【摘要】 为了实现燕麦脂肪含量的无损、准确、快速测定,采集燕麦400-1000nm范围的高光谱图像数据,使用多元散射校正方法处理原始光谱,采用分段线性回归分析方法对燕麦光谱数据进行降维,将光谱按照全波段、100nm间隔、200nm间隔进行划分,并分别计算出每段的特征拐点波长,将光谱特征拐点波长的反射率作为BP神经网络的输入向量,中间计算隐含层神经元的传递函数采用S型正切函数tansig,输出向量是一维目标向量,采用BP神经网络对燕麦脂肪含量进行预测。结果表明:燕麦脂肪含量的预测值与国标测定值之间的决定系数R~2达到0.78,预测均方根误差(RMSEP)达到0.006。研究结果为燕麦脂肪含量的快速无损检测提供了理论依据和实用方法。更多还原

【Abstract】 In order to achieve rapid, accurate, and non-destructive determination of fat content in oat. We acquired the hyperspectral of oat in the range of from 400 to 1000nm, preprocessed the original spectral by using multiplicative scatter correction method, and processed the spectral data with dimension-reduction treatment by segmented linear-regression analysis. The spectrum is divided according to the whole band, 100nm interval, and 200nm interval, and the characteristic inflection point wavelength of each segment is calculated respectively. We used the reflectance of the wavelength of the spectral characteristic inflection point as the input vector of the BP neural network and the content of fat in oat as the output vector of the BP neural network. At the same time, the hidden layer neurons were operated by s-shaped tansig as the transferring function, we predicted the content of fat in oats by the BP neural network. The correlation coefficient R~2 between the prediction value of the BP neural network and the national standard measurement value was 0. 78, and the root-mean-square error reached 0. 006. The results provide a theoretical basis and practical method for rapid non-destructive testing of fat content in oats,更多还原