IEEE - Aerospace and Electronic Systems - February 2023 - 11

Yang et al.
Table 2.
Parameters for Testing Algorithms
Parameters
Launch distance
Entry angle
Total number of
missile
Type of
maneuvering
Group interval (s)
Number of
bombing groups
Number of
missiles per group
Missile interval (s)
Similarly, the following calculation can be performed:
P f C ¼ 0 ¼ðÞj
Value
7000 m
Within 360, select
trajectory from	10 and
30
48
No maneuvering, Left
maneuvering, Jump
0.4, 0.7
12, 24
2, 4
0.1
Y16
i¼1
Pgi PaðgiÞ
¼ Pðg1 ¼ e1;C ¼ 0Þ Pðg2 ¼ e2 g1
pðg3 ¼ e3 g2
;C ¼ 0Þ
;C ¼ 0Þ pðg14 ¼ e14 g15
pðg15 ¼ e15 X13j ;C ¼ 0Þ pðe16 ¼ e16 g12
;C ¼ 0Þ
;C ¼ 0Þ: (16)
By comparing PðfjC ¼ 1ÞPðC ¼ 1Þ and PðfjC ¼
0:ÞPðC ¼ 0Þ, the result of the probabilistic inference is
obtained. If
P f C ¼ 1 PðC ¼ 1Þ >P f C ¼ 0 PðC ¼ 0Þ (17)
ðÞj
ðÞj
the candidate region belongs to the object region; otherwise,
it belongs to the artificial interference region.
ALGORITHM STEPS
C ¼ arg max
C
C
PðCjg1;g2; ... ;g16Þ
¼ arg max a PðCÞ
Y16
i¼1
PðgijPaðgiÞÞ
(14)
where a is a constant irrelevant to the determined
class, and C is class variable with a value of either 1
or 0, representing object and interference, respectively.
For a set of feature descriptors f ¼½e1;e2; ... ;e16
T
Algorithm 1. Our proposed object recognition
algorithm
Require: test image x;
Ensure: recognition result;
of
the candidate region, PðfjC ¼ 1ÞPðC ¼ 1Þ and
PðfjC ¼ 0ÞPðC ¼ 0Þ are calculated respectively based
on the network parameters obtained in the previous
section. The prior probability PðCÞ of each class is
obtained by counting the training samples. According
to the algorithm in this section, the network structure
includes
P f C ¼ 1 ¼ðÞj
Y16
i¼1
pðg3 ¼ e3 g2
pðg5 ¼ e5 g9
pðg7 ¼ e7 g6
pðg9 ¼ e9 g11
pðg11 ¼ e11 g10
pðg13 ¼ e13 g14
Pgi PaðgiÞ
¼ Pðg1 ¼ e1;C ¼ 1Þ Pðg2 ¼ e2 g1
;C ¼ 1Þ pðg4 ¼ e4 g3
;C ¼ 1Þ pðg6 ¼ e6 g8
;C ¼ 1Þ pðg8 ¼ e8 g4
;C ¼ 1Þ pðg10 ¼ e10 g7
;C ¼ 1Þ pðg12 ¼ e12 g8
;C ¼ 1Þ pðg14 ¼ e14 g15
;C ¼ 1Þ
;C ¼ 1Þ
;C ¼ 1Þ
;C ¼ 1Þ
;C ¼ 1Þ
;C ¼ 1Þ
;C ¼ 1Þ
pðg15 ¼ e15 X13j ;C ¼ 1Þ pðe16 ¼ e16 g12
FEBRUARY 2023
;C ¼ 1Þ: (15)
1: x image preprocessing(x)
2:
cn Conv(x)
3:
4:
cn MaxPooling(cn)
c MixPooling(cn)
5: P(object, Interference) Table(c)
6:
7: x is object
8:
else
9: x is interference
if P(object) > P(Interference) then
We use the pretraining model trained by large-scale datasets
to better represent the difference between target and
interference. Bayesian network must design network
structure and connection form according to feature
vector extracted by 2DPCANet, and design conditional
probability table. The recognition steps of the proposed
algorithm are as given in Algorithm 1.
EXPERIMENTAL RESULT ANALYSIS
EXPERIMENTAL CONDITIONS
We describe the battlefield situational confrontation simulation
experiment platform of the National Natural Science
Foundation of China; this platform generates
infrared air combat simulated image datasets for algorithm
testing, in which infrared band, frame frequency, and resolution
ratio are mid-wave (3 5mm), 100 Hz, and 256
256, respectively. Open Scene Graph (OSG) rendering
IEEE A&E SYSTEMS MAGAZINE
11
IEEE - Aerospace and Electronic Systems - February 2023

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - February 2023
