Abstract: In this paper,the wavelet packet was used to analyse the impact acoustic signals of un-damaged kernels,insect damaged kernels and moldy damaged kernels.The characteristic features including node energy,singularity value,power spectral entropy and spectral entropy arm of node envelope signals were extracted,and the features were classified in support vector machine.The recognition accuracy rate in classification of three types of wheat kernels were above 92%.The experimental result shows that each type of wheat impact signals features are much different,and this research has a more comprehensive value in application,and provides a new method for wheat kernels sorting.