
420 Mathematical Tools
A.4 Least squares procedures
In this section, ordinary linear regression is extended to a recursive proce-
dure for sequential data. This forms the basis of one of the neural network
training algorithms derived in Appendix B. In addition, the orthogonal lin-
ear regression procedure used in Chapter 9 for radiometric normalization is
explained.
A.4.1 Recursive linear regression
Consider the statistical model given by Equation (2.92), now in a slight
ly
different notation:
Y (j) =
N
X
i=0
w
j
x
i
(j) + R(j), j = 1 . . . ν. (A.1)
This model r elates the independent variables x(j) = (1, x
1
(j) . . . x
N
(j))
⊤
to
a measured quantity Y (j) via the