Detection Of Mutton Fat And Protein Contents Using Near-infrared Hyperspectral Imaging Technique

The contents of mutton fat and protein are the most important indicator to evaluate nutrition value, but the traditional detection methods are operational complexity, time-consuming and power-consumption, it is more difficult to realize on-line and rapid detection for large quantities of samples. As a new photoelectric detecting technique, NIR hyperspectral imaging technique can obtain the spectral and image information of each point of samples space simultaneously. Because of the rapid, non-destractive, environmental protection and accuracy advantages, NIR hyperspectral imaging technique are becoming more and more attention in detecting the meat quality recently.As an object of small-tail Han Sheep in Ningxia, this study adopted NIR hyperspectral imaging system (900-1700nm) to get image information of72mutton samples, established the quantitative prediction model combined with chemometrics methods, and provided a theoretical basis for the on-line and non-destructive detection of the mutton quality. The main contents are as follows:(1) To build the hyperspectral imaging system, proceed system calibration and the determination of acquisition parameters.(2) To acquire the spectral and image information and proceed spectral preprocessing by MSC.(3) To use linear regression analysis coupled with actual chemics value, obtain optimal wavelengths by absolute value of correlation coefficient.(4) To establish the prediction models on the contents of mutton fat and protein by BP neural network and partial least squares, and get the optimal models. The results showed that the prediction effect of models were very well by the partial least squares method. Correlation coefficients (R) of the contents of fat and protein were0.95and0.92, root mean squared error of prediction (RMSEP) were0.40and0.80respectively.