FACT: Predicting Wheat Hagberg Falling Number from Near Infrared Spectrometers

Our overallgoalis to improve overall end-use quality of wheat by isolating low falling number wheat and developing varieties with resistance to low falling number.Our five objectives are:Objective 1: Predict wheat falling number using near infrared spectroscopy on wheat meal.NIR spectrometers have been widely and successfully used to evaluate protein, oil, and moisture content. Although research has been conducted to predict FN from NIR on either wheat meal or kernels, the resulting prediction accuracy is still far from satisfactory. Unlike protein, oil, and moisture content, the causes of low FN are catalytic. A single enzyme can make thousands of cuts into starch and protein within seconds during dough making, so even a small amount of enzyme can cause significant damage. Detecting this enzyme in low amounts requires precise identification of the wavelength patterns related to the enzyme's molecular bonds, including N-H, C-H, and O-H. Our preliminary results showed the potential for improving prediction accuracy by increasing the training sample sizes using ANN. Our target is to limit prediction error within 40 seconds, compared to 30 seconds of FN standard deviation of technical replication.Objective 2: Predict wheat falling number using hyperspectral imaging of individual kernels.FN prediction using HSI of individual kernels not only provides a non-destructive assessment tool, but also a mechanism to improve prediction accuracy. Measuring the kernel surface provides direct access to the relevant area for enzyme activity, which is concentrated in the aleurone layer of the wheat kernel. Mixing with the inside starchy endosperm, which makes up over 80% of the wheat kernel, substantially dilutes the enzyme concentration, reducing FN prediction accuracy. Detecting and measuring only the surface components of the wheat kernel is achievable using HSI. Our target is to limit prediction error within 35 seconds, compared to 30 seconds of FN standard deviation of technical replication.Objective 3: Validate prediction accuracy on independent samples outside the Pacific Northwest.Low FN is a nationwide problem that degrades wheat quality and causes heavy financial losses to farmers, millers and bakers. As the prediction methods and tool will be developed by using samples from Pacific Northwest region, it is critical to validate these methods and tool to predict FN on independent samples outside the region.Objective 4: Develop and maintain an online computing tool to predict wheat falling number.The input of the computing tool is NIR and HSI data and the output is the prediction of FN. The calibration will be integrated to the tools and documented online. The tool can be directly used as a plugin for firmware update on devices such the detectors on combine harvesters and elevators.Objective 5: Stakeholder outreach for improving grower economics.Active engagement of wheat growers, breeders, millers, and bakers will allow them to incorporate our results as soon as they are available. For example, at the current 20% (R square) prediction accuracy using fast and nondestructive HSI screening, wheat breeders can efficiently eliminate the bottom 30% of early lines that have very little chance to get into the top 10% for field trials. Now breeders can eliminate them earlier and work on making more crosses.