Target Planning for UAV Merchant Ship Recognition Based on KNN

Nearest Neighbor Algorithm

Meili Zhang

, Yue Yang, Hongmei Pei and Weili Liu

Dalian Naval Academy, Dalian, China

Keywords: Merchant Ship Identification, UAV, KNN Nearest Neighbor Algorithm, Shortest Path Algorithm.

Abstract: The planning of the port closure and control force is an important issue related to the port defence in the

face of war. According to the given information and the given coordinate information, this paper makes an

initial classification of them, and uses KNN nearest neighbour algorithm to classify the location information

of the given coordinates and obtain the central point by using its clustering information. After classification

training, its accuracy is high, and it is most appropriate when k=5. After obtaining the best classification

central point, take the classification central point as the initial starting position of the UAV, and take five

points as the starting points to establish the shortest path problem based on the Dijkstra algorithm. Based on

the idea of the shortest path, build the path planning strategy, analyse, model, and solve from a new

perspective, and optimize the closure and control force planning.

1 INTRODUCTION

In recent years, with the increasingly fierce

competition for maritime rights and interests of

various countries and the increasingly severe anti-

terrorism situation, coastal ports and bases have

gradually become the targets of terrorists and

enemies. With the military's awareness of the threat

to the security of ports and important coastal bases,

security measures at these strategic locations have

been gradually strengthened in recent years, leaving

fewer and fewer opportunities for terrorists and

adversaries to carry out sabotage from the road.

Therefore, the sealing-control operation planning of

the port is an important issue concerning the port

protection in the state of battle. In order to complete

the combat task, it is necessary to identify and

classify the merchant ships entering the port, and

study how to use UAVs to plan and intercept the

target and find the optimal strategy in the case of

port sealing and control.

At present, the research on port containment and

control operation planning mainly focuses on the

safety identification of merchant ships and the target

planning of coast guard ships and frigates in

blockade and interception operations. Although most

of the literature gives calculation methods from

different perspectives, most of these calculation

formulas are recursive formulas, which is rather

complicated. This paper mainly solves the problem

of optimizing the use of UAVs in the class

identification of all merchant ships and the problem

of sealing, control and interception and force

adjustment strategy.

2 PERPROBLEM DESCRIPTION

It is assumed that the port's external route is located

in a fan-shaped area with the port as the centre of the

circle and the orientation between 20°and

70°clockwise with the direction of due north, and the

course of merchant ships heading to a certain port

points to the centre of the circle, and the interception

disposal area is shown as the figure ABCD, as

shown in Figure 1.

Figure 1: Port diagram.

Zhang, M., Yang, Y., Pei, H. and Liu, W.

Target Planning for UAV Merchant Ship Recognition Based on KNN Nearest Neighbor Algorithm.

DOI: 10.5220/0012272700003807

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 2nd International Seminar on Artiﬁcial Intelligence, Networking and Information Technology (ANIT 2023), pages 11-15

ISBN: 978-989-758-677-4

There are currently two frigates and three coast

guard ships performing containment and control

tasks. According to the given data, it is generally

required that all Class III ships be intercepted, and

the task planning of containment forces with the

highest interception rate for Class II ships is given,

including the initial position of deployment,

interception objects, and movement trajectory.

3 RECOGNITION OF

MERCHANT SHIPS --

ESTABLISHMENT AND

SOLUTION OF KNN MODEL

OF NEAREST NODE

ALGORITHM

Target identification refers to the identification of

target attributes, including motion attributes

(tonnage, velocity, acceleration, etc.), friend or foe

attributes (our side, friend, enemy, unknown), inter-

class attributes (air, sea, underwater), type attributes

(ship type or model) and class. The important

significance of target recognition is that it is

combined with target state estimation to form the

basis of battlefield situation assessment and threat

estimation and is an important basis for tactical

decision-making.

In general, merchant ships are divided into three

categories: merchant ships carrying general

necessities (Class I), merchant ships carrying major

strategic materials such as oil or gas (Class II), and

merchant ships carrying contraband such as weapons

and equipment (Class III). The significance of

merchant ship identification is that it is combined

with target state estimation, which forms the basis of

battlefield situation assessment and threat estimation

and is an important basis for tactical decision-

making.

At present, the methods that can be used for

security classification are decision tree (Kleinberg J

M, 1999), genetic algorithm (Gongde Guo, 2006),

neural network (Feng Guohe, 2012), naive Bayes

(Tang Huxin, 2016), vote-based method (Liu Tong,

2018), Rocchio classification (Zhang Zitong, 2019),

KNN classification (Shaozhong Cao, 2012),

maximum entropy (Shen Yuqing, 2004), etc. KNN

algorithm is one of the simplest methods in data

mining classification technology.

