Non-destructive detection of color in chilled mutton based on hyperspectral technique and extreme learning machine

Near- infrared hyperspectral imaging system that was ranging from 400 nm to 1000 nm was used to finish the research of non- destructive and rapid testing of the chilled mutton colors. Hyperspectral images were taken from the 140 chilled mutton samples( storaging 1 ~7 d),and it measured the colors parameters: brightness( L*),redness( a*),yellowness( b*) and saturation( C*). Then it selected an interested area to get a representative sample spectra,and utilized the spectral data which was gotten from the preprocess synergy interval partial least squares regression( si PLS),the first derivative( FD),multiplicative scatter correction( MSC) and SNV and it was used to select the characteristic bands.It established si PLS- ELM neural network calibration model that was based on the colors parameters of the chilled mutton.The correlation coefficient( RP) of the model prediction was 0.9219,0.9391,0.9603,0.8839 which was corresponded to the L*,a*,b*and C*. The root mean square error prediction( RMSEP) was 1.1935,0.2333,0.6009,0.3586. The results showed that: the near- infrared hyperspectral imaging technologies combing with si PLS- ELM neural network was feasible for the non- destructive and rapid testing of the the chilled mutton colors.