Using Real-Time Industrial Hyperspectral Imaging

Sweetness is the main driver for consumer liking of tomatoes and all other fruits, and Brix is commonly accepted as the proxy for the total concentration of the associated sugars. Measuring Brix has become the primary way to access the taste quality of tomatoes in the industry. Traditionally, destructive methods, such as refractometry, are employed to quantify the Brix levels in tomatoes. These techniques, while accurate and simple, present challenges when applied in phenotyping programs that require largescale, high-throughput analysis. Destructive approaches are labor-intensive, time-consuming, and involve considerable resource consumption when applied to large sample numbers. This application note highlights an alternative, non-destructive, and real-time approach for Brix estimation using hyperspectral imaging technology involving instruments designed for in-line systems operating at high speeds.

Overview
The study included over one hundred individual tomatoes (still attached to trusses) from seventeen different, carefully selected, cherry type tomato varieties to ensure a broad range of Brix levels. Spectra were collected using dual scanning with Headwall’s MV.C VNIR and MV.C NIR cameras with the perClass Mira Stage and analyzed in perClass Mira Software. All fruits were subsequently submitted to traditional, destructive, Brix measurements using a digital refractometer. Then, a regression model was developed based on the collected data in both ranges.
The dual scanning set-up allowed a complete study on comparable data. For this specific application, results were good in the VNIR, but even better in the NIR range presented in Figure 1.
For the NIR data, the average error of prediction was approximately 0.5 Brix units on the test set. A tenfold cross-validation performed on 20% of randomly selected samples consistently demonstrated high-
accuracy regression models. Also, a leave-one-variety-out cross-validation demonstrated robust performance, indicating that the regression model can be effectively applied to unseen tomato varieties.


When applying the regression model to the individual fruits along a single truss not included in the model, a noticeable, but expected, trend emerged (Figure 2): highest Brix values (most tasty) were observed in the top (first formed and ripest) fruits, and progressively decreasing values (less tasty) towards the lowest (last formed and less ripe) fruits. Considering that consumers will eat all fruits along a truss, a small Brix gradient is preferred by the industry, and with the setup used in this study, that information can be quickly assessed within seconds.
Applying hyperspectral imaging to entire trusses significantly improves the phenotyping workflow compared to traditional methods involving a fruit to fruit approach. The ability to quickly estimate Brix levels and truss gradients classifies hyperspectral
imaging as a reliable and efficient tool for high throughput tomato phenotyping as required for selection and advancement decisions during development of new varieties, and quality assessments within the commercial pipelines.
