IEEE - Aerospace and Electronic Systems - February 2023 - 14

Anti-Interference Recognition for Aerial Infrared Object Based on Convolutional Feature Inference Network
algorithm. The specific test conditions are listed in Table 2.
There are 24465 point source interferences. The recognition
rate is defined as follows:
Pright ¼
Nright
Ntotal þNwrong
100%
(18)
where Nright represents the number of correctly recognized
objects; Ntotal and Nwrong represent the total number
of objects and the number of interferences that are
misjudged as objects, respectively.
EXPERIMENTAL RESULTS AND EVALUATION ANALYSIS
After preprocessing and feature extraction are performed
on the test set sequence, the classifier is used to recognize
aircraft objects in various situations for anti-interference.
Some of the results are shown in Figure 9 (the object is
marked with a red rectangular box, and the interference is
marked with a green rectangular box).
The test results show that, under the above test conditions,
the anti-interference recognition algorithm based on
the 2DPCANet-NB classifier can resist point-source artificial
decoys in the whole process. In this experiment,
23,589 objects (including 584 false alarms) are recognized
for the selected 24,465 test images. Hence, the object recognition
rate is 94.17%.
The comparison of the recognition rates of the five
algorithms listed in Table 3 shows that the overall recognition
rate of the anti-interference recognition algorithm
based on the 2DPCANet-NB classifier achieves a higher
recognition rate than the algorithms based on the NBbased
classifier, the SVM-based classifier, the multifeature
fusion, and the TAN-based classifier.
The entry angle of the missile and the maneuvering
mode of the object both have a certain impact on the performance
of the proposed algorithm. In terms of pointsource
interference, for an entry angle between 10 and
60, the missile is in a tail chase state. The object tail has
obvious characteristics at a long distance and clear features
at a medium and short distance. Hence, the overall
recognition rate is high. Besides, for an entry angle
between 60 and 100 and a lateral trajectory, the shape of
the long-range object is not obvious, resulting in a low
object recognition rate in the first dozen frames of the
sequence. However, the maneuvering of the object in a
short distance can cause tailing interference and approximate
the shape of the object without maneuvering, which
makes a high false alarm rate at this stage. For an entry
angle between 100 and 160, the missile and the object
are in an attacking state. The object characteristics are not
obvious in the long distance. In the medium and short distance,
the distance between the missile and the object
decreases faster, and the characteristics change obviously.
14
In this case, the overall recognition rate is high. Especially,
for an entry angle of 100, the object makes a jump
maneuver, and the interference is 2 decoys per group. In
this case, the short-range interference presents a form similar
to the object, as the missile approaches the object. The
feature descriptor extracted by the algorithm proposed in
this article shows strong adaptability, which can well capture
the difference between the object and the interference
at the feature level. Also, compared with other algorithms,
it achieves a greatly improved recognition rate.
CONCLUSION
Aiming at the problem of anti-interference object recognition
faced by infrared imaging-guided air-to-air missiles,
this article proposes an object recognition algorithm based
on a convolutional feature inference network to recognize
aerial infrared objects. In the simulation performed on the
air combat interference confrontation image dataset, the
recognition rate of the proposed anti-interference recognition
algorithm reaches 94.17%, which is 3.11% higher
than the algorithm based on the TAN classifier. To a certain
extent, the proposed algorithm can solve the problems
such as false objects and object occlusion that the TAN
classifier-based statistical pattern recognition method cannot
cope with. Besides, the proposed algorithm has certain
feasibility and effectiveness. In future work, the sample
database needs to be expanded to obtain more accurate
prior information. Meanwhile, more features especially
shape features and invariant features and the relationship
between the features need to be mined to obtain a more
comprehensive and stable depiction of the objects and
interference.
REFERENCES
[1] G.Marsiglia,L.Fortunato,A.Ondini, andG.Balzarotti,
" Templatematching techniques for automatic IR target recognition
in real and simulated scenarios: Tests and evaluations, "
Autom. Target Recognit. XIII, vol. 5094, pp. 159-169, 2003.
[2] A. C. Berg, T. L. Berg, and J. Malik, " Shape matching and
object recognition using low distortion correspondences, "
in Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern
Recognit., 2005, vol 1, pp. 26-33.
[3] S. Ullman, " Object recognition and segmentation by a
fragment-based hierarchy, " Trends Cogn. Sci., vol. 11,
no. 2, pp. 58-64, 2007.
[4] J. Yang, D. Zhang, A. F. Frangi, and J. Y. Yang, " Twodimensional
PCA: A new approach to appearance-based
face representation and recognition, " IEEE Trans. Pattern
Anal. Mach. Intell., vol. 26, no. 1, pp. 131-137, Jan. 2004.
[5] C. Cortes and V. Vapnik, " Support-vector networks, "
Mach. Learn., vol. 20, no. 3, pp. 273-297, 1995.
IEEE A&E SYSTEMS MAGAZINE
FEBRUARY 2023

IEEE - Aerospace and Electronic Systems - February 2023

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