F-750 Instruction Manual rev. 3/26/2014
1554 NE 3
rd
Ave, Camas, WA 98607, USA Phone (360) 833-8835 Fax (360) 833-1914
[email protected] http://www.felixinstruments.com
16
F-750 Vocabulary Definitions
Factor/Principle Component:
a vector that describes the information in the spectra (pixel
count/intensity gradient). The set of factors used to build regression coefficients must describe all the
selected spectra in the training set.
Feature/Spectra Selection Window:
where you select the region of wavelength to use for the PLS
regression.
K-1 Cross Validation or Leave One out Cross Validation:
the iterative method used to determine how
good the mathematical model generated by PLS is. This is done by measuring how well the
mathematical model can predict the samples used to make its self.
Mahalanobis Distance:
the perpendicular distance from a sample to an axis on a scores plot (traditional
visualization method). This measure can be used to detect outliers and is typically applied with
spectroscopy and biological applications. It can be normalized for mathematical models with more than
2 principle components.
Model Set:
the term used to describe the model file used by the 750. This file can consist of multiple
mathematical PLS models (i.e. Dry Matter, Total Soluble Solids, etc.) and/or traditional index band
(wavelength ratios) and/or the differential absorbance of chlorophyll. This will make up a set of
properties that will be predicted and displayed after a model file is loaded on the F-750.
NIPALS (Non-linear Iterative Partial Least Squares)
: iterative computer implementation of PLS.
Outlier:
a member of the training set that is not consistent with the others and thus introduces a large
error. For example, if building a model for predicting when to harvest apples from the orchard, the
training set should consists of samples taken prior to and at the time of harvest. If a rotten or usually
over-ripe apple was included in the sample, it would not be representative of the goal of the model and
introduce error as it would be inconsistent with the samples.
Partial Least Squares (PLS):
mathematical method used by the F-750 to generate a mathematical model
of regression coefficients. PLS1 only predicts a single trait with a single regression vector. PLS2 can be
used to predict multiple traits with a single regression vector. The F-750 will do PLS1 multiple times to
predict multiple traits.
Reference Method:
determines the known concentrations or value of constituent (i.e. trait, quality, or
property of interest). This could be the use of a refractometer for total soluble sugar or a scale to
determine dry matter content.
Regression Coefficients:
a series of weighted values generated by PLS. When these are applied to a scan
of an unknown sample, a prediction of concentration is made. These values are the core of the
mathematical model. PLS takes multiple scans and combines it with known values to generate a
Summary of Contents for F-750
Page 1: ...F 750 Produce Quality Meter Operation Manual...
Page 2: ......
