A simulated annealing feature extraction approach for hyperspectral images

In this paper, a novel study of the simulated annealing feature extraction (SAFE) for high-dimensional remote sensing images is proposed. The approach is based on the greedy modular eigenspace (GME) scheme. GME was developed by clustering highly correlated bands into a smaller subset based on the greedy algorithm. Unfortunately, GME doesn’t guarantee to reach a global optimal solution by the greedy algorithm except by the exhaustive search method. Accordingly, finding an optimal (or near-optimal) solution is very expensive. In order to overcome this disadvantage, the SAFE scheme is introduced to improve the performance of GME feature extraction optimally by modifying the correlation coefficient operations and taking sets of non-correlated bands for hyperspectral images based on a heuristic optimization algorithm. It presents a framework, which consists of two algorithms, referred to as SAFE and the feature scale uniformity transformation (FSUT). SAFE is designed to extract features by a new defined three-dimensional simulated annealing modular eigenspace (SAME) to optimize the modular eigenspace, while FSUT is performed to fuse most correlated features from different spectrums associated with different data sources. The performance of the proposed method is evaluated by applying it to hyperspectral and airborne synthetic aperture radar (SAR) images. The experimental results demonstrated that SAFE is not only an effective scheme of feature extraction but also an alternative to the existing dimensionality reduction methods.

Research highlights
► Greedy modular eigenspace feature extraction (GMEFE) groups highly correlated bands.
► 3D simulated annealing feature extraction (SAFE) outperforms 2D-GMEFE.
► 3D-SAFE is an effective alternative to existing dimensionality reduction methods.