Double-Branch Network with Pyramidal Convolution and Iterative Attention for Hyperspectral Image Classification

Deep-learning methods, especially convolutional neural networks (CNN), have become the first choice for hyperspectral image (HSI) classification to date. It is a common procedure that small cubes are cropped from images and then fed into CNNs. However, standard CNNs find it difficult extract discriminative spectral–spatial features. How obtain finer features improve performance now hot topic of research. In this regard, attention mechanism, which has achieved excellent in other computer vision, holds exciting prospect. paper, we propose double-branch network consisting novel convolution named pyramidal (PyConv) an iterative mechanism. Each branch concentrates on exploiting spectral or spatial with different PyConvs, supplemented by module refining feature map. Experimental results demonstrate our model can yield competitive compared state-of-the-art models.