A nonnegative constrained principal component analysis method was proposed to construct a low-dimensional space by which the conversion between it and the multi-spectral space could be achieved.This method overcame the shortcoming that the reconstructed spectral reflectance may be negative when using the classic principal component analysis (PCA) to reduce the dimension of the multi-spectral image data. The reason behind the negative data produced by the PCA was analyzed firstly. According to this, a nonnegative constraint was imposed on the classic principal component analysis model and an iteration equation was constructed. Then through solving that equation, a set of nonnegative linear independence weight vectors of principal components was obtained, by which a low-dimensional spectral space was spanned. Finally a nonlinear optimization technique was used to determine the projection vectors of the multi-spectral image data in the constructed space. Experiments showed that comparing with the classic PCA, the new method can make the reconstructed spectral reflectance data in the interval of [0, 1], which maintains the physical significance of the spectral reflectance. The precision of the space founded by the new method is almost equivalent to that by classic PCA.
