基于撞击声分析的榛壳开口判别

Discriminating intact and crack hazelnuts using impact sound analysis

撞击声分析是新型的坚果品级检测方法。为了研究开、闭口榛果的撞击声特性，通过不同的数据处理方法提取声信号的梯度累积特征、线性频谱特征和梅尔倒谱系数（MFCC）特征参数。研究结果显示：撞击声信号的累积梯度特征参数计算方便，阈值判断简单易行，识剐精度可达80％左右；撞击声信号的线性频谱特征明显，闭口榛果撞击声能量集中在5～8kHz频段内，开口榛果撞击声能量趋于全频段内分散。较难寻找固定的频率特征参数；开闭、口榛果的辨识不能直接采用撞击声信号的MFCC特征参数，需要引入特定统计模型或应用人工神经网络，对特征参数进行训练，进而实现开闭、口榛果的模型识别。

Impacting sound analysis is a new method for nuts grading and detection. The acoustic characteristic of the intact and crack ha zelnuts is presented in this paper. The gradient accumulation, frequency spectrum and mel-frequency cepstral coefficients （MFCC） were extracted from the impacting sound signals. The results showed that the gradient accumulation parameter was easily calculated, and the accuracy of discriminating crack hazelnuts was about 80%. For frequency spectrum analysis, the spectral distribution of the intact hazelnuts sound concentrates was in 5-8 kHz and that of the crack hazelnuts tends to disperse in wide frequency range. It is difficult to choose the special detection spectrum. The MFCC parameters cannot be used directly. For detection or grading, it should be combined with the statistical models or artificial neural network.