The objective of this study is to compare portable visible spectral imaging (443–726 nm) and conventional RGB imaging for detecting products stored beyond the recommended “use-by” date and predicting the number of days poultry products have been stored. Packages of chicken thighs with skin on were stored at 4 °C and imaged daily in pack through plastic lidding film using spectral and RGB imaging over 10 days. K-nearest neighbour (KNN) models were built to detect poultry stored beyond its recommended “use-by” date and partial least squares regression (PLSR) models were built to predict the storage day of samples. Model overfitting in the spectral PLSR model was prevented using a geostatistical approach to estimate the number of latent variables (LV). All models were built at the object level by using mean spectra and colour values per image. The KNN model built using spectral images (acc. = 93 %, sen. = 75 %, spec. = 100 %) was more suitable than the model built using RGB images (acc. = 80 %, sen. = 42 %, spec. = 96 %) for detecting poultry stored beyond its “use-by” date. The PLSR model built using spectral images (R² = 0.78 RMSEC = 0.92, RMSEV = 1.11, RMSEP = 1.34 day) was more suitable than the model built using RGB images (R² = 0.60, RMSEC = 1.66, RMSEV = 1.67, RMSEP = 1.92 day) for predicting storage day of poultry products.

Keywords: portable spectral imaging, K-nearest neighbour (KNN), partial least squares regression (PLSR), poultry, “use-by”, plastic lidding film