanism is introduced to dynamically adjust kernel size in convolution layer and aggregate multi-scale information. Experiments on EuroSAT dataset demonstrate that the proposed method can outperform greatly state-of-the-art methods, especially when the number of labeled data is relatively small.