【摘要】 不同种属动物源肉骨粉的快速鉴别分析技术是加强饲料监管、防范疯牛病传播的重要保障。为了探索使用显微X射线计算机断层成像技术(Micro-computed tomography, Micro-CT)快速鉴别分析不同种属动物源肉骨粉的可行性，本研究以制备的哺乳动物源牛骨颗粒和非哺乳动物源鸡骨颗粒各100个作为样品集，以不同相对位置鸡、牛骨颗粒以及鸡骨颗粒中牛骨颗粒质量分数约0.97%分别制备验证集，使用Bruker Skyscan 1275 Micro-CT对所有样品进行扫描和图像重构(管电压80 kV、管电流125μA,图像分辨率10μm,重构灰度图像灰度阶为0～255,对应X射线吸收系数为0～0.035);提取不同骨颗粒样品的感兴趣区域进行图像分割，并结合PLS-DA和SVM-DA机器学习算法分别构建鸡和牛骨颗粒分割模型。研究结果表明，鸡、牛骨颗粒图像分割感兴趣区域灰度区间为165～255,基于PLS-DA和SVM-DA模型的鸡、牛骨颗粒鉴别交互验证总准确率均为94%,验证集样品的Micro-CT三维原位可视化表征结果经验证与样品实际结果一致。结果表明，Micro-CT结合PLS-DA和SVM-DA机器学习算法进行哺乳和非哺乳动物源骨颗粒的鉴别分析是可行的。本研究为不同种属动物源饲料的快速、无损鉴别提供了新的三维原位可视化表征手段。更多还原

【Abstract】 Rapid discriminant analysis of meat and bone meal from different animal origin species is an important guarantee to strengthen feed supervision and prevent the spread of BSE disease. In order to explore the feasibility of using advanced micro-computed tomography(Micro-CT) to rapidly identify and analyze meat and bone meal from different species of animals, a calibration sample set and two validation sample sets consisted of different amount of avian origin and bovine bone particles were prepared. The Bruker Skyscan 1275 Micro-CT system was used to build method for 3D in situ visual characterization. The Micro-CT conditions for sample scaning and images reconstructing were: tube voltage of 80 kV, tube current of 125 μA, image resolution of 10 μm, reconstructed gray-scale image of from 0 to 255, and the corresponding X-ray absorption coefficient was from 0 to 0.035. The regions of interest of different bone particle samples were extracted for image segmentation. Combined with PLS-DA and SVM-DA machine learning algorithms, avian origin and bovine origin bone particle image segmentation models were constructed, respectively. Finally, the Micro-CT in situ 3D visual discriminant analysis of avian origin and bovine bone particles were carried out. The main results were as follows: the gray range of the regions of interest for image segmentation of chicken and bovine bone particles was 165～255. The total accuracy of cross validation of chicken and bovine bone particles based on PLS-DA and SVM-DA models was 94%. The Micro-CT 3D in situ visualization results of the verification set samples were verified to be consistent with the actual results of the samples. The verification results showed that the established Micro-CT 3D in situ visual discriminant analysis method achieved very consistent results with that of the actual samples. The research result can provide an imaging methodology for 3D in situ visual discriminant analysis for rapid and non-invasive identification of different species of animal origin material in feed.更多还原