Passive reflectance sensing using regression and multivariate analysis to estimate biochemical parameters of different fruits kinds

Food quality control monitoring is crucial in food processing, due to the potential of adverse effects on the health of entire populations. The traditional biochemical measurements are based on chemical analysis techniques in the laboratory, which, despite being effective, are expensive, laborious, and time consuming, making them infeasible to obtain information on biochemical measurements in time and at large scales. In this study, the performance of non-contactless high throughput passive sensing was evaluated to estimate the biochemical parameters as well as to discriminate between fruit kinds via the application of chemometric techniques based on principle component regression (PCR), partial least square regression (PLSR) as well as simple regressions. Models of PCR or PLSR included data of the (i) spectral reflectance reading from 400 to 1000 nm and (ii) selected sixteen spectral indices that were calibrated and cross-validated for biochemical parameters prediction. Results show that the selected spectral indices showed close and highly significant associations with all measured parameters of guava, mandarin and orange fruits at three different ripening degrees with coefficient of determination (R2) reach up to (R2 = 0.87; p ≤ 0.001, R2 = 0.86; p ≤ 0.001, R2 = 0.86; p ≤ 0.001, R2 = 0.80; p ≤ 0.001 and R2 = 0.42; p ≤ 0.001) for Chlorophyll a (Chl a), Chlorophyll b (Chl b), Chlorophyll t (Chl t), soluble solids content (SSC) and titratable acidity (T. Acidity), respectively. Multivariate analysis of PCR and PLSR models showed a good prediction performance of the measured parameters. For example, the PCR based on the selected sixteen spectral indices showed that a good prediction performance was obtained with coefficient of determination (R2) of 0.85, 0.85, 84, 0.76 and 0.39, and root mean square errors of prediction of 0.052 (μg cm−2), 0.099 (μg cm-2), 0.152 (μg cm-2), 0.683 (%) and 0.0485 (%) for Chl a, Chl b, and Chl t, SSC and T. Acidity for guava fruits, respectively. As well as the PLSR based on selected sixteen spectral indices showed that a good prediction performance was obtained with coefficient of determination (R2) of 0.80, 81, 82, 0.73 and 0.22, and root mean square errors of prediction of 0.100 (μg cm−2), 0.202 (μg cm-2), 0.290 (μg cm-2), 0.457 (%) and 0.0822 (%) for Chl a, Chl b, and Chl t, SSC and T. Acidity for orange fruits, respectively. The overall results demonstrate that passive reflectance sensing can be used to evaluate the quality of different fruit types via the application of chemometric techniques as well as simple regression.