Low probability detection(LPD) is an approach for hyperspectral imagery analysis;it use the eigenvectors of the imagery's correlation matrix in detecting small targets.Unfortunately

bacause of noise and the orthogonality constraints among the eigen vectors

the results of detection are non ideal.In this paper

we use the method of Convex Cone Analysis(CCA) to improve the detectability of LPD and to eliminate the stripe noise.The experimental results are given by applying the method to the data from Operative Modulor Imaging Spectrometer(OMIS) system.