3.1 KNN Algorithm

KNN algorithm is mainly based on a limited number

of adjacent samples to determine the category, so it

can be classified by measuring the distance between

different eigenvalues. Based on the KNN algorithm,

the proximity of the distance between the initial

information samples of merchant ships and the

mathematical values of these points is generally

measured by Euclidean distance. Assuming there are

two points p and q in the position samples of

merchant ships, the Euclidean distance formula

between the two points is

In addition, its testing phase is slow and

expensive in terms of time and memory, requiring

large memory to store the entire training data set for

prediction. Because KNN uses the Euclidean metric

between two data points to find the nearest

neighbour, we need to scale the number. In other

words, KNN algorithm is not suitable for large-

dimensional data, so it needs to reduce the

dimension to improve performance, but it has a good

classification effect for small data sets in this paper.

Moreover, to get better results, we standardize data

on the same scale, choosing a standardized range

considered between 0 and 1.

In the KNN algorithm, k is the number of nearest

neighbours, and this number of neighbours is the

decisive core factor. If the number of classes is 2,

then k is usually an odd number. When k = 1, the

algorithm is called the nearest neighbour algorithm.

Specific calculation steps are as follows, as shown in

Figure 2.

Figure 2: Specific calculation steps.

ANIT 2023 - The International Seminar on Artiﬁcial Intelligence, Networking and Information Technology

The KNN algorithm is more accurate at performing

a smaller number of features than a larger number of

features. As the number increases, the amount of

data required exceeds it. At the same time, the

increase in size will also lead to the problem of

overfitting. Its neighbourhood number k is a hyper

parameter that needs to be selected during modelling,

and k can be regarded as the control variable of the

prediction model, as shown in Figure 3.

Figure 3: Category distance diagram.

To sum up, the nearest node algorithm is used to

classify the location information of a given

coordinate and obtain the central point. The position

of each merchant ship is displayed through KNN

model clustering by MATLAB software program, so

that the merchant ship is recognized and detected by

UAV. Figure 4 shows the positions of the three

classes of merchant ships on the left, and the red dot

on the right of Figure 4 shows the central points of

the three classes of merchant ships.

Figure 4: Classification and central points of merchant

ships.

The above results are summarized in Figure 5,

which shows the positions of class I, II, and III

merchant ships and their central points respectively.

Figure 5: Merchant vessel category classification and

central point.

The problem of identification and safety

classification of merchant ships uses the thought

information of clustering

to classify the location

information of a given coordinate and obtain the

central point. After classification training, its

accuracy is high, that is, the most appropriate K=5.

3.2 Dijkstra Algorithm

The existing means can find out the geographical

location, course speed and other information of the

commercial ship to enter the port at a long distance,

but the nature of the cargo carried by it needs to use

the UAV close-in reconnaissance.

The unmanned reconnaissance aircraft is

equipped with TV cameras, forward-looking

infrared instrument, synthetic aperture radar and

other loads, with an endurance time of more than 30

hours, a cruise speed of 120km/h, and strong

mobility (turning radius is not considered) and can

be independently deployed in important sea areas to

implement long-term and continuous reconnaissance

and surveillance tasks on maritime targets. The

UAV approaches the target at sea according to the

location information provided by the intelligence.

When the weather conditions are good, the target

can be found at 20-30km, and the altitude can be

lowered after approaching 10km. The UAV will

circle the target for several weeks at a distance of no

more than 2km and take pictures from multiple

perspectives. Shore-based operators can identify the

ship name (side number), ship type (such as

container ship, oil tanker, natural gas tanker, cruise

ship), material type (such as oil, natural gas, coal,

iron ore, container material), flag (nationality), etc.,

and compare with the information and ship database

obtained by the red side to determine whether the

ship name and appearance are consistent. Whether

the ship name is altered or forged, whether the bad

records are in the case, whether the weapons and

Target Planning for UAV Merchant Ship Recognition Based on KNN Nearest Neighbor Algorithm

equipment are smuggled, etc., climb and fly away

after the completion of the identification, the entire

identification process takes 10 minutes, and the

identification results can be shared to any demand

party in real time.

The most commonly used routing algorithms

include Dijkstra algorithm (Liu Xuhong, 2005),

SPFA algorithm (Nannicini, 2008), Bellman-Ford

algorithm (Bang-Jensen, 2000), Floyd algorithm

/Floyd-Warshall algorithm (Shi Ren, 2009) to adjust

and intercept strategy schemes. In order to solve the

UAV reconnaissance route after the classification

results in the model, the shortest path can be solved

for the UAV starting from the fixed point. It is

solved by Dijkstra algorithm. The position

information is turned into a graph to facilitate its

subsequent calculation of the shortest flight path.

The Dijkstra algorithm is effective to solve this

problem. The interception problem and the merchant

ship entering the port need to be calculated and

analysed immediately, and the computational

complexity of this algorithm is low.

The complexity of Dijkstra's algorithm: (1) Time

complexity , where e is the number of

edges and v is the number of vertices. (2) Space

complexity .

The Dijkstra algorithm is effective to solve this

problem. The interception problem and the merchant

ship entering the port need to be calculated and

analysed immediately, and the computational

complexity of this algorithm is low, as shown in

Figure 6.

Figure 6: Dijkstra solution.

