Multi-Scale U-Shape MLP for Hyperspectral Image Classification

Hyperspectral images have significant applications in various domains, since they register numerous semantic and spatial information the spectral band with variability of signatures. Two critical challenges identifying pixels hyperspectral image are respectively representing correlated among local global, as well abundant parameters model. To tackle this challenge, we propose a Multi-Scale U-shape Multi-Layer Perceptron (MUMLP) model consisting designed MSC (Multi-Scale Channel) block UMLP (U-shape Perceptron) structure. transforms channel dimension mixes feature to embed deep-level representation adequately. is by encoder-decoder structure multi-layer perceptron layers, which capable compressing large-scale parameters. Extensive experiments conducted demonstrate our can outperform state-of-the-art methods across-the-board on three wide-adopted public datasets, namely Pavia University, Houston 2013 2018