In the base of kernel signature space orthogonal projection (KSSP), a composite kernel signature space orthogonal projection (CKSSP) technique, which combines spectral information with spatial information, is proposed for target detection in nonlinearly mixed hyperspectral imagery. The grey mathematical morphological transform is extended into multivariate mathematical morphological transform based on marginal ordering and reduced ordering, respectively. The pixel distance is used as ordering scale function to establish reduced ordering. Extended mathematical morphological method with multi-structure elements is used to extract spatial information of hyperspectral images. Combining the spectral and spatial information, the composite kernel function is constructed and improved according to kernel function definition. Target is detected by CKSSP. The proposed method not only sufficiently applies the spectral information, but also effectively takes into account the spatial information. Experimental results of simulated data demonstrate that root mean square error of CKSSP is 0.03 less than that of KSSP, Experimental results of real data and the receiver operating characteristic curves show that CKSSP approach slightly outperforms the KSSP method in target detection.