The model takes points 18, 23, 49, 53, and 60 as

the starting points, and follows 18, 15, 7, 6, 48, 57,

17, 67, 45, 79, 84, 66, and 23, 55, 64, 46, 73, 56, 63,

71, and 39, respectively. Python was used to analyse

and solve the model, and the results were obtained

after 100 iterations. Iterating the python solution 100

times. By substituting the model for calculation, it is

obtained that its initial position is (113.6410,

353.4564), and its intercepting objects include all of

the ship marks of the three types and can intercept

83% of the part of the second type. Its motion

trajectory starts from point (15) and intercepts along

the shortest path of the three types, and most of the

ships of the second type can be intercepted.

Using MATLAB software, we get the best route

for the UAV to travel through all merchant ships,

and the UAV can travel according to this route,

which can make the UAV spend the shortest time

and experience the shortest path, as shown in Figure

Figure 7: The shortest flight path of UAV.

4 CONCLUSION

Through the research and demonstration in this

paper, we first use KNN algorithm to classify and

identify merchant ships. This method has strong

robustness to noise training data. Starting from the

geographical coordinate position of merchant ships,

this algorithm solves the problem of merchant ships

category well. Through the research and

demonstration in this paper, we know that it is easy

to implement KNN algorithm, and this method has

strong robustness to noise training data. From the

perspective of algorithm performance comparison,

take Floyd algorithm and Dijkstra algorithm as an

example. On the one hand, if Floyd algorithm is

applied to a certain vertex successively, then

compared with Dijkstra algorithm, many path and

result calculations are repeated. Although the

complexity is the same, the calculation amount is

much different. At the same time, more importantly,

Floyd's algorithm requires no loops with a sum less

ANIT 2023 - The International Seminar on Artiﬁcial Intelligence, Networking and Information Technology

than 0, while Dijkstra's algorithm uses the

prerequisite that as long as the length of the path in

the graph is greater than or equal to 0. Therefore, we

use Dijkstra algorithm to calculate the shortest path

of UAV classification of merchant ships and

optimize Floyd algorithm.

ACKNOWLEDGMENTS

This paper is one of the stage achievements of the

Research Fund project of Dalian Naval Academy of

the People's Liberation Army Navy (DJYKKT2023)

and the research project of Dalian Naval Academy

students (DJYKKT2022-007, DJYKKT2022-003).

REFERENCES

Kleinberg J M. Authoritative sources in a hyperlinked

environment[J].Journal of the ACM, 1999, 46 (5): 604

－632.

Gongde Guo, Hui Wang, David Bell, Yaxin Bi & Kieran

Greer. Using KNN model for automatic text

categorization[J]. Soft Computing, 2006, 10(5): 423-

430. https://doi.org/10.1007/s00500-005-0503-y

Feng Guo he, Wu Jingxue. Advances in the improvement

of KNN classification algorithm [J]. Library and

Information Work, 2012, 56(21):97-100.

Tang Huxin, Website Operation and Data Analysis two-

colour edition, Aviation Industry Press, 2016, (12):81-

82.

Liu Tong, Gao Shichen. Optimization of KNN method

based on data spatial structure [J]. Mathematics in

Practice and Understanding, 2018, 048(014):197-206.

Zhang Zitong, Zhang Chun lei, Zhang Yan, et al. Data

space structure and its application in KNN algorithm

[J]. Mathematics in Practice and Understanding, 2019,

49(1):8.

Shaozhong Cao, Ji Tu. Double Motor Coordinated Control

Based on Hybrid Genetic Algorithm and CMAC [J].

Physics Procedia, 2012 Vol. 33, pp. 975-982

https://doi.org/10.1016/j.phpro.2012.05.163

Shen Yuqing, Xu Lihong. A Genetic algorithm based on

pattern clustering [J]. Microcomputer Applications,

2004(04):9-11.

LI Ping. Improvement and Implementation of Dijkstra

Algorithm [J]. Information Construction,

2020(02):345.

Gao Wu, Zhou Gongjian, Wang Yang. Application of

Dijkstra Algorithm in Temporary speed limiting

operation [J]. Electronic World, 2020(08):31-32+35.

Li Min qiang. Basic Theory and Application of Genetic

Algorithm [M]. Science Press, 2002.

Liu Xuhong, Zhang Guoying, Liu Yushu, et al. Path

planning based on multi-objective genetic algorithm

[J]. Journal of Beijing Institute of Technology, 2005,

25(7):4.

Nannicini, G., Delling, D., Liberti, L., Schultes, D.:

Bidirectional A* Search for Time-Dependent Fast

Paths. In: McGeoch, C.C. (ed.) WEA 2008. LNCS, vol.

5038, pp. 334–346. Springer, Heidelberg, 2008.

Bang-Jensen, Jørgen; Gutin, Gregory. Section2.3.4: The

Bellman-Ford-Moore algorithm. Digraphs: Theory,

Algorithms and Applications[M], ISBN: 978-1-84800-

997-4, 2000.

Shi Ren, Wang Kai. Research on Shortest path Planning

of Mobile Robot based on Floyd Algorithm [J].

Chinese Journal of Scientific Instrument, 2009,

30(10):2088-2092.

Target Planning for UAV Merchant Ship Recognition Based on KNN Nearest Neighbor Algorithm