Data Analysis: Wavelength Selection Methods

1
Data Analysis Wavelength Selection Methods

• Yang, Chao-Lin
• 20021225

2
Introduction

• There is an almost linear relation between the
  concentration of an absorbing component and its
  contribution to the log(1/R) value at the
  wavelength absorbed.

3
Introduction

• Beer/Lambert law

4
(No Transcript)
5
(No Transcript)
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
Developing a model

• Measurement
• Calibration
• Validation

10
(No Transcript)
11
Developing a model

• Measurement
• Quantification
• Remark
• Determined quantitative
• Detected qualitative

12
Developing a model

• Calibration
• Single-term regression

13
(No Transcript)
14
Developing a model

• Calibration
• Multiterm linear regression

15
Developing a model

• Validation
• To compare NIR measurements with reference method
  measurements on a new set of samples

16
Developing a model

• Step
• To record the spectra and reference data
• To determine the errors
• To choose data treatments
• To validate the acceptable calibration
• To understand why it works

17
(No Transcript)
18
(No Transcript)
19
Sources of error

• Sampling Error
• Reference Method Error
• NIR Method Error

20
Sources of error

• Sampling Error
• Lack of homogeneity in the material being sampled

21
Sources of error

• Sampling Error (conti.)
• reload (repack)
• rotate

22
(No Transcript)
23
(No Transcript)
24
(No Transcript)
25
Sources of error

• Reference Method Error
• Laboratory error, reproducibility
• The standard deviation of differences between
  measurements on duplication samples

26
Sources of error

• NIR Method Error
• Spectral measurement errors, lack of intrinsic
  correlation between spectral and reference
  methods data, and poor choice of data treatment

27
(No Transcript)
28
Sources of error

• NIR Method Error (conti.)
• Running mean (boxcar smooth)
• Stavitzky-Golay smooth

29
(No Transcript)
30
Single-term Linear Regression
31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34
(No Transcript)
35
(No Transcript)
36
(No Transcript)
37
(No Transcript)
38
(No Transcript)
39
(No Transcript)
40
Multiterm Linear Regression
41
(No Transcript)
42
(No Transcript)
43
(No Transcript)
44
(No Transcript)
45
Mathematical treatment

• Derivative

46
(No Transcript)
47
Mathematical treatment

• Derivative (conti.)
• Resolution of overlapping peaks

48
(No Transcript)
49
Mathematical treatment

• Derivative (conti.)
• Removal of linear baselines

50
(No Transcript)
51
(No Transcript)
52
(No Transcript)
53
(No Transcript)
54
(No Transcript)
55
Mathematical treatment

• Fourier Transform

56
(No Transcript)
57
(No Transcript)
58
(No Transcript)
59
(No Transcript)
60
(No Transcript)
61
(No Transcript)
62
(No Transcript)
63
Mathematical treatment

• Other methods
• Component Spectrum Reconstruction
• Fast Correlation Transform
• Normalizing Spectra
• Discriminant Analysis
• Multiplicative Scatter Correlation

64
(No Transcript)
65
(No Transcript)
66
Thanks for Your Attention!