Janel Kane reports on a novel deployment of miniature spectrometers for grain protein concentration sensing
Grain protein concentration (GPC) is a critical measure in grain production, with commodity pricing based on protein content. Protein content is also directly proportional to nitrogen uptake. The measurement of protein content and nitrogen uptake allows the producer to manage application of nitrogen fertilisers to reduce over-use and operate more economically while maximising production yield and return. Near-infrared absorbance spectroscopy is a standard and nondestructive analysis technique that has been used to measure the GPC in wheat and other grains. Since 2000, several systems have been developed to deploy this technology in the field, on-combine, but these systems are too expensive for many small producers and have not been widely adopted for that reason. Researchers at the United States Department of Agriculture (USDA) Oregon Field Offices endeavoured to develop a low cost near infrared analyser for in field, real-time grain analysis.
Miniature spectrometer system description
The instrument adapted for the USDA study was the Avantes AvaSpec ULS2048x16-USB2 spectrometer (next generation instrument is the AvaSpec-ULS2048X64-Evo) with a 2,048 pixel charged coupled device (CCD) linear array detector and configured with a 600 lines mm-1 grating to measure the range 650-1,150 nanometres. The spectrometer is fibre coupled to a fibre optic cable that connects to an integrating sphere (AvaSphere 50-LS-HAL) with built-in halogen lamp. Light is collected through an aluminosilicate glass covered aperture for a sealed, light transmitting collection window. Thermal stabilisation of the system was achieved with the help of a thermoelectric cold plate and a styrofoam insulated enclosure. An inexpensive notebook computer, off-the-shelf GPS and programmable controller completed the equipment list for a total cost of less than US$9,000, putting this system in reach of a wider segment of small producers.
Designing the system began with lab testing, but it was also important to mimic field conditions. A test stand was used to simulate grain flow conditions on an auger conveyor system like that on a combine harvester. The probe head was secured to an opening in the auger housing at 30° above the centre of the grain stream at a flow rate of 1.63 kg/s-1. This allowed further testing of the calibration model under simulated harvest conditions.
Field evaluation of the adapted system was conducted with the integrating sphere head mounted to the housing of the exit auger of a Case International Harvester 1470 combine. Steel mesh screens were used to protect the fan ports on the sphere from debris and fibre optics from the sphere led to instrumentation mounted in the cab. Dark and light referencing was performed each time the combine was stopped for unloading the bulk collection tank, multiple times per day. Data was collected on the notebook computer during operation with processing at the end of each harvest day. The ruggedised Avantes system withstood the rigors of four growing seasons and saw about 200 hours of run time.
Development of the chemometric model
The USDA pursued a research product aimed at adapting relatively inexpensive NIR reflectance spectrometers to map GPC across a field. Development of the chemometric calibration model began with a three-year field study to obtain grain samples of soft white wheat grown under a variety of nitrogen and water level conditions. Custom code interfaces to import Avantes AvaSoft data into DeLight spectroscopic software where a process called binning grouped spectra into 126 bands of 16 pixels, a smoothing technique that simplifies a set of variables for modelling. Binning increases the signal to noise ratio while sacrificing resolution, in this case the resolution was reduced to 11.2nm, which kept resolution within acceptable limits. The application of the second derivative removed baseline shifts caused by the random orientation of grains. And finally, DeLight used partial least squares to find wavelength regions of interest where the grain sample was the independent variable and GPC as the dependent variable and derive a calibration model for grain protein concentration.
Miniature spectrometer system results
The resulting maps of GPC and grain yield demonstrated the inverse relationship between yield and protein content indicative of moisture-limited environments. Low yield area corresponds to thin unproductive soils where increased nitrogen fertiliser has little benefit. The protein map derived from the USDA test system was comparable to one created using a much more expensive Polytec spectrometer system.
Predictive models from local samples are limited in data, but the use of machine learning or artificial neural networks can increase system accuracy, but even so, not many farmers have the luxury of time to learn to apply advanced chemometric analysis. Adoption of this technology could be dependent on the availability of commercial services for chemometric modelling.
The final result of this study was proof of the potential for moderately priced alternatives to costly (>US$20,000) combine-mounted spectral mapping systems.
Janel Kane is with Avantes