Abstract: This paper uses HHT method to analyze the impact acoustic signal of un-damaged kernels, IDK and moldy damaged kernels, extracts five characteristic features from three types of wheat kernels, and classifies the wheat kernels with BP neural network, gets a good recognition result in the end. The experimental result shows that each type of wheat kernels is much different from other types in frequency domain. The research provides a feasible method for the wheat kernels identification and separation.

Key words: impact acoustic signal, Hilbert-Huang Transform（HHT）, Back Propagation（BP） neural network

摘要： 采用希尔伯特-黄变换方法分析了小麦完好粒、虫害粒和霉变粒的碰撞声信号，提取了3种类型的声信号在频域的5个特征量，使用BP神经网络进行分类，得到较好的识别结果。实验结果表明，不同类型小麦碰撞声信号在频域存在较大差异，此项研究为实现小麦颗粒的自动识别提供了可行方法。

关键词: 碰撞声, 希尔伯特-黄变换（HHT）方法, 反向传播（BP）神经网络