Noisy band selection based on the integration of the Stacked-Autoencoder and Convolutional Neural Network approaches for hyperspectral data.

• DOI: 10.5016/geociencias.v42i2.16976
• Publication Date: 2023-09-27T13:01:18Z
• Authors: Mario Ernesto JIJ...; Caisse AMISSE; Jaime Carlos MACU...; Jorge Antonio Sil...
• Keywords: Autoencoder; Pooling

The presence of noise on hyperspectral images causes degradation and hinders efficiency processing for land cover classification. In this sense, removing or detecting noisy bands automatically becomes a challenge research in remote sensing. To cope problem, an integrated model (SAE-1DCNN) is presented study, based Stacked-Autoencoders (SAE) Convolutional Neural Networks (CNN) algorithms the selection exclusion bands. proposed employs convolutional layers to improve performance autoencoders focused discriminating training data by analyzing signature pixel. Thus, SAE-1DCNN model, information can be compressed, then redundant detected extracted taking advantage deep architecture pooling layers. Hyperspectral from AVIRIS (Airborne Visible/Infrared Imaging Spectrometer) sensor were used evaluate automatic method feature selection. results showed effectiveness identify automatically, suggesting that methodology was found promising alternative within scope pre-processing.
 Keywords: bands; selection; neural network; stacked-autoencoders;