Location: Sugarbeet and Bean Research2012 Annual Report
1a. Objectives (from AD-416):
1) Evaluate NIR spectra in diffuse reflectance, transmittance, and scattering modes for determination of potato density/specific gravity. 2) Evaluate NIR spectra in diffuse reflectance, transmittance, and scattering modes for determination of specific sugars levels in potatoes. 3) Determine optimal sensing configuration (mode and wavelengths) for moving toward development of a prototype device for potential real-time field-based measurement.
1b. Approach (from AD-416):
Potatoes covering a range of density and sugar levels will be obtained from the potato industry. Samples will be collected and evaluated at multiple intervals over the length of the storage season. Each cleaned and intact potato tuber will be measured with the laboratory instrumentation set-up under the following three modes: diffuse reflectance over the visible and NIR region from 500-1700 nm; transmittance over the range of 500-1300 nm; and scattering over the range of 500-1300 nm. The transmittance and scattering ranges do not extend as far into the NIR range due to absorption. Following whole (intact) tuber measurement, a tissue sample will be cut from the same tuber and the same spectroscopic measurements will be obtained on this controlled and uniform sized sample. Additionally, each potato will be measured for density and specific sugars using conventional hydrometer and wet chemistry techniques to provide a baseline to which the spectroscopic measurements will be compared and correlated. A local company, Techmark Inc., specializes in potato handling and analysis and is supportive and willing to assist with conventional analysis (as a no cost collaborator). While the goal is to successfully measure whole/intact tubers, additionally evaluating samples of tissue will provide the opportunity to compare the results of this study against published results and also tissue versus whole tuber measurement. Several potato varieties will be included in the study to determine robustness or specificity of findings. Data analysis will involve determining portions of the spectra and the mode, or combinations of spectra and modes, capable of best predicting density and sugars levels. The progression of the research would include the first year of broad spectral measurement of whole tubers and tissue samples and critical analysis of data. A second year would be important (similar budget) to validate first year findings and, in parallel, focus on a particular sensing configuration based on what we learned from year 1. This sensing configuration would be a step toward a prototype device incorporating dedicated electronics for real-time in-field measurements.
3. Progress Report:
Previous reports of this project document the progress of promising response in the first year of study of correlating electronic measurements of visible and NIR transmittance, interactance, and hyperspectral imaging to wet chemistry measurements of sucrose and glucose sugars. These sugars are important for the potato industry to monitor both during storage and at harvest time to determine the physiological status of the potato and thus appropriately handle and manage the crop for optimal profitability. The second year of the study attempted to validate the promise and demonstrate robustness of the approach by repeating the measurements with a second season of samples having an improved range and more even distribution of samples. This proved challenging and the results were non-confirming, however, significant data analysis was able to demonstrate some high levels of correlation, especially for the Russet variety which has higher levels of sugars and more so for glucose than sucrose in both varieties. This past year the project involved a third season of sample collection with again the effort to achieve a more uniform distribution of sugar levels across the sample set. Samples were collected from different fields, and locations within the field, and the number of storage temperatures was increased and broadened. Seven laboratory evaluation/measurement dates from October to May were included. More preprocessing methods were used on the data to help reduce signal noise and increase the performance of calibration and consequently prediction models. The focus in constructing the prediction models is built upon three main methods: partial least squares regression (PLSR), radial basis functions neural networks (RBFNN), and feed forward neural networks (FFNN). Data fusion between system variables (measurement modes) is being applied as well as applying calibration models for one season on the data for another season. For PLSR results, the Russet Norkotah variety continues to yield better correlation with glucose and sucrose than the Frito Lay (chipping) variety. Some correlations as high as 0.95 have been achieved for prediction of a single constituent for a single variety. Analysis of the most recent samples is ongoing for both RBFNN and FFNN to determine if improvement of performance of prediction models for both varieties can be achieved. Because a single electronic mode and analysis technique does not, at this point, appear to best predict all constituents over all varieties, it has been a challenge to focus on the final objective of the development of a dedicated low-cost rapid evaluation instrument for field use measurement of potato constituents. The project is beginning to organize and publish the findings.