Stand-alone LED sensors for future field monitoring of grape (Vitis vinifera L.) ripeness

In the winemaking industry, the grape maturation control is a complex process that is critical to produce high-quality wines, but currently, maturation control is cumbersome and inefficient. This inefficient control of the maturation is related to a reduced value of the wine. The current state of the art for grape maturation control is based on multiple wet-chemistry assays that are: (i) destructive, (ii) time consuming, (iii) labor intensive, (iv) performed on a sparse basis and (v) based in on complex sampling process. To shift the current paradigm of grape maturation monitoring (based on lab analysis) it is needed a new technology that: (i) allows a real-time and stand-alone monitoring with a low-cost, (ii) is capable to drastically reduce the need of manpower and (iii) provides information with temporal and spatial resolution. This work aims to develop of a fully integrated stand-alone optical devices for grape quality monitoring directly in field. The main steps to fulfil the project purpose were: 1) setting up of a miniaturized low-cost and stand-alone optical prototype composed by LEDs suitable for diffuse reflectance measurements, photodetectors (PDs, CMOS), interference filters, sensors controller and power management; 2) multivariate predictive models development for the prediction of the main grape ripening parameters; 3) test the prototype in field conditions. Ergo, during the sampling campaign 2019 a first prototype version of a fully integrated optical device was developed by INL following a “stripe” design in which the spectrometric components were mounted on a long, flexible substrate which can be placed onto or inside the grape bunch. The multiple spectrometers were placed along the stripe (currently 2, module 1, M1 and module 2, M2), enabling simultaneous measurements at different parts of the grape bunch to have more representative information of the entire bunch. Four optical bands associated to the evolution of the maturation parameters of the grapes such as the development of anthocyanins and sugars, chlorophyll degradation, and decrease of water content were integrated. Four light-emitting diodes (LEDs) in the Vis and SW-NIR range were used for illumination of the grape. Placed close to these, but optically isolated using an opaque barrier, four photodiodes (with an active area of 520 × 520 µm2) assembled with spectral filters to allow intensity measurements at the desired wavelengths have been used. The components were encapsulated in a hermetically sealed yet optically transparent layer, assuring weatherproof operation of the entire system and reducing stray light. The light emitted from each LED hits the sample and the diffuse reflectance light is collected by each PD. The electromagnetic signal is converted into electronic signal and expressed in counts (arbitrary units from 0 to 4095). From each sample, 20 readouts were obtained (one readout from each PD at each LED on and one with LEDs off for background info). The optical data were collected on grape berries in a commercial vineyard owned by Sogrape, using the prototype and one commercial handheld spectrometer which works among 400 and 1000 nm, with a resolution of 0.3 nm. As reference values, the common technological parameters were analyzed on each sample. Results were encouraging underlining a small loss of information for the MLR models employing the prototypes compared to the PLS models calculated using the commercial spectrometer. Thanks to these encouraging results, a second experimental campaign was performed during the crop season 2020 using the i-Grape sensor of second generation. The data collection was performed in lab environment (using the lab version of i-Grape sensor) and in field using i-Grape sensors placed directly inside the grape bunch. In this case, four PLS models were developed for the prediction of the qualitative parameters of interest. The models were developed using 70% of the total amount of the data for calibration and 30% as external validation (prediction). In detail, it was concluded that: - The best models were obtained for the prediction of the total solids soluble, and consequently for the potential alcohol considering an R2 about 0.90 and RMSEP of 2.22 and 1.54, respectively using 6 LVs; - A very promising model was also obtained concerning the prediction of the titratable acidity with an R2 equal to 0.93 and RMSEP of 1.39 (using 4 LVs); - A pH predictive model (using 4 LVs) was developed showing a lower performance than previous parameters (R2 of 0.76 and RMSEP 0.15) but still with potential for being used with further improvements.