基于近红外光谱和机器视觉融合技术的板栗缺陷检测

Detection of chestnut defect based on data fusion of near-infrared spectroscopy and machine vision

为提高合格和缺陷板栗分级检测识别精度,提出了近红外光谱和机器视觉的多源信息融合技术的板栗缺陷检测方法。试验以湖北京山板栗为试验对象,利用BP神经网络方法建立了基于近红外光谱、机器视觉和多源信息融合技术的板栗分级检测模型。试验结果表明,3种识别模型对对训练集板栗回判率分别为96.25%、96.67%和97.92%;对测试集板栗的识别率为86.25%、83.75%和90.00%。基于近红外光谱和机器视觉的多源信息融合技术进行板栗分级检测的方法是可行的,融合模型较单独采用机器视觉技术或近红外光谱分析技术建立模型的识别率均有显著提高。

In order to improve the detecting precision of the eligible and defected chestnut,a multisource information fusion technique based on near-infrared spectroscopy and machine vision was proposed.Chestnuts from Jingshan area in Hubei province were taken as test samples.BPNN models for discriminating chestnuts based on near infrared spectroscopy,machine vision and multisource information fusion technique were built respectively.The discriminating rates of 3 models were 96.25%,96.67% and 97.92% in training set,and 86.25%,83.75% and 90.00% in prediction set.The overall results showed that it is feasible to discriminate chestnut quality using multisource information fusion technique.The recognizing rate of integration model was higher than that of the model built by machine vision technology or near-infrared spectroscopy alone.