Prediction of quality traits in packaged mango by NIR spectroscopy

Packaging with paper bags is essential for protecting mangoes during growth, but strongly berries near infrared (NIR) spectral signals used for non-invasive analysis of their internal quality. This study focused on eliminating or minimizing the interference of paper bags on the NIR spectra of mangoes and developed innovative solutions to accurately assess mango firmness (FI), dry matter content (DMC), soluble solids content (SSC), and titratable acidity (TA). Specific NIR signals at around 1150-1250 nm and 2100-2400 nm were highlighted, which significantly reduced the precision of NIR predictions for these quality traits. A deep learning-based fully connected neural network (FNN) combined with Gaussian spatial (GS) filtering were applied as an effective strategy to mitigate the spectral interferences of packaged mangoes. Additionally, partial least squares regression (PLSR) consistently outperformed principal components regression (PCR) across all quality traits based on the spectra of packaged mangoes after FNN correction and GS filtering (PMs-FNN-GS), with R_p² and RMSEP values of 0.847 and 10.705 N, 0.932 and 0.320 % for DMC, 0.821 and 1.211 % for SSC, and 0.907 and 0.032 % for TA, demonstrating that reliable predictive accuracy for packaged mangoes was effectively achieved through the combination of deep learning and GS filtering.