Recursive Band Processing Algorithms for Finding Unknown Targets in Hyperspectral Imagery

Author/Creator

Author/Creator ORCID

Date

2016-01-01

Department

Computer Science and Electrical Engineering

Program

Engineering, Electrical

Citation of Original Publication

Rights

This item may be protected under Title 17 of the U.S. Copyright Law. It is made available by UMBC for non-commercial research and education. For permission to publish or reproduce, please see http://aok.lib.umbc.edu/specoll/repro.php or contact Special Collections at speccoll(at)umbc.edu
Distribution Rights granted to UMBC by the author.

Abstract

Hyperspectral imaging has emerged as a remote sensing image processing technique in many applications. The reason that hyperspectral data is called hyperspectral is mainly because hundreds of contiguous spectral bands provide the massive spectral information used for data analysis. Such high spectral resolution leads to a challenging issue in how to effectively utilize vast amount of spectral information. In order to resolve this dilemma one general approach is to use Band Selection (BS) to judiciously select a desired band subset that can well represent the data. Unfortunately, two prime issues, (1) the number of bands need to be selected p, and (2) how to select appropriate bands, need to be addressed. Recently, an alternative approach, Progressive Band Selection (PBS) was proposed which can process data band by band without knowing the value of p. This dissertations looks into PBS and develops a rather different approach according to Band SeQuential (BSQ) data acquisition format, called Recursive Band Processing (RBP) which focuses on PBS specified by applications. With the custom-designed algorithmic architectures RBP can be carried out by PBS recursively in a similar manner that a Kalman filter does. Since the utility of RBP must be realized by applications, three applications of interest are investigated: (1) unsupervised active hyperspectral target detection where the well-known Automatic Target Generation Process (ATGP) is extended to RBP-ATGP; (2) unsupervised target identification where one of most popular algorithms Pixel Purity Index (PPI) is re-derived as RBP-PPI; (3) endmember finding where a recently developed Orthogonal Projection (OP) based Simplex Growing Algorithm (OPSGA) is further extended to RBP-OPSGA. Several advantages result from the proposed RBP-based algorithms. First it is specifically designed based on BSQ data acquisition format and it is particularly suitable for hyperspectral data communication and transmission with limited bandwidth. Second, RBP provides spectral profiles of changes in target detection which allow users to screen preliminary results while data collection is ongoing without waiting for completion of full data set. Most importantly, the recursive nature in RBP can facilitate the hardware design which can significantly reduce computational complexity in chip design.