基于多源指纹信息融合的新疆核桃产地溯源技术研究

Walnut（Juglans regin L）is an important featured produce resources of Xinjiang, the quality and safety are closely related to their geographical origin. The multi-source information fusion can be applied to walnut geographical origin tracing to improve the accuracy and stability significantly, for its complementariness of multi-source fingerprints. In this research, the near infrared spectroscopy fingerprint, mineral elements fingerprint and fatty acids fingerprints in the different varieties and GI walnuts were determination by NIR, ICP-MS/OES and GC methods, respectively. In order to select the identification indexes group of walnut geographical origin tracing and definite the reasons for changing, the regional differences and factors effect (regional, varieties, interannual, etc.) of multi-source fingerprint informations were analysed by chemometric methods. And then the origin discrimination of Xinjiang walnut was analysed in order to build the discrimination model by different multi-source information fusion, such as DS theoretical argument, Bayes discrimination and BP artificial neural network. The research findings can provide theoretical basis for the geographical origin tracing of walnut in Xinjiang provence.

核桃是新疆重要的特色农产品资源，核桃的品质、安全性与其原产地密切相关。将多源信息融合技术应用于核桃产地溯源，可实现多源指纹的信息互补，显著提高判别的准确率和稳定性。本项目以新疆不同品种和地理标志的核桃为研究对象，分别采用近红外光谱法（NIR）、电感耦合等离子体质谱/发射光谱法（ICP-MS/OES）、气相色谱法（GC）采集近红外光谱指纹、矿物元素指纹及脂肪酸指纹，建立核桃多源指纹信息库。进一步采用化学计量学方法进行多源指纹信息地域差异性及影响因素（地域、品种、年际等）分析，筛选核桃产地溯源鉴定指标群，明确地域之间指纹信息形成的原因。进而结合多源信息融合技术（DS证据理论、Bayes判别、BP人工神经网络）进行核桃产地判别分析、比较，初步建立新疆核桃的产地溯源模型。为新疆核桃产地多源指纹信息溯源技术的应用与推广提供理论依据。

本项目以新疆三个主产区（和田地区、喀什地区、阿克苏地区），以及全国11个省份的核桃样品为研究对象，分别对新疆核桃和全国核桃进行了产地溯源技术研究。对于新疆不同产地的核桃，基于脂肪酸含量和矿质元素含量的产地溯源模型，判别准确率较低。基于近红外光谱指纹建立新疆核桃产地溯源的PLSDA模型，平均产地判别准确率均达到99.00%以上，对“扎343”、“新二”和“新丰”三个主要品种平均判别准确率均达100.00%。基于核桃中红外光谱指纹建立的PLSDA判别模型对产地的平均判别准确率达85.00%以上；对“扎343”和“新二”品种的平均判别准确率均达100.00%。.对于全国11个省的核桃样品，基于全波段和特征波段近红外光谱指纹建立的溯源模型，对河北、山西、贵州、云南四个省的准确率达到90.00%以上；基于全波段近红外光谱建立的溯源模型，对“温185”、“新二”、“清香”、“香玲”和“辽核3号”的品种判别准确率分别达到100.00%、90.00%、100.00%、80.00%和100.00%。基于全波段中红外光谱建立的PLSDA溯源模型对山西、贵州、云南核桃的产地判别准确率达到90.00%以上，对“温185”、“新二”、“清香”、“香玲”、“辽核3号”品种的判别准确率分别高达100.00%、100%、100.00%、90%和100.00%。综上所述，基于近红外光谱指纹的判别模型溯源效果优于基于中红外光谱和其他品质指纹的判别模型，基于全波段光谱指纹的溯源模型判别效果显著优于基于特征波段建立的溯源模型。