Quality evaluation and variety discrimination of pear, apple, nectarine and potato using optical properties coupled with machine learning

The optical properties of fruits and vegetables are essential for elucidating light propagation mechanisms within tissues, critical for enhancing spectroscopy-based quality assessments. This study employed a single integrating sphere system to determine two key parameters, which are the absorption coefficient ( μ a ) and reduced scattering coefficient ( μ s ′) across ten varieties, including four pear cultivars, and two each of apple, nectarine, and potato cultivars. Based on these measurements, conventional chemometric methods and machine learning algorithms were applied to develop models predicting soluble solids content (SSC) and flesh firmness (FF), in individual- and combined-variety samples. The partial least squares regression (PLSR) models based on μ a showed superior performance for pears, with R p 2 values of 0.877 and 0.884 for SSC and FF, respectively. In mixed fruit samples (apples and nectarines, or combinations with pears), μ a -based PLSR models yielded comparable SSC prediction performance. For qualitative classification, the RF model using μ a achieved high accuracies, with 92.9 % for pears and apples, and 100 % for nectarines and potatoes. Random forest (RF) models based on μ s ′ also demonstrated robust performance, exceeding 85 % accuracy for apples and nectarines. This study demonstrates an effective approach that combines optical properties and machine learning for quality evaluation and variety discrimination.