Skip to main content

Table 2 Selection of best pre-treatment methods (combination of linearization, de-trending and multiplicative correction) from non-Gaussian information indicators and Bhattacharyya distance

From: Hyperspectral imaging: a novel approach for plant root phenotyping

 

Linearization

De-trending

Multiplicative correction

Non-Gaussian measure rank

Bhattacharyya distance R (rank)

Bhattacharyya distance Ri − Rj (rank)

Topsoil

1

Non*

Non

Non

15

0.28 (15)

9.40 (3)

2

Non

Polynomial

Non

16

0.12 (16)

9.94 (1)

3

Non

Derivative

Non

6

4.89 (2)

8.24 (4)

4

Non

ALS

Non

4

4.05 (5)

9.86 (2)

5

Non

Non

SNV

14

4.55 (4)

5.21 (12)

6

Non

Polynomial

SNV

11

9.05 (1)

5.52 (9)

7

Non

Derivative

SNV

10

1.78 (12)

3.81 (15)

8

Non

ALS

SNV

9

2.16 (11)

4.07 (14)

9

Log(1/R)

Non

Non

8

0.30 (14)

5.30 (11)

10

Log(1/R)

Polynomial

Non

7

1.00 (13)

5.46 (10)

11

Log(1/R)

Derivative

Non

1

2.57 (10)

4.94 (13)

12

Log(1/R)

ALS

Non

2

4.63 (3)

3.67 (16)

13

Log(1/R)

Non

SNV

12

3.08 (7)

6.77 (7)

14

Log(1/R)

Polynomial

SNV

13

3.65 (6)

6.46 (8)

15

Log(1/R)

Derivative

SNV

5

3.03 (8)

7.87 (5)

16

Log(1/R)

ALS

SNV

3

2.69 (9)

7.14 (6)

Subsoil

1

Non

Non

Non

13

0.26 (15)

3.18 (16)

2

Non

Polynomial

Non

14

0.11 (16)

3.48 (15)

3

Non

Derivative

Non

15

2.76 (12)

7.04 (11)

4

Non

ALS

Non

10

2.56 (13)

6.20 (12)

5

Non

Non

SNV

9

3.68 (9)

9.36 (6)

6

Non

Polynomial

SNV

7

3.87 (8)

9.88 (5)

7

Non

Derivative

SNV

16

3.89 (7)

4.52 (14)

8

Non

ALS

SNV

12

5.84 (6)

7.39 (10)

9

Log(1/R)

Non

Non

8

7.44 (2)

11.95 (3)

10

Log(1/R)

Polynomial

Non

11

1.10 (14)

10.61 (4)

11

Log(1/R)

Derivative

Non

5

3.61 (10)

8.87 (7)

12

Log(1/R)

ALS

Non

2

5.91 (5)

13.62 (1)

13

Log(1/R)

Non

SNV

3

6.60 (3)

8.19 (8)

14

Log(1/R)

Polynomial

SNV

4

6.37 (4)

7.56 (9)

15

Log(1/R)

Derivative

SNV

6

3.36 (11)

5.07 (13)

16

Log(1/R)

ALS

SNV

1

10.52 (1)

11.51 (2)

  1. For non-Gaussian measures the rank represents a mean of six different indicators; for ROIs (R, single wavelengths, Ri − Rj, wavelength differences) the absolute value of Bhattacharyya distance and the ranking of methods (in brackets) is given. The three best pre-treatment methods suggested by each selection approach are marked in italic
  2. * Non no treatment, Log(1/R) log-linearization, polynomial 2nd order polynomial centring, derivative 1st order Savitzky–Golay derivative, ALS asymptotic least square correction, SNV standard normal variate