Reference clusters based feature extraction approach for mixed spectral signatures with dimensionality disparity

Abstract

This paper addressed two main challenges in trace detection: a) dimensionality disparity, i.e., the unfavorable ratio between the high dimensionality of the hyperspectral cube and the small size of the training data set, and b) mixed spectral signatures, i.e., the presence of several material substances embedded in one single pixel due to an insufficient resolution. Under these circumstances, traditional pure pixel-based image processing techniques may not be applicable and it remains an open area for more transformative research among academia, government, industry and other non-government organizations. To address these issues, we proposed a new feature extraction approach (FEA) prior to supervised classification of hyperspectral cubes based on local and global spatio-spectral analysis. FEA adapted its parameters to different levels of mixtures using both local gradients and reference clusters. The adaptive feature selection (AFE) approach selected the minimum number of spectral bands without losing discriminating power. We tested the effect of different number of selected spectral bands. The AFE with two spectral bands gave classification accuracy, in which the area under the curve (AUC) was given.

A Robust Context‐Based Deep Learning Approach for Highly Imbalanced Hyperspectral Classification

“…The second step of our approach is to calculate the reference clusters [ 64 ]. Figure 5 shows a pivotal pixel p (5, 5, 5) in its 9-by-9 reference neighborhood .…”

“…Gradients. e first step of our approach is to calculate the local gradients [64]. Figure 4 shows a pivotal pixel p(1, 1, 1) in its 3-by-3-by-3 local neighborhood.…”