In order to select optimal spectral features at the minimum cost, a Filter type algorithm for spectral feature selection is proposed from the perspective of information loss. The algorithm sorts the features according to the joint mutual information size and uses a zero information loss principle to judge and reduce the redundant features of the sorted feature set. In this way, a small and optimal subset of spectral features can be obtained, which can represent the whole original spectral features, and the information loss caused by the deletion of redundant features can be reduced.