Development of low cost optical sensors for produce quality evaluation

Abstract:

Near infrared (NIR) spectroscopy is powerful tool for non/destructive measurements of various quality parameters. Moreover, the performance of NIR spectroscopy for quality evaluation is dependent on two key components 1) suitable spectrometers and 2) appropriate calibration models. The aim of this study was 1) to develop predictive models for quality parameters of mangoes and tomatoes using different commercial spectrometers, 2) to construct a prototype of an in/house NIR spectrometer and investigate the possibility to use it as a source of spectral data for the development of calibration models for quality parameters of mangoes and tomatoes. This work focuses on the goal of determination quality parameters of fruits and vegetables: mangoes and tomatoes. Dry matter (DM), total soluble solids (TSS), titratable acidity (TA), pH, and firmness were selected as the key quality parameters in this study. The calibration models were developed using partial least squares regression (PLSR) and the data analysis used both unprocessed data and preprocessed data (e.g. Savitzky/Golay derivative, SNV). The possibility to perform the prediction of quality parameters of mango and tomato samples was evaluated using different commercial spectrometers (SCIO, Linksqure, Texas Instruments NIRscan Nano, and Neospectra). In case of mango samples, good predictive models were developed for DM, TSS, TA, and pH using the spectroscopic measurements from SCIO and Linksqure operating in both visible and NIR modes. The best model for DM using SCIO spectrometer exhibited a cross validation values of 0.92 and 0.739% for R2 and RMSE, respectively. The best predictive models for TSS, TA, and pH parameters were developed using Linksqure operated in the visible mode. The R2 values of calibration and cross/validation (brackets) for TSS, TA, and pH were 0.91 (0.75), 0.91 (0.79), and 0.93 (0.81) respectively. The RMSE values of calibration and cross/validation (brackets) for TSS, TA, and pH were 1.03 oBrix (1.76 oBrix), 0.38% (0.58%), and 0.21 (0.35), respectively. Poorly performing predictive models with modest R2 values were obtained using spectral data from Texas Instruments NIR Scan Nano and Neospectra instruments. For the work with tomatoes, cherry tomato was chosen for the test of quality parameters. Only three commercial spectrometers (SCIO, Linksqure and Texas Instruments) were utilized in this part because of the sampling window of Neospectra is too large to allow the spectroscopic measurements. Good predictive models were developed for predicting DM and firmness using the spectroscopic measurements taken with SCIO and Linksqure operating in both visible and NIR modes. The best model for DM was obtained using spectral data from the SCIO spectrometer and has exhibited a cross validation values of 0.89 and 0.27% for R2 and RMSE, respectively. For the firmness, the best results were obtained using spectral data acquired using the Linksqure instrument operating in visible mode. The R2 values of calibration and cross/validation (brackets) were 0.91 (0.87). The RMSE values of calibration and cross/validation (brackets) for firmness were 0.91 N (0.87 N). The performance of models for predicting quality parameters based on spectral data acquired using the Texas Instruments NIRscan Nano were poor with modest R2 values exhibiting similar results as for the work carried out with mangoes. Given the encouraging results obtained with commercial low cost NIR instruments in the first part of this work, we proceeded to the second part where an in/house NIR spectrometer prototype was constructed and evaluated. The performance of an NIR spectrometer depends on three key components: light source, wavelength selector, and detector. The prototype of a potentially low cost portable NIR spectrometer has been constructed around the Hamamatsu C14384MA/01 sensor. The in/house spectrometer prototype had been made in two version using different light sources. The first version used an NIR LED (SFH 4376, OSRAM) light source while the second version used a tungsten halogen filament bulb (TH). These spectrometers operated in the wavelength range from 650 to 1050 nm. The performance of the spectrometer prototype was then tested by using it to collect spectral data from mangoes and tomatoes for the purpose of developing predictive models for selected quality parameters. In case of mango samples, good predictive models were obtained for predicting DM, TSS, TA, and pH using both NIR LED and TH light sources. The best models for predicting DM were obtained using the spectrometer version with the TH filament light source. The R2 values of the test set was 0.82. For the best models for TSS, TA, and pH were obtained using data acquired with the prototype equipped the NIR LED. The R2 values of the test sets for TSS, TA, and pH were 0.86, 0.92, and 0.86, respectively. Models developed for the prediction of firmness were poor with moderate R2 values in the case of both spectrometer versions. In conclusion, the in/house spectrometer prototype has been used to collect spectroscopic data from Nam Dok Mai mangoes, which were collected in two different harvesting seasons. Predictive models for mango quality parameters (DM, TSS, TA, pH, firmness) were developed from this spectroscopic data. Models with satisfactory quality (R2 > 0.80 in the test set) were developed for DM, TSS, TA, and pH. The results indicate that the constructed instrument can collect usable spectroscopic data from produce samples. In the case of tomato samples, predictive models of modest quality were developed for all quality parameters, with R2 values of the test set below 0.70 in all instances. The performance of predicting DM, TSS, TA, pH, and firmness using both NIR LED and TH filament light sources were significantly worse than predictive models reported in previous publications. On the other hand, the predictive models of in/house spectrometers show better performance in comparison with previous prototype (MOEMS technology) for predicting, TSS, DM, TA, and pH for tomato samples. In conclusion, the potential of low cost NIR spectrometer using new generation of MOEMS technology (C14383MA/01) for rapid and non/destructive measurement of tomato samples was evaluated. The results showed that the predictive models can be used to predict DM, TSS, and pH. The predictive models with satisfactory quality (R2 > 0.50) have been developed for DM, TSS, and pH. But for the TA and firmness yielded poor prediction performance.