High-tech and Artificial Intelligence Make the Invisible Visible

Research and Development in Biosystems Engineering

Quantification of Crop Constituents

Non-invasive quantification of crop constituents based on optical methods has great research and business potential. Such systems are used for plant phenotyping, e.g. in high throughput screening of large mutant collections, or in precision farming.

Unlike commercially available systems that typically detect and analyze only a few wavelengths characteristic for very specific constituents (e.g. nitrogen), the advantage of the hyperspectral approach favored by the researchers of the Biosystems Engineering Expert Group is its evaluation of the spectral fingerprint throughout a broad range of the spectrum as a complete pattern. Any number of constituents can be quantified simultaneously.

The key challenge in this approach is analyzing extremely high-dimensional and complex spectra so that quantitative information on constituents relevant to the user can be extracted from them. To do so, we employ methods of machine learning developed specifically for this application. Using appropriate biochemical reference measurements, they can be properly trained and thus adapted to a user’s concrete needs.

Schematic teaching phase (top) and operational phase (left).
A few reference samples teach the mathematical model that assigns spectral data to relevant constituents. In the operational phase, the trained knowledge is applied to the spectral data processed. Relevant constituents are thus quantified without having to perform complex laboratory analysis. This requires more than standard machine learning architectures, however.