Abstract: In order to realize the automatic classification of wheat kernels, a new approach that combines the bispectrum and support vector machine is introduced to classify and recognise wheat impact sounds of undamaged kernels, insect damaged kernels and moldy kernels. The impact acoustic signals are processed by bispectrum estimation. Features in bispectrum and diagonal slices spectrum are extracted. Then the features are classified in support vector machine. The recognition accuracy rates in classification of undamaged kernel, insect damaged kernel and moldy kernel are above 84%. The experimental results show that this research has a more comprehensive value in application, and it provides a new method for wheat kernels classification.

Key words: wheat impact acoustic signals, bispectrum estimation, support vector machine

摘要： 为实现小麦颗粒的自动分类，采用双谱和支持向量机相结合方法对小麦完好粒、虫蛀粒和霉变粒的碰撞声进行分类识别。对碰撞声信号进行双谱估计，提取信号双谱峰值和对角切片谱两部分特征，用支持向量机分类器进行分类，对完好粒、虫蛀粒和霉变粒3种小麦颗粒识别正确率均达84%以上。实验结果表明，该研究具有较强的实际应用价值，为小麦颗粒的分类提供了新的方法和依据。

关键词: 小麦碰撞声, 双谱估计, 支持向量机