Revolutionary Non-Contact Prediction of Blueberry Brix with Hyperspectral Imaging

Blueberries are prized for their flavor diversity, with early-season varieties often being sweeter for snacking and later varieties ideal for baking due to their tartness.

Traditionally, blueberry inspection involves manual labor to detect defects, ripeness, and bruising, and sweetness is usually measured through destructive methods. Headwall’s hyperspectral imaging (HSI) technology introduces a breakthrough in blueberry assessment: a non-destructive, real-time, and rapid way to evaluate large quantities of berries for both quality and Brix (sweetness) levels. This technology allows for instant, contactless grading, enhancing the value chain by providing accurate data for quality assessment.

Efficient Sorting of Blueberries with VNIR Technology

Using Headwall’s perClass Mira Stage and MV.C VNIR imaging system, growers can distinguish between high- and low-quality blueberries by detecting ripeness, bruising, and defects. The blueberries are scanned, and the PerClass Mira Software builds a classification model, differentiating quality grades as determined by growers. Testing of this model demonstrated a high accuracy rate, correctly identifying 24 out of 25 good berries and 23 out of 24 bad ones. VNIR hyperspectral imaging thus enables effective sorting of blueberries by quality, supporting large-scale, efficient sorting.

Efficient Sorting of Blueberries with VNIR Technology

blueberries-scanned-headwall-mv.c-nir-perclass-mira.png

Predicting Blueberry Brix with NIR Regression

In a further study, Headwall’s MV.C NIR sensor with the PerClass Mira Stage and Software predicted the sweetness (Brix) of blueberries from various harvest times across the growing season. Hyperspectral images were captured, and Brix levels were measured with a digital refractometer to train a regression model that accurately predicts blueberry sweetness. Repeat scans confirmed the model’s robustness, and cross-validation across early and late season blueberries achieved high predictive accuracy, as shown by a strong correlation between predicted and measured Brix.

The model also categorized each pixel as either “good berry,” “bad berry,” or “background,” grouping at least 1,000 contiguous berry pixels to identify distinct berry objects. Berries were then classified by the proportion of good or bad pixels, achieving a 92% success rate in accurately grading berries.

Advancing Blueberry Grading with Hyperspectral Imaging

Headwall’s advanced hyperspectral imaging solutions, combined with PerClass Mira Stage and Software, enable users to quickly build and refine classification and regression models for real-time actionable insights. These models are adaptable for use on quality control test stands, inline at packing facilities, or in laboratory environments, empowering producers to optimize grading and assign value based on precise quality metrics.

Photo: An example of cross-validation results from a full season model. Berries from all lots were used in training (shown in blue diamonds). The model proved strong in its predictive power with a regression power of 0.882 for the training set and an RPD of 2.08.

More information?

Contact our specialist in spectroscopy Kees van der Sar.

More information about the MV.C NIR® and PerClass Mira® Scanning Stage.

Contact Inventech today

Revolutionary Non-Contact Prediction of Blueberry Brix with Hyperspectral Imaging

Inventech is the specialist in real time measurements with NIR, VNIR and SWIR.