Image-based neural architecture automatic search method for hyperspectral image classification

Convolutional neural networks (CNNs) have shown excellent performance for hyperspectral image (HSI) classification due to their characteristics of both local connectivity and sharing weights. Nevertheless, with the in-depth study network architecture, merely manual empirical design can no longer meet current scenario needs. In addition, existing CNN-based frameworks are heavily affected by redundant three-dimensional cubes input result in inefficient description issues HSIs. We propose an image-based architecture automatic search framework (I-NAS) as alternative CNN. First, alleviate spectral–spatial distribution, I-NAS feeds a full into via label masking fashion. Second, end-to-end cell-based structure space is considered enrich feature representation. Then, it determined optimal cells employing gradient descent algorithm. Finally, well-trained CNN automatically constructed stacking cells. The experimental results from two real HSI datasets indicate that our proposal provide competitive classification.