掺假羊肉的温度响应及高光谱图像和热成像融合识别研究

Mutton adulteration becomes a more and more prominent food safety problem in China. With the enhancement of adulteration level, the traditional method of sensory examination is difficult to identify adulteration. In addition, physical and chemical analysis methods, for instance, PCR, are time and manpower consuming, and suitable for precise laboratory identification. These methods cannot meet the development requirement of rapid and efficient market regulation. The project takes the mutton, which is a famous product with distinct regional characteristic, as the research object and is aiming at the common adulteration phenomena of pork, duck and fox into mutton. By heating pretreatment to introduce the temperature disturbance, the difference between true and false mutton is expanded and their true nature is presented. Firstly, starting from the aspect of the thermal physical properties, the rules of heat conduction of different meats under different heating treatment conditions are studied, and the thermal infrared imaging technology is used to obtain the temperature distribution characteristics of meat under the optimized conditions. Secondly, starting from the aspect of the morphology and chemical properties, the change rules of surface morphology and ingredient contents of different meats under different heating treatment conditions are studied, and the hyperspectral imaging technology is used to obtain the meat morphology and ingredient variation characteristics under the optimized conditions. Then by using different chemometric and pattern recognition methods, the mutton adulteration recognition models using the thermal infrared imaging information, the adulteration recognition models using the hyperspectral imaging information, and the fusion recognition models using the hyperspectral imaging and thermal infrared imaging information together are established and compared. Finally, the eventually rapid and accurate pattern recognition method of mutton adulteration is determined and developed, which can provide technical support and theoretical basis for the supervision of mutton quality and the development of fast detection equipment for mutton adulteration.

羊肉掺假是我国比较突出的食品安全问题。随着掺假水平提升，感官检验难以识别，PCR等理化方法耗时耗力，适用于实验室精确鉴定，难以满足市场监管快速高效的筛查要求。本项目以掺假羊肉为对象，针对其中常掺入猪肉、鸭肉和狐狸肉等现象，通过加热引入温度扰动的预处理手段，将真假羊肉的差异进行扩大，使其在加热过程中“原形毕露”，首先从热物理特性出发，研究不同加热处理下不同肉类的热传导规律，采用热红外成像获取优化条件下肉品的温度分布特征；其次从形态和化学特性出发，研究不同加热处理下不同肉类的表面形态和成分变化规律，采用高光谱图像获取优化条件下肉品的形态和成分含量变化特征；然后采用不同化学计量学和模式识别方法，分别建立基于热成像、基于高光谱图像、基于高光谱图像和热成像信息融合的掺假羊肉识别模型；最终比较确定不同掺假羊肉快速、准确的优化识别方法，从而为羊肉品质监管和掺假羊肉快速检测设备开发提供技术支撑和理论依据。

本项目针对羊肉中常掺入猪肉、鸭肉和狐狸肉等现象，分别基于热成像、RGB 图像、高光谱图像（HSI）及其融合技术开展了掺假羊肉检测机理和方法研究。首先通过试验分析确定了掺假羊肉热成像和 RGB 成像的加热方式、功率、时间等关键采集参数，研究加热条件下真假羊肉的图像感兴趣区域（ROIs）提取方法及温度特征参数变化规律、温度分布差异、外部形态特征参数变化规律。其次，基于 HSI 研究并解析加热处理对肉品光谱的影响，分析了不同肉类（羊肉、猪肉、牛肉、鸡肉和鸭肉）的光谱特征变化规律，并结合二维相关光谱建立肉品分类方法。然后，基于掺假羊肉的热像图、RGB 图像和 HSI 图谱信息，结合不同化学计量学和深度学习方法分别建立了不同肉类（猪肉、鸭肉和狐狸肉）掺假羊肉的定性和定量检测模型。此外，利用 HSI 技术分别探索了脂肪影响和食品添加剂（羊肉精和染色剂）作用下的掺假羊肉的光谱和图像信息提取方法和特征波段筛选方法，使用特征波段下的光谱和图像信息分别建立了掺假羊肉定性与定量识别模型，并开发了相关应用软件。最后结合多特征融合思想，构建了特征波段下 HSI 图谱信息融合的猪肉掺假羊肉判别和预测模型，改善了脂肪影响下掺假羊肉的预测精度；针对羊肉精和染色剂作用下多部位猪肉（背脊、前腿、后腿）掺假羊肉，构建并行卷积神经网络提取单一部位深层特征然后进行融合，基于深层融合特征建立的多部位猪肉掺假羊肉判别模型外部验证集的准确率有较大提高；通过图像拼接和通道堆叠的方式获取热像图和 RGB 图像的融合图像，并基于多源融合信息建立了掺假羊肉定性和定量模型，研究结果表明，外部热场的引入可以扩大真假羊肉间的信息差异，基于热成像、RGB 图像、HSI 及其多源特征融合方法可以有效实现掺假羊肉的快速检测和分析。该项目研究成果可以为羊肉品质监管和掺假羊肉快速检测设备的开发提供技术支撑和理论依据。