Biomedical Engineering Reference
In-Depth Information
where
E
j
is the excitation of
j-
neuron of the
R
-layer,
a
i
is the excitation of
i-
neuron
of the
A
-layer, and
w
ij
is the connection weight between the
A
-layer neuron
i
and the
R
-layer neuron
j
.
In the neural classifier, the neuron from the
R
-layer having the highest excitation
determines the class under recognition. This rule is used always in the process of
recognition, but in training, it should be changed. Let the neuron-winner have
excitation
E
w
and its nearest competitor have excitation
E
c
.If
ð
E
w
E
c
Þ=
E
w
<
T
E
(11.2)
the competitor is considered the winner. Here,
T
E
is the superfluous excitation of the
neuron-winner.
Distinct from the Rosenblatt perceptron, LIRA neural classifier has only positive
connections between the
A
-layer and the
R
-layer, so the training rule is as follows:
1. Let
j
correspond to the correct class under recognition. During recognition, we
obtain excitations of
R
-layer neurons. The excitation of neuron
R
j
corresponding
to the correct class is decreased by the factor (1 -
T
E
). After this, the neuron
having maximum excitation
R
k
is selected as the winner.
2. If
j
=
k
, nothing must be done.
3. If
k
,
W
ij
(
t
+1)=
W
ij
(
t
)+
a
i
, where
W
ij
(
t
) is the weight of the connection
between the
i
-neuron of the
A
-layer and the
j
-neuron of the
R
-layer before
reinforcement,
W
ij
(
t
+ 1) is the weight after reinforcement, and
a
i
is the output
signal (0 or 1) of the
i
-neuron of the
A
-layer.
j
6¼
W
ik
t
ð
þ
1
Þ ¼
W
ik
t
ðÞ
a
i
;
if
W
ik
t
ð
ðÞ >
0
Þ;
-
W
ik
t
ð
þ
1
Þ ¼
W
ik
t
ðÞ;
if
W
ik
t
ð
ðÞ¼
0
Þ;
where
W
ik
(
t
) is the weight of the connection between the
i
-neuron of the
A
-layer and
the
k
-neuron of the
R
-layer before reinforcement, and
W
ik
(
t
+ 1) is the weight after
reinforcement. According to this rule, connection weights have only positive
values.
We adapt the classifier LIRA for pattern recognition in assembly of microde-
vices. The experiments were made with the pin, ring, and hole having the following
parameters:
- diameter of the pin: 1 mm;
- outer diameter of the ring: 1.2 mm;
- inner diameter of the hole: 1.25 mm.
11.3 Neural Interpolator for Pin-Hole Position Detection
The neural interpolator [
6
,
8
,
9
] differs from the neural classifier because the
excitations of the output neurons are considered as a set of values of continuous
functions
f
(
dx
) and
'
(
dy
). To determine the functions
f
(
dx
) and
'
(
dy
), we use a
