The Research Based On Hyperspectral Image Technol Ogy On The Nondestructive Examination For Beef Quality

Beef is very popular for its high nutrition and edible value. With the improvement of material standard of living, people’s demand for the quality and edible species of beef also gets higher and higher. Generally, beef should be edible with 48 hours’ refrigeration after being butchered. However, the quality of beef will be affected on all kinds of physical effects and chemistry actions due to various processing methods and transport means during refrigeration. Usually, people can not directly identify the beef quality by the naked eye and traditional test methods do not have high efficiency and accuracy. This paper aims to test the beef quality though hyperspectral image technology to provide theoretical basis for nondestructive examination of agriculture products. Following is the main contents of this experiment:Hyperspectral images based on spectral characteristics of moisture content of beef is forecasted. By automatically on beef of hyperspectral image segmentation,By comparing the predicting results of the model after being matched of different methods, after variable standardization pretreatment, best result of PLS calibration model under characteristic wavelength can be achieved though principal component analysis, the determination coefficient set of correction is 0.94, the RMSE is 4.973, the determination coefficient of prediction set is 0.92 and the RMSEP is 5.868, which is most suitable to inspect beef ’s moisture content.Hyperspectral image technology on prediction of beef ’s protein content. After automatic segmentation of beef ’s hyperspectral image, different pretreatment methods are adopted to process the obtained spectral data and characteristic variables are selected by principal component analysis to build the PLS calibration model. The results show that best function of the model can be achieved after variable standardization pretreatment on spectral data. The determination coefficient set of correction is 0.94, the RMSE is 3.137, the determination coefficient of prediction set is 0.93 and the RMSEP is 5.298, which realizes the prediction of beef ’s protein content.Based on image texture feature to predict the moisture content and protein content of beef. Collect beef samples in 400-1000 nm band of hyperspectral image data, through the principal component analysis(pca), 6 and 3 characteristic wavelengths respectively. In a characteristic wavelength image were extracted based on gray level co-occurrence matrix of energy, entropy, moment of inertia and the correlation of four texture characteristic parameters. The use of principal component analysis(pca) for secondary dimension reduction to identify variables, moisture content and the application of BP neural network method to establish beef protein recognition model.