Page 71 - RAC_CIAW_ a_I_n_01_2021.pdf

processing to find small parts of an image that cor- the template matching mechanism for cyber-electronic
respond to a model (template) image. To do so, it is attack was implemented and simulated in Python.
defined a template to be searched in a main image. The It is assumed that the cyber component of the at-
main image in analysis and the template are divided in tack (the malware) is already installed in the radar,
pixels, as shown in Figure 2. Then the template is moved given that the exploitation mechanisms to install it in
over the main image, in a search process throughout the radar computational system is out of the scope of
all the main image’s area. For each position assumed this paper. Also, considering that the implementation
by the template in this scanning process over the main of the EA component of the attack is not in the scope
image, a similarity index is computed. The similarity of this paper, it is assumed that the remote command (a
index quantifies the similitude between the template sequence of false echoes) is generated and transmitted
and the piece of the main image being compared. If through a Digital Radio Frequency Memory (DRFM)
the index is higher than a previously defined threshold, technique [11]. The command is received and pro-
then the template image is considered to be detected cessed by the radar, and displayed as an image in the
in the main image. This exhaustive search operation PPI screen, such as any other received echoes (false or
demands a considerable computational cost, propor- not).
tional to the sizes of the images. On the other hand, it In the simulations herein presented, the Python
provides a high degree of effectiveness in searching for code scans the graphical interface (the PPI) produced
patterns in images [14].
by the radar simulator in order to identify the attack
Note that the degree of similarity between the commands received from the EA component of the of-
template and a piece of the main image is established fensive. To evaluate the effectiveness of the proposed
by comparing intensity values of each of their pixels. mechanism, in this work, the chosen attack command
Among the available methods to compute the similar- consists of a sequence of five false echoes, which pro-
ity coefficient, there are: the Sum of Absolute Differ- duces a sequence of five points displayed in the direc-
ences (SAD), the Sum of Squared Differences (SSD), tion where the DRFM transmitter is. Once this pattern
and normalized cross correlation. In this paper, the is detected, it can be used to trigger a malicious action
Pearson cross correlation (PCC) [14] is used (1): in the naval radar system.

To validate this hypothesis and test the effective-
ness of the command detection method, 30 fictitious
scenarios were generated using the radar simulation
software in order to represent real situations where
a naval platform could be. Clutter/target echoes that
wherein, pi is the intensity of pixel i in the tem- might affect the command detection were randomly
plate; p is the average intensity of the pixels of the tem- inserted in the PPI. Figure 3 shows an example of sce-
plate; ai is the intensity of the pixel i in the patch of nario used in the simulations. The triggering command
the image; a is the average intensity of the pixels in the is highlighted.
patch of the image; N is the number of pixels. Note It is worth mentioning that the attacker is trans-
that this method presents a normalizing term in the de- mitting the EA signal that generates triggering com-
nominator, which gives it invariance to global changes mand shown in the screen, and that the signal can
in brightness [14], and the results always lie within a come from any direction, depending on the DRFM
defined range [−1, 1]. transmitter location. Thus, it is necessary to consider
different angles from which the triggering command
RESULTS could be received. For the sake of simplicity, varia-
tions of 1 degree are considered, so the attacker could
The mechanism for cyber-electronic attack de-
scribed in Section III was evaluated through simula- emit from the directions 000, 001, 002, 003 and so
tions on a computer with an Intel i7 processor of 2.5 on. Considering these possible different Angles Of Ar-
Ghz, 8G RAM DDR3 memory, running Microsoft rival (AOA), the template containing the triggering
Windows 10, 64 bits. The radar environment was command pattern is also rotated in steps of 1 degree
simulated in the Cinematic Radar Simulator v.2.0 and during the search process throughout the PPI. This
template matching search is executed throughout all

CIAW – EFICIÊNCIA, CULTURA E TRADIÇÃO 71