基于拉曼散射图像的低温肉制品品质无损检测技术研究

Due to special nutritional components and organizational structure, low-temperature meat products have been widely enjoyed by consumers and become the trend of future meat product development. However, low-temperature meat products are prone to spoilage, resulting in potential food safety risks. In this proposal, the Raman scattering imaging technology is introduced to detect the quality of low-temperature meat products, which can solve the problem that the quality of low-temperature meat products is difficult to detect (monitor) in real-time because of the influence of packaging materials. Fluorescence background correction and separation analysis methods of Raman scattering images will be studied to overcome the influence of packaging materials on Raman scattering images. The regularity of Raman scattering images acquired from low-temperature meat products changing with food additives, physicochemical and microbiological criteria will be analyzed, and the mechanism of meat quality detection based on Raman scattering images will be studied. Image features of Raman scattering images and light intensity and distribution characteristics of scattering profile will be extracted to realize the full excavation of feature information. The kernel transform and sparse self-coding will be applied to fuse and reconstruct the original features, and then develop intelligent detection and analysis models of meat quality. Finally, a complete real-time nondestructive detecting system for low-temperature meat products based on Raman scattering images will be developed. The results will be of great significance to ensure the safety of food consumption and improve the nondestructive detecting technique for the quality of packaged meat products in China.

低温肉制品因其独有的营养成分和组织结构，受到消费者的广泛欢迎，已成为未来肉制品发展的方向。但低温肉制品易发生腐败变质，造成潜在的食品安全风险。项目将拉曼散射图像技术引入到低温肉制品的品质检测中，解决当前低温肉制品品质由于包装材料存在，而难以大批量实时检测（监测）的问题。研究拉曼散射图像的荧光背景校正及分离解析方法，以克服包装材料对拉曼散射图像的影响；分析低温肉制品添加剂、理化和微生物指标与其拉曼散射图像的变化规律，探明拉曼散射图像进行肉制品品质检测的机理；提取拉曼散射图像的图像特征、空间散射曲线光强及其分布特征，实现其特征信息的充分挖掘；利用核变换、稀疏自编码方式对原始特征进行深度融合和重构，并构建肉制品品质智能检测与分析模型；形成一套完整的基于拉曼散射图像技术的低温肉制品品质实时无损检测方法体系。研究成果对于保障食品消费安全，提高我国包装肉制品品质无损检测技术水平具有重要意义。

项目针对传统光学检测方法难以大批量实时检测带包装低温肉制品品质问题，以拉曼散射图像技术为检测手段，并提出一系列解决算法和模型，最终形成一套基于拉曼散射图像技术的低温肉制品品质实时无损检测方法体系。研究内容包括：（1）在拉曼散射图像处理方面，利用点-线光源转化技术减少手动操作误差，提高数据检测速度和可靠性；利用基于信息熵的区域分割技术获取有效信息，进一步提升检测效率和精度；利用基于插值的拉曼光谱连续背景的校正方法剔除拉曼检测中的荧光干扰；基于Voigt函数拉曼谱峰拟合方法和结合LMF算法重叠峰分离方法，对光谱曲线进行重构；对于建模分析过程，通常还采用多元散射校正补偿拉曼光谱中的加性和乘性效应。通常根据应用背景和分析目的不同使用一种或多种以上的数据处理方法。（2）在拉曼散射图像检测分层或有表面干扰的低温肉制品品质中，首先采用次表层的信号识别和分离手段提取数据。通过选择最佳偏移距离能有效增强次表层食品的拉曼信号；由于谱峰对与成分检测的重要作用，采用广义高斯模型对次表层拉曼谱峰进行识别；针对无先验信息的多层物质，通过盲源分离手段，重构次表层拉曼的全谱，保留弱峰和重叠峰信息。（3）在拉曼散射图像特征选择与降维方面，首先对拉曼峰对物质的表征和对应关系进行了研究；通过谱峰位置、形态、面积等先验知识，PCA等无监督降维方法，随机森林特征重要度排序等方法，提取有效特征并进行数据降维。（4）在低温肉制品品质模型构建与优化方面，针对不同的检测需求探究了支持向量数据描述单分类模型、偏最小二乘多分类模型、随机森林回归预测模型对冷鲜肉质品品质检测，同时还通过搭建谱库，研究基于稀疏非负最小二乘、相似度的混合物拉曼光谱定性分析方法以及已知混合物增量搜索的定量分析模型，研究成果对于提高我国包装肉制品质量检测的自动化水平具有重要意义。