Preliminary Evaluation of NIR Spectroscopy Combined with PLS-DA for Non-Destructive Detection of Mechanical Injury in Avocado

Avocado (Persea americana) is a high-value tropical fruit rich in lipids, essential fatty acids, and lipophilic vitamins, representing an important agricultural commodity worldwide. However, avocados are highly perishable and particularly vulnerable to mechanical injuries during harvesting, handling, packaging, and transportation. Impact injuries can cause plastic deformation, cell wall disruption, and biochemical alterations that promote tissue softening and enzymatic pulp browning. Importantly, the thick avocado peel can conceal early internal tissue damage, preventing visual detection at the time of injury. Consequently, damaged fruits may pass through postharvest supply chains undetected, contributing to accelerated deterioration, increased pathogen susceptibility, and economic losses. Near-infrared (NIR) spectroscopy offers a rapid and non-destructive approach capable of probing molecular vibrations, particularly O–H and C–H bonds, associated with structural and compositional changes in fruit tissues. This study evaluated the preliminary feasibility of NIR spectroscopy (1350–2500 nm) combined with Partial Least Squares Discriminant Analysis (PLS-DA) for the non-destructive detection of mechanical injury in avocados. Two classes were considered for PLS-DA modeling: injured and uninjured fruits. Spectra were acquired using handheld portable NIR (Neo-Spectra ProxiScout^TM) from six fruits (12 measurements), 2 measures per fruit 72 h after a controlled impact of 0.77 J, generated using a 131 g steel sphere dropped from a height of 60 cm. Spectral data were preprocessed using a first derivative with a Savitzky–Golay filter (2nd-order polynomial, 7-point window), followed by mean centering. A preliminary PLS-DA model with one latent variable (LV) was constructed and evaluated using Venetian blinds cross-validation with eight splits. In calibration, the model achieved sensitivity, specificity, and accuracy values ranging from 87.5% to 100% for both classes. In cross-validation, injured fruits showed 75% sensitivity, 50% specificity, and 60% accuracy, whereas uninjured fruits showed 50% sensitivity, 75% specificity, and 56% accuracy. These preliminary findings indicate that NIR spectroscopy captures spectral changes associated with mechanical injury in avocado; however, discrimination performance remains limited by the small dataset size. Further studies should include a larger number of samples from different fruit lots and maturity stages to improve model robustness and evaluate its generalization capability through external validation using independent samples.

Acknowledgments: The authors acknowledge financial support from CNPq - Process nº 142639/2026-5].