FC-TLBO: fully constrained meta-heuristic algorithm for abundance estimation using linear mixing model

Abstract

A study of abundance estimation has vital importance in spectral unmixing of hyperspectral image. Recently, various methods have been proposed for spectral unmixing to achieve higher performance using an evolutionary approach. However, these methods are based on unconstrained optimisation problems. Their performance was also based on proper tuning parameters. We have proposed a new non-parametric algorithm using teaching-learning-based optimisation technique with an inbuilt constraints maintenance mechanism using the linear mixing model. In this approach, the unmixing problem is transformed into a combinatorial optimisation problem by introducing abundance sum to one constraint and abundance non-negative constraint. A comparative analysis of the proposed algorithm is conducted with other two state-of-the-art algorithms. Experimental results in known and unknown environments with varying signal-to-noise ratio on simulated and real hyperspectral data demonstrate that the proposed method outperforms the other methods.