Hyperspectral Image Classification Using Transfer Learning

Hyperspectral images offer abundant spectral details for spatially related material to be identified and distinguished thorough analysis. Based on spectral data and spatial correlation, a broad range of advanced classification techniques is possible. Computer technological advances have fostered the growth of modern, efficient deep-learning (DL) techniques that show a wide variety of applications with encouraging performance. Particularly in the field of remote sensing information gathered by Earth Observer (EO) instruments, deep-learning techniques have been successfully used. Given the abundance of information contained in this type of pictures, hyperspectral imaging (HSI) is one of the major topics in remote sensing research which enables greater earth surface analysis and exploitation through the combination of rich spatial and spectral information. Given the high dimensions of the data and the restricted supply of training samples, HSI presents significant difficulties for supervised classification methods. Transfer learning architectures having the great potential in the classification of HSI information have been established recently to resolve some constraints. This paper provides an experiment of HSI classification using transfer learning. Experiment results show that limited number of training sample model gives better accuracy.