Hyperspectral image-based target detection makes good use of the advantage of continuous spectral information of hyperspectral image

distinguishing the target and background mainly by spectral difference. However

fully considering the spatial relationship between image pixels in some spectra analysis algorithms can overcome the shortcomings of these algorithms

such as the lack of constrained energy minimization (CEM) algorithm estimating the background information with the information of full map and orthogonal subspace projection (OSP) algorithm being difficult to accurately construct background subspace. In this paper

spatial dimension information of targets are introduced and possible targets are filtered out by means of morphological opening operation to construct an accurate background

based on which OSP and CEM algorithms are implemented respectively for target detection. The 3D ROC curve is used to evaluate the accuracy of the detection results to overcome the shortcomings of traditional 2D ROC curve. The experimental results show that morphology-based CEM and OSP algorithms can effectively reduce the false alarm rate and improve detection efficiency.