高光谱成像技术检测延边黄牛肉的新鲜度

为了提高黄牛肉的质量以及食用安全性,本研究采用高光谱成像技术对延边黄牛肉新鲜度进行检测。通过成像系统采集黄牛肉样品的高光谱图像,对高光谱图像做黑白校正处理,结合S-G滤波和均值滤波对处理后的高光谱图像进行去噪处理,并采用主成分分析法对高光谱数据进行降维。采用半微量定氮法测定延边黄牛肉的挥发性盐基氮含量,通过多个BP-ANN弱分类器组建强分类器,将延边黄牛肉的挥发性盐基氮含量输入强分类器中,以挥发性盐基氮含量为指标,根据分类结果实现延边黄牛肉新鲜度的检测。实验结果表明,当波长为700~750 nm时,延边黄牛肉的挥发性盐基氮含量最高,表明肉内部被破坏的程度较高,肉类的新鲜度越差;在第10~11 d时,延边黄牛肉TVB-N质量分数变化较为明显;采用所提方法对高光谱图像去噪后,信号频率由-15 d B^15 d B区间波动变为-5 dB^5 dB区间波动,信号的频率波动较小,趋于稳定,表明所提方法可有效的去除图像中存在的噪声;所提方法的检测结果准确率最高可达99%,具有较高的检测准确率,且去噪效果较好。

Detection of the Freshness of Yanbian Yellow Beef by Hyperspectral Imaging

In order to improve the quality and safety of Yanbian yellow beef, hyperspectral imaging technique was used to detect its freshness. Hyperspectral images of Yanbian yellow beef sample were collected by the imaging system, and then subjected to black-and-white correction. The processed hyperspectral images were denoised via the combined use of S-G filter and mean filter before being processed by principal component analysis to reduce dimension. The total volatile basic nitrogen (TVB-N) content of Yanbian yellow beef was determined by the semi-micro nitrogen determination method. The strong classifiers of Yanbian yellow beef were established through multiple BP-ANN weak classifiers. The TVB-N content of Yanbian yellow beef was placed as an input into the strong classifier and taken as the index, and then the freshness of Yanbian yellow beef was detected according to the classification results. The results showed that at the wavelength of 700~750 nm, the TVB-N content was the highest, suggesting a higher degree of internal damage in meat and poorer freshness of meat. On the 10th to 11th day, the TVB-N content of the Yanbian yellow beef changed significantly. After the hyperspectral image was denoised, the signal frequency fluctuated from-15 dB~15 dB to-5 dB~5 dB, and the fluctuation of signal frequency gradually decreased and tended to be stable. These results indicate that the proposed method was highly accurate with high denoising ability: The detection method can effectively remove the noise existing in the image, and the accuracy of this method was as high as 99%.