AMSD 2020
Journal of Physics: Conference Series
1791
(2021) 012099
IOP Publishing
doi:10.1088/1742-6596/1791/1/012099
3
(
)
( ) (
)
∑
=
−
−
−
=
=
N
i
i
N
i
k
k
k
k
k
k
k
1
,...,
1
,
1
|
P
H
)
(
K
1
|
P
)
(
P
;
where
( )
k
z
i
– vector of observations;
( ) (
)
1
|
x
H
0
0
−
=
n
n
n
z
– vector of estimates of observations;
( )
n
0
x
– estimate
the state vector;
(
)
1
|
x
0
−
k
k
– estimate the state lead vector;
( )
k
Φ
– transition matrix;
( )
k
H
– observation matrix;
( )
k
i
K
– coefficient matrix;
(
)
1
|
P
−
k
k
– dispersion matrix of a state vector;
( )
k
P
– dispersion matrix of the state vector;
( )
k
U
– control vector;
( )
k
F
– vector of measured signals
from the object output;
( )
k
B
– matrix of control coefficients;
( )
k
D
–
matrix of measurement
coefficients;
( )
k
S
i
– meter sign;
( )
0
=
k
S
i
.
The lead value of the contour point is specified as
( ) ( ) ( ) ( )
1
-
xˆ
Ф
C
ˆ
k
k
k
k
y
=
.
The difference between the anticipated and actually observed points is set as
( ) ( ) ( )
k
k
k
e
yˆ
y
−
=
.
The Kalman filter gain is defined as
( ) (
) ( ) ( ) (
) ( )
( )
(
)
1
Q
C
1
P
C
C
1
P
K
−
+
−
×
−
=
k
k
k
k
k
k
k
M
T
T
.
The proposed image filtering mechanisms for solving
smoothing problems are considered in three-
dimensional space and implemented on the basis of the
cyclic multigrid method in a parallel computing
environment. The study was
carried out according to
the coefficient of gain in filtration, the values of which
are determined as the ratio
σ
σ
/
p
, where
p
σ
– the
filtering error on
σ
– the variance of the input process.
In figure 1. shows graphs of the gain in filtering
depending on the values of the parameter
B
P
G
σ
=
,
where
P
– the probability of relaxation of the reference
points of the image contour due to noise;
B
– the
range
of change in the coordinates of reference points on the
plane, usually specified by
10
=
B
. Charts 1- strategies of rule
σ
3
±
are illustrated; 2 - threshold
adaptive control; 3 – linear Kalman filter. Graphs
are obtained with
3
10
−
=
P
values;
( )
.
10
;
7
.
0
;
05
.
0
=
=
=
B
R
τ
σ
The efficiency of the mechanism is increased by using a differential operator based on adjusting the
orders of the approximation model depending on the mesh size. A constructive approach is proposed,
aimed at the recognition and classification of micro-objects with implementations of computational
schemes of a three-layer neural network (NN), learning algorithms, and retranslation of the process
dynamics to an image identification optimization model under conditions of nonstationarity and
parametric uncertainty.
Mechanisms for searching for correlations, tendencies, relationships, patterns in the dynamics of
images have been used and implemented. Tools for interpreting dependencies,
using templates, and
regulating variables have been obtained.
2.2.
Mechanisms for adaptive identification of micro-objects of various dimensions
The solution to the problem of identifying micro-objects is based on scaling tools, selection of
reference points of the contour,
threshold control, reduction of zero points, calculation of the root-
mean-square error from the corresponding functions of the input and reference images. Many micro-
objects have different dimensions, which negatively affect the quality of image identification and the
loss of useful information occurs, therefore, it is necessary to study the effect of the operation of
reducing the size of digital images to a certain threshold. A technique has been developed,
mechanisms obtained by which do not depend on the size of the image. And
it is aimed only at finding
the required function by its integrals in the family of lines on the basis of the Mellin transformation.
Do'stlaringiz bilan baham: