Generalized evidence theory(GET) is a useful method to address the problem of multi-sensor information fusion over the incomplete framework of discernment(FoD). Due to the limitations of cognition in the era, people inevitably considered the incomplete FoD as the complete FoD, and the classical evidence theory is not fully applicable in this case. Therefore, a new generalized basic probability assignment(GBPA) determination method based on the GET is proposed. According to the training data, the method first generates the generalized triangular fuzzy number(GTFN) models of the classes and the test samples, respectively. Then the GBPAs are determined by calculating the generalized triangular fuzzy distance between samples and classes. Finally, the generalized combination rule fuses all the bodies of evidence to obtain the final conclusion. The experimental results on the Iris dataset show that the proposed method is reasonably effective and has relatively high classification accuracy even in the case of insufficient samples.