Endmember extraction is a key step in the decomposition of mixed pixels in hyperspectral remote sensing images. The traditional linear endmember extraction method ignores the nonlinear mixing factors of the geological features in the pixel, which restricts the improvement of the decomposition accuracy of the mixed pixel. Considering the nonlinear structure of hyperspectral image data, an orthogonal projection endmember extraction algorithm based on geodesic distance (OPEEGD) is proposed. The geodesic distance was introduced into the endmember extraction process, and the endmembers were extracted one by one using the orthogonal projection method. In order to reduce the amount of geodesic distance calculation, the original hyperspectral data was reduced by the automatic target generation process method and the unconstrained least squares method before endmember extraction. Experiments on simulated and real hyperspectral remote sensing images show that this method can characterize nonlinear factors in spectral data, and the endmember extraction results are better than traditional automatic target generation process endmember extraction method.