Infrared-radiative ceramic materials for grain drying were prepared based on the characteristic infrared absorbance of grains. The sintering raw materials are mainly composed of Al₂O₃, SiO^₂, while Fe₂O₃ , CuO and Y₂O₃ are used as doping materials. The infrared radiances of the ceramics（mullite,mullite doped）, which was measured by infrared radiator in the wave band range from 2.5 to 15 μm, are varied from 0.845 to 0.933. An optimized composition, which is obtained by using an orthogonal experimental design method, is as follows:mullite（Al₂O₃∶SiO₂=3∶2）, Fe₂O₃ 2.5%, CuO 5%, and Y₂O₃ 2.5%.

For in small sample case, within-class scatter matrix has zero space which degenerates the vectors' discriminant ability. A modified Kernel-based uncorrelated set of discriminant vectors algorithm(MKUFDA) was proposed. Experiments on ORL face dataset show that our modified algorithm has a better performance than original kernel-based uncorrelated discriminant vectors.