How machine learning enables non-contact moisture monitoring in rockwool substrates
Source: VFD.com
Widely utilized growing media such as rockwool remain spectrally under-characterized, while conventional sensing approaches are frequently contact-based or restricted to laboratory conditions. To address these operational constraints, an engineering research team has introduced an integrated laboratory-to-deployment framework designed for the non-contact estimation of volumetric water content in horticultural rockwool cubes.
The experimental investigation initiated moisture characterization by acquiring high-resolution ultraviolet, visible, and near-infrared reflectance spectroscopy spanning wavelengths from 200 nm to 2500 nm. These high-resolution spectral readings were gathered across eleven distinct volumetric water content levels to establish the baseline optical response of the fibrous substrate.
To classify the full-spectrum optical data across moisture variations, the investigators developed and trained a one-dimensional convolutional neural network. In comparative laboratory evaluations, this multi-learner framework improved the baseline class-average F1-score by 20.2 percentage points.
Why this matters: For operators, this is a water-management story. The useful signal is that direct substrate measurements can help cut drain loss materially without giving up yield or fruit quality, which is exactly the kind of controllable efficiency gain a facility can build on.
Frequently Asked Questions
Why does substrate sensing matter in free-drain strawberry systems?
Because drain percentage tells a grower what already happened, while substrate moisture and EC data show root-zone conditions directly. That makes it easier to cut water loss without guessing.
What is the operator takeaway from this trial?
If the thresholds are understood well enough, growers can reduce drain water materially while protecting yield and fruit quality, which makes sensing an operational tool instead of a reporting tool.