Abstract: Based on high-order nonlinear mixing model, a multi-objective hyperspectral unmixing algorithm was proposed, which solved the problem that the traditional method cannot obtain higher unmixing accuracy due to the outliers of hyperspectral image data. The proposed algorithm took the reconstruction error and spectral angle mapper as the objective functions and optimized them in order to reduce the outliers influence of hyperspectral data on the solution of optimization model and improve the two evaluation indicators. Then, the difference search algorithm was used to solve the multi-objective optimization model and overcame the tendency of the traditional gradient-based optimization method to fall into the local extremum problem and further improved the unmixing accuracy. The experiment results show that the proposed algorithm has more accurate endmembers abundance estimation and higher unmixing accuracy.