Hyperspectral Imagery Further Unmixing Based On Analysis Of Variance

Hyperspectral imagery unmixing model based on sparse regression uses the existing endmembers' library as priori information. Usually, the existing endmembers' library contains almost all kinds of ground objects. Even though sparse regression-based imagery unmixing method added sparse constraint to the original unmxing model, the solution is still far away as sparse as real scenario. Therefore, we propose a hyperspectral imagery further unmixing method based on the analysis of variance. In this method, fractional abundances unmixed by sparse regression-based approach are analyzed with t-test. If the fractional abundances are not significant enough, the corresponding endmembers will be removed and a new optimal endmember subset will be extracted. Then the unmixing process was redid with acquired optimal endmember subset and the final result will be acquired. The experimental results indicate that the proposed method could acquire sparser solution, which is closer to the real sparsity of abundance, both in simulate scenario and real scenario. Furthermore, the precision of the endmember recognition of proposed method is more than 97%, which is a pretty good result.