Abstract: The peroxidation value of pork belly was rapidly and nondestructively detected by near infrared(NIR)hyperspectral technique combined with continuous projection algorithm(SPA). Spectral images of samples were acquired through the high spectral spectroscopy imaging system,and their reflection spectrum information was extracted,after seven kinds of pretreatment,including baseline correction(BC),gaussian filter smoothing(GFS),and median filtering smoothing(MFS),savitzky golay smoothing(SGS),moving average smoothing(MAS),standard normal variate(SNV),and multiplicative scatter correction(MSC),partial least squares(PLS)were used to establish the prediction model,and used the regression coefficient method(RC)and SPA to select the optimal wavelength,multiple linear regression(MLR)and PLS optimization model. The results showed that the full-band PLS model(F-PLS)with BC preconditioning(R_P=0.960,RMSEP=5.15×10^-4 g/100 g)and RAW data(R_P=0.960,RMSEP=4.89×10^-4 g/100 g)had better prediction effect on peroxidation value,and the optimization showed that the MLR optimization model of RAW(R_P=0.968,RMSEP=4.12×10^-4 g/100 g)had better prediction effect. The results showed that this method can predict the changes of pork under different refrigeration conditions.