Abstract:In order to rapidly and nondestructively detect the excessive soybean protein content in fresh minced chicken meat, hyperspectral images of three typical soybean protein adulterated in minced chicken meat were collected. Soybean protein flour (SPF), soybean protein concentrate (SPC) and soybean protein isolates (SPI) with 0~30% adulteration levels were mixed into the minced chicken meat. The hyperspectral images of prepared samples were collected in the visible and near infrared (400~1000nm) spectral range. The average spectra of regions of interest were extracted based on band operation and models based on the original and preprocessed spectra were established. Partial least squares regression (PLSR) quantitative models showed that the optimal performance were best for SPC (R2p=0.9984), followed by SPF, and then SPI. The twodimensional correlation spectrum (2DCOS) was used to extract the characteristic wavelengths to further establish multispectral models. The limits of detection (LODs) of the multispectral models were found to be 0.53%, 0.58% and 1.02% for SPF, SPC and SPI, respectively. The multispectral models above were also applied back to the raw multispectral images to visualize the distribution of different soybean proteins, as well as their adulteration levels. Results showed that the distribution differences within samples and among samples could be well visualized.