分别在2台傅里叶变换近红外光谱仪上采集了43个栽培和野生中药材灯盏花样品的近红外漫反射光谱,提取光谱信息的15个主成分,方差贡献率达到99%以上。以20个灯盏花样品作为建模集,15个主成分作为网络学习输入层的15个节点,在2台仪器上用2套光谱分别建立了识别栽培和野生灯盏花样品的BP-神经网络模型,并对预测集的23个样品用于实际鉴别分析。两台仪器上的建模集样品模型回代正确识别率均为100%,预测集样品的正确识别率分别为100%和95.7%,结果表明,利用近红外光谱法进行栽培和野生中药材灯盏花的快速鉴别是可行的。
Forty three cultivated and wild Chinese medical herbs erigeron breviscapus were scanned on two Fourier transform near-infrared spectroscopy instruments. Twenty samples were used to set up the BP-NN models and the others were used to validate the models. Fifteen principal components, whose variance contribution rate is above 99%, were collected as input nodes for BP-NN models. The correct identification rates of calibration samples were 100% for the models on both the two instruments, and the correct identification rates of validation samples were 100% and 95.7%, irrespectively. The results showed that using NIR to fast detect cultivated and wild Chinese medical herbs erigeron breviscapus was feasible.