The existing supervised fusion classification models for target feature data reorganization cannot achieve effectively on incomplete information. Aiming at the characteristics of non-perfect sample library involved in the discrimination of targets, the generalized belief classification model based on sample class centers is proposed to identify similar targets or that have not appeared in training sets. According to the range between the target to be identified and the category acceptance domain, as well as the unknown domain, the model for generating basic belief assignments of the target category is established. Then, by analyzing the correlations of target feature and the unknown domain of the category, a fusion method is introduced considering belief redundant control strategy. The simulation results show that the proposed generalized belief classification method can divide the targets into single element class, compound class and unknown class. In additionally, the misclassification rate under the incomplete training sample is reduced with effect.