基于高光谱技术的有机鸡蛋与普通鸡蛋鉴别

当今全球范围内有机食品行业发展迅速，体现出消费者对食品质量安全的重视。相比于普通鸡蛋，有机鸡蛋因严格的生产条件以及更高的营养价值生产成本更高、售价更贵。市面上所销售的有机鸡蛋虽取得了严格有机食品认证标识，但依旧不能阻止不法份子将普通鸡蛋冒充有机鸡蛋销售，从而谋取利润。这一行为不仅会损害有机鸡蛋生产商的利益，也降低了人们对有机食品的信任度，为此需要一种有效的对有机鸡蛋和普通鸡蛋无损鉴别方法。使用高光谱技术透射成像的方式，可以获取物质内部的信息，以有机鸡蛋和普通鸡蛋为试验对象，采集鸡蛋样本364～1 025 nm波长范围的高光谱图像数据，从采集到的数据中提取出鸡蛋蛋清和蛋黄感兴趣区域(ROI)的平均透射光谱数据。根据透射光谱曲线图筛选出有机鸡蛋与普通鸡蛋光谱响应差异明显的波段区间，分别通过偏最小二乘判别分析(PLS-DA)和支持向量机(SVM)建立鸡蛋类别的鉴别模型，结果表明模型分别根据蛋黄区域与蛋清区域数据进行建模的鉴别准确率相近，进一步对蛋黄区域数据进行分析。由于高光谱数据量大且存在大量冗余信息，给数据采集、存储、传输和建模处理都带来不便，因此分别通过连续投影算法(SPA)和竞争性自适应重加权算法(CARS)对蛋黄ROI数据进行降维处理，剔除了大量冗余信息后再建模。最终，使用对蛋黄ROI区域运用SPA降维后得到的23个特征波长建立的SPA-SVM鉴别模型在测试集的准确率最高达到94.2%。结果表明，通过高光谱技术对有机鸡蛋和普通鸡蛋进行无损鉴别有一定效果。

The Discrimination of Organic and Conventional Eggs Based on Hyperspectral Technology

Today, the organic food industry is developing rapidly around the world. It reflects consumers’ attention to the quality and safety of food. Organic eggs are produced under strict conditions, and it have nutrition, so the more price it is compared with conventional eggs. Although there are some strict certification processes to the eggs sold in the market, they still cannot prevent illegal elements from making profits by replacing organic eggs with conventional eggs. This phenomenon harms the interests of organic producers, and consumers will have less faith in organic food. Therefore, an effective non-destructive method is needed to identify the organic eggs from conventional eggs. One material’s inner information can be obtained by hyperspectral transmission image technology. In this paper, the organic eggs and conventional eggs were used as the experimental objects, and hyperspectral image data of egg samples were collected in the wavelength range from 364 to 1 025 nm, and the average spectral of the ROI in the area of albumen and yolk were abstracted from the collected data respectively. According to the transmission spectrum curves, bands with obvious differences in spectral response between organic eggs and conventional eggs were selected out. The Partial Least Squares Discriminant Analysis (PLS-DA) and Support Vector Machine (SVM) were used to establish the discrimination model. The results show that the accuracy of the four models based on the yolk and albumen area respectively are closed, further analysis was carry on the datas of yolk area. Due to a large amount of hyperspectral data and redundant information, it is inconvenient for data storage, transmission and modeling. Therefore, Successive Projections Algorithm (SPA) and Competitive Adaptive Reweighted Sampling (CARS) were used to reduce the dimensions of data. After removing a lot of redundant information, the SPA-SVM discrimination model based on 23 wavelengths selected by using SPA on the hyperspectral data of yolk area has the highest accuracy, reaching 94.2%. The results show that the hyperspectral technique has some effect on the non-destructive identification of organic eggs and conventional eggs by hyperspectral technique has some effect.