基于计算机图像处理的沁州黄小米产地判别研究

为了沁州黄小米的产地溯源和品牌保护，本研究采集沁州黄小米地理标志保护区及域外小米样品，利用计算机图像处理技术对来源于沁州黄小米保护区（核心区和非核心区）和域外区3个产区的小米有效颜色特征进行提取。分别建立Fisher判别、K最近邻判别和BP人工神经网络3种判别模型并进行比较。采用紫外可见光谱法和荧光定量PCR法对不同产区小米的总类胡萝卜素含量及其降解途径中的关键基因进行差异分析。结果表明：不同产区的沁州黄小米在RGB色彩模式中B值存在显著差异（P＜0.05）；3种判别模型中，BP人工神经网络判别模型正确率最高为96.7%，具有可行性。进一步分析表明，小米颜色参数变化与总类胡萝卜素含量有显著相关性，并且不同产区之间总类胡萝卜素含量存在显著差异（P＜0.05），核心产区小米的总类胡萝卜素含量最高，达到15.89 mg/kg；此外，沁州黄核心区小米中的类胡萝卜素裂解双加氧酶的表达量显著低于非核心区和域外区（P＜0.05）。本研究运用计算机图像处理技术和人工神经网络相结合的判别方法对沁州黄小米的不同产地进行了正确区分，具有一定的应用价值，可为沁州黄小米的产地溯源和品牌保护提供技术支持，同时对类胡萝卜素含量及降解途径中关键基因进行比较分析，从生化和分子水平揭示了引起不同产区沁州黄小米图像和品质差异的本质原因。

Research on Origin Identification of Qinzhouhuang millet Based on Computer Image Processing

In order to trace the origin and protect the brand of Qinzhouhuang millet, this study collected the Qinzhouhuang millet from geographical indication area (core area and non-core area) and other millet samples. The effective color features of millet from three production areas were extracted. Three discriminant models, Fisher discriminant, K-nearest neighbor discriminant and BP artificial neural network, were established and compared. Subsequently, the total carotenoid content of millet and the expression of key genes in the carotenoid degradation pathway were determined by UV-vis spectroscopy and fluorescence quantitative PCR, respectively. The results showed that significant differences (P<0.05) were observed in B value of RGB color pattern. Moreover, among the three discriminant models, the BP artificial neural network discriminant model has the highest accuracy rate of 96.7%, which was feasible and credible. Further analysis suggested that the color parameters of millet and the total carotenoid content showed significant correlation, the total carotenoid content differed significantly (P<0.05) among different millet production areas, and the carotenoid content of core areas up to 15.89 mg/kg, which was higher than non-core area and outside area. In addition, the expression level of carotenoid cracking dioxygenase in the core area of Qinzhouhuang millet was significantly (P<0.05) lower than that in the non-core area and outside area. Overall, the computer image processing technology with artificial neural network method could be used to distinguish the different production areas of Qinzhouhuang millet, and provide technical support for the origin tracing and brand protection of the millet. Meanwhile, the carotenoid content and key genes in the degradation pathway were compared and analyzed, this study indicated essential reasons for the difference in image and quality of Qinzhouhuang millet from different production areas at biochemical and molecular levels.