An autonomous Internet of Things spectral sensing system for in-situ optical monitoring of grape ripening: design, characterization, and operation

[Display omitted] •An IoT spectral sensing system for monitoring the grape berry ripening process through reflectance signals was developed.•The system was evaluated in both lab and field environments for red and white grapes.•The PCA analysis of the reflectance data collected showed a trend related to the grape berry ripening process.•A Partial Least Squares model for predicting the Total Soluble Solids content in red and white grapes was applied. The present work proposes a novel autonomous Internet of Things (IoT) spectral sensing system for in-situ optical monitoring of grape ripening through reflectance signals. To this end, tailor-made hardware for this IoT end node was developed, characterized, and operated in both lab and field conditions. It included three complementary modules: the optical module, the host module, and the controller module. The optical module included four photodetectors and four LEDs with maximum emission wavelength centered at 530, 630, 690, and 730 nm that was placed in direct contact with the grape berry. The host module included the LED driver and the analog front-end for signal acquisition. Finally, the controller module provided full control of the system and ensured data storage, power management, and connectivity. The system was capable of measuring reflectance in the range of 4 – 100 % with a linear response (r2 > 998) and with a high reproducibility among different optical units. This design made it possible to collect reflectance signals from red (cv. Touriga Nacional) and white (cv. Loureiro) grape varieties in both lab and field environments. The relationship between this optical fingerprint (comprised of the different reflectance intensities recorded) and the evolution of grape berry quality parameters throughout the ripening period (for approximately two months), was analyzed and discussed. Lab data was used to establish a multivariate model based on Partial Least Squares for the prediction of the Total Soluble Solids (TSS) content in both varieties. The model error (Root Mean Square Error in Cross Validation) was 2.31 and 0.73 °Brix for Touriga Nacional and Loureiro, respectively. This model was applied to data acquired in the field in an illustrative example of the potential of the system to predict TSS in real time. The field observations collected during the monitoring period also provided relevant information about the potential issues that may occur during the unattended operation of the optical sensors. Add