Pattern recognition in noisy environment using principal component analysis and spectral angle mapping

Abstract

This paper proposes an algorithm for detecting object of interest in hyperspectral imagery using the principal component analysis (PCA) as preprocessing and spectral angle mapping. PCA has found many applications in multivariate statistics which is very useful method to extract features from higher dimensional dataset. Spectral angle mapper is a widely used method for similarity measurement of spectral signatures. The developed algorithm includes two main processing steps: preprocessing of hyperspectral dataset and detection of object of interest. To improve the detection rate, the preprocessing step is implemented which processes hyperspectral data with a median filter (MF). Then, principal component transform is applied to the output of the MF filter which completes the preprocessing step. Spectral angle mapping is then applied to the output of preprocessing step to detect object with the signature of interest. We have tested the developed detection algorithm with two different hyperspectral datasets. The simulation results indicate that the proposed algorithm efficiently detects object of interest in all datasets.