Aerospace and Electronic Systems - March 2019 - 20

Robust Cooperative Target Detection for a Vision-Based UAVs Autonomous Aerial Refueling Platform via the Contrast . . .

Figure 2.
Schematic drawing illustrating the cooperative target detection method based on the CSF of eagle's eye.

frequency. Here, a two-dimensional (2-D) exponential
function is established to express the relationship between
the center and surround:
SFðx; yÞ ¼ fðcenterÞ À fðsurroundÞ
¼ Kc expðÀapðx2 þ y2 ÞÞ À Ks expðÀbpðx2 þ y2 ÞÞ

(1)
where the four constants are set as Kc ¼ 16:99,
Ks ¼ 33:73, a ¼ 0.02630, and b ¼ 0.1076, respectively,
in this paper. This function can well fit to the whole
eagle's CSF properties mentioned above.
The neurons located in the primary visual cortex can
efficiently dispose the basic features of image, including
local edge and orientation [33]. Various filters such as Gabor
filter or canny detector simulate this property of neurons,
however, they lack inhibitive effect. When the stimulus
from the classical receptive fields activate the neurons, the
inhibitions effect from the nonclassical receptive fields will
suppress the neurons. The CSF possess many similar properties of surround suppression and the attention effect.
We apply it to extract the contour of the cooperative target
from disorderly background textures.
Our proposed method is able to produce a strong
response to the salient contours and restrain the texture
information. The proposed method consists of five layers,
as shown in Figure 2. The edge responses with different
scales and orientations are firstly obtained by the orientation-selective neurons layer (OSNL). Then the denoising
layer (DL), contrast sensitivity based surround inhibition
layer (CSSIL) and multiresolution fusion layer (MFL) are
conducted to the input image. Finally, the cooperative
target is determined with the target selection recognition
layer (TSRL).

where X ¼ x cos u þ y sin u, Y ¼ x sin u þ y cos u, the
other related parameters are given in Table 1. In this
paper, the ellipticity of the receptive field g and the spatial
frequency bandwidth s= are constants and set as 0.5 and
0.56, respectively. The orientation and frequency selectivity of the simple cell receptive fields is modeled using the
Gabor functions [35], and the convolving operator is also
utilized to calculate the neuron response Rs;u;' ðx; yÞ
of simple cell with the input image stimulus Iðx; yÞ:
Rs;u;' ðx; yÞ ¼ ðI Ã Gs;u;' Þðx; yÞ

At the same time the energy filter function is used to
establish the model of the spatial property of complex cell
[36]. When orientation u and the scale s is chosen the
energy value Es;u ðx; yÞ of each pixel is defined by:
Es;u ðx; yÞ ¼

qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
R2 s;u;0 ðx; yÞ þ R2 s;u;p=2 ðx; yÞ

(4)

For several different orientations ui , MAX-like operation is utilized to construct the energy maps for complex
cells [37]. Here, we selected four scales and eight orientations for the construction of the energy maps:
Es ðx; yÞ ¼ maxfEs;ui ðx; yÞji ¼ 1; 2; . . . ; 8g

(5)

where the orientation ui ¼ ði À 1Þp=8; ðiÀ ¼ 1; 2; . . . ; 8Þ
and the scale s will be set according to the actual situation.
The preferred orientation map is determined by the corresponding maximum response.

Table 1.

Parameters for the Cooperative Target Detection Method

ORIENTATION-SELECTIVE NEURONS LAYER
In the neurophysiology, visual neurons can respond to an
edge of a certain orientation. Two types of orientation-selective neurons, located in visual cortex, have been found. One
is called simple cells and the other is called complex cells.
Simple cells appear to very sensitive to the contrast polarity,
while complex cells are formed by simple cells.
A 2-D Gabor function is utilized to establish the
model of the receptive field's spatial property of a
simple cell [34]:




X2 þ g 2 Y 2
2p
Gðx; yÞ ¼ exp À
X
þ
'
Â
cos
2s 2


20

(3)

(2)

Parameter

Description

Value

g

The ellipticity of the
receptive field

0.5

u

The preferred
orientation

0 -p

s

The size of the
receptive field

1-20



The wavelength

k

The inhibition factor

'

The phase offset

IEEE A&E SYSTEMS MAGAZINE

s=0:56
1.5
5
MARCH 2019
Aerospace and Electronic Systems - March 2019

Table of Contents for the Digital Edition of Aerospace and Electronic Systems - March 2019
