Hyperspectral imaging is a non-contact, nondestructive technique that combines spectroscopy and imaging to extract information from a sample. This technology has recently emerged as a powerful technique for food analysis. In this study, the potential of hyperspectral imaging (HSI) to predict white button mushroom moisture content (MC) was investigated. Mushrooms were subjected to dehydration at 45 ± 1 C for different time periods (0, 30, 60 and 120 min) to obtain representative samples at different moisture levels (93.40 ± 0.62%, 82.76 ± 2.11%, 73.20 ± 2.60% and 60.89 ± 4.32% wet basis [wb]). Hyperspectral images of the mushrooms were obtained using a pushbroom system operating in the wavelength range of 400–1000 nm. Hunter L, a and b colour values of the mushrooms were also measured. The average reflectance spectra of samples at different MC levels were obtained and Partial Least Square Regression (PLSR) models were built to predict mushroom moisture content. To reduce the spectral variability caused by factors unrelated to MC such as scattering effects and differences in sample height, different spectral pre-treatments were applied. The Standard Normal Variate (SNV) transformation was found to be the best approach among the wavelength range studied, resulting in the greatest reduction in Root Mean Square Error of Cross Validation (RMSECV) and Root Mean Square Error of Prediction (RMSEP) for a 4-component PLSR model. RMSECV of 5.50 (% wb) and RMSEP of 5.58 (% wb) were obtained for the calibration and test sets of data, respectively. Prediction maps were generated from hyperspectral data to show the predictive model performance at pixel level. This study shows the potential of hyperspectral imaging for prediction of mushroom moisture content in the studied wavelength range. The implemented method highlighted contrast between areas of different moisture content to achieve better knowledge of dehydration distribution over the mushroom surface.

Keywords