基于高光谱图像技术的大豆分级识别方法研究

Identification of different soybean grades based on hyperspectral imagery

对大豆进行快速准确分级，采集1—5等级大豆波长在1000～2500nm范围的高光谱图像数据，获得光谱图像；对不同大豆等级样本的光谱曲线进行分析；通过主成分分析法，从每个等级大豆样本中优选出四个特征波长，得到特征图像；从每个特征图像中分别提取基于灰度共生矩阵的4个纹理特征参数——能量、熵、惯性矩和相关性，从16个特征变量中选取8个主要特征变量，应用BP神经网络建立大豆品质分级识别模型。模型预测准确率为92％。结果表明，高光谱图像技术对大豆等级具有较好的识别作用，可为大豆的在线无损检测分级提供参考。

In order to fast and exact classification of soybean,collection 1-5 grades soybean 1 000- 2 500 nm range of hyperspectral image data to obtain spectral image; analysis of different samples of soybean grade spectral curve; application of principal component analysis （PCA）, from the 4 features of each variety selected optimal wavelength, extracted four texture feature parameters （moment of inertia, energy, entropy and correlation）from each feature in the image based on statistical moment. Select 8 main characteristic variables from 16 characteristic variables, establishment of soybean grade identification model based on BP neural network. Experimental results showed that discriminating rate was 92% in the prediction set. Results showed that the hyperspectral image technology had better recognition effects on soybean grade, Provided a good reference for soybean online non-destructive testing classification.