Assessment of Different Spectral Unmixing Techniques on Space Borne Hyperspectral Imagery

Spectral unmixing decomposes the mixed pixels into constituent land cover features present in that pixel. This can be understood through the concepts of affine, convex and projective geometries. Spectral unmixing is difficult to implement in coarser spatial resolution space-borne hyperspectral data, due to the natural heterogeneity of the different land cover features. Linear spectral unmixing (LSU) follows linear equations for generating fractional coefficients; however, it contains limitations like its inability to handle noisy pixels, least-square error calculation, etc. Mixture tuned matched filtering (MTMF) is a partial unmixing technique in which user-defined targets are mapped. This approach uses a matched filter (MF) and linear mixture theory in combination. Whereas simplex projection unmixing (SPU) technique is nonlinear and is utilized for resolving problems such as fully constrained least square and projecting a point onto a simplex. In this study, Hyperion data was used for performing spectral unmixing using LSU, MTMF, and SPU techniques. The unmixing results obtained were compared and validated using available images from geo-portals. The abundance images of SPU were observed better than MTMF and LSU in terms of the material identification. The variation in the percentage aerial coverage of the land cover features in the mixed pixel is found closer in the abundance results of SPU, i.e., 0.1–3.4% whereas MTMF and LSU have a variation of 0.6–5.2% and 1.9–8.7%, respectively. Rule-based classification was performed on the “abundance images” and SPU classification outperformed the other two techniques, as it enabled differentiation of most of the land cover features. © 2022, The Author(s), under exclusive licence to Springer Nature Switzerland AG.