Deep Diffusion Processes for Active Learning of Hyperspectral Images
This method addresses the problem of efficiently labeling hyperspectral images for researchers and practitioners working with HSI data.
This paper proposes an active learning method for hyperspectral images (HSI) that combines deep learning with graph diffusion processes. It uses a deep variational autoencoder to extract denoised features from HSI, which are then used for labeling queries based on graph diffusion, leading to strong performance on real HSI.
A method for active learning of hyperspectral images (HSI) is proposed, which combines deep learning with diffusion processes on graphs. A deep variational autoencoder extracts smoothed, denoised features from a high-dimensional HSI, which are then used to make labeling queries based on graph diffusion processes. The proposed method combines the robust representations of deep learning with the mathematical tractability of diffusion geometry, and leads to strong performance on real HSI.