Tomtat luanan nâng cao chất lượng lọc, bám quỹ Đạo mục tiêu Đa Điểm chói trên biển

To drawback with the problem, the dissertation provides a new method for improving the tracking accuracy of marine target in command and control.. Research goal Research and evaluate a

MINISTRY OF EDUCATION AND TRAINING MINISTRY OF DEFENSE

ACADEMY OF MILITARY SCIENCE AND TECHNOLOGY

VO XUNG HA

IMPROVED FILTERING AND TRACKING OF MULTI-PLOT

MARINE TARGETS AT SEA

Specialization: Radar and navigation engineering Code : 9 52 02 04

SUMMARY OF PH.D THESIS IN ENGINEERING

HA NOI – 2024

THE THESIS HAS BEEN COMPLETED AT

ACADEMY OF MILITARY SCIENCE AND TECHNOLOGY-

GENERAL STAFF

Scientific Supervisors:

1 Dr Nguyen Trung Kien

2 Dr Nguyen Phung Bao

Reviewer 1: Prof Dr Vu Van Yem

Hanoi University of Science and Technology

Reviewer 2: Prof D.Sc Do Duc Luu

Vietnam Maritime University

Reviewer 3: Assoc Prof Dr Le Vinh Ha

Academy of Military Science and Technology

The dissertation was defended at the Doctoral Thesis Evaluating Council at Academy level held of Academy of Military Science and Technology at ……… , 2024

The dissertation can be found at:

- The library of Academy of Military Science and Technology

- Vietnam National Library

INTRODUCTION

1 Overview of dissertation topic

Radar systems play an important role in wars and conflicts The radar systems are used for estimating and determining position and type of targets, based on received signal analysis Also, the radar systems provide a real electronic situation for commanders and supports them to decide in time

In real situations, the command and controls systems require updated and accurate information about type of targets and their trajectories But the problem is that many types

of electronic devices are in the conflict zones, especially for marine targets such: reflected signals from sea surface, ships, low flying targets and transmission loss They are main causes of errors in determining targets of the marine radar systems

There are two groups of methods for improving accurate information The first one uses high precision filters such as: alpha-beta filters, Kalman filter and its modification and combination of filters The second way uses complex techniques to analysis received signals However, recent radar systems use only simple technique for estimating position and tracking target based on binary and barycentric method

To drawback with the problem, the dissertation provides a new method for improving the tracking accuracy of marine target in command and control So, the topic of

dissertation is “Research on improving tracking trajectory of multi-plot marine targets”

2 Research goal

Research and evaluate a new method for improving tracking trajectory of multi-plot marine targets The time processing and accuracy of trajectories are used for evaluating performance of the proposed method

4 Research content

The dissertation includes four stages The first stage is used for estimating plots of targets and their parameters such as: area, reflected energy and coordinates (range and azimuth) The second stage is used for clustering estimated plots into represented targets and calculating position and direction of targets Based on estimated parameters, a new method for improving tracking trajectories is proposed in the third stage Finally, the performance of the proposed method is evaluated by simulated and real data

5 Methodology

The theoretical of the proposed methodology is analysis and detail described in the first step The next step provides evaluation performance of the proposed method by simulation and real data by using MATLAB

6 Scientific significance and practical significance of the dissertation

Scientific significance: The dissertation proposes an algorithm for estimating

characteristics of targets such as: position (range, azimuth), movement direction, reflected

areas and energies of targets Also, a new method for improving tracking trajectories of targets is proposed based on their estimated parameters

Practical significance: The results of the dissertation can be applied to real radar with

powerful signal processing system

7 Content of dissertation

The dissertation includes three parts 1 Introduction, 3 chapters and conclusion

CHAPTER 1 OVERVIED OF MULTIPLOT MARINE TARGETS

1.1 Reflected signals and radar images of multiplot marine targets

1.1.1 Reflected signals and radar images of multiplot marine targets

The scatter field of a multi-plot target is the sum of the vectors of the fields of the independent scattering centers that make up the target itself Modeling useful signals reflected from multi-plot targets can be done by calculating the signals reflected from the

"brightest" plots of the target’s RCS, which are represented as point reflections (known

as glare plots)

4

5 6 7

10

11

X Y

Figure 1.1 Model of a medium-sized military ship Figure 1.1 shows a model of a target model of 12 point-glare reflections applied to military ships Accordingly, the medium-range military ship target model has 12 dispersion centers located on its surface

The composite signal reflected from the multi-plot marine target is the result of the interference of the component signals reflected from the bright plots that constitute the multi-plot target:

ultra-Thus, the RCS of a bright multi-plot target is a vector whose magnitude is approximated to the sum of the vectors of the RCS of the component elements of the target

1.1.2 Characteristics of detected marine targets

With the multi-monitoring output heads of marine targets, the resolution is determined

by two parameters, the pulse length l xg and the length of the arc, which correspond to the width of the antenna bud

0.5

l in the azimuth plane

The magnitude of the return arc length depends on the detection distance in the form of 𝑙𝛽0.5:

0.5

And the resolution is calculated by (1.12)

1.1.3 Radar image of multi-plot marine targets

Figure 1.4 shows a display region with multiple targets (right) and a 3D image of a hand-selected target for analysis (left) Analysis of this image through cycles shows that the large-sized marine target has many plots (protruding and pointed areas in the figure) that are relatively stable corresponding to the positions of antennas, conning towers and their metal structures Figure 1.5 shows a radar images of single marine targets It is shown that the target has multiple brightness plots

Figure 1.4 Score3000 radar display

Figure 1.5 Radar image of marine target

Figure 1.7 Image of a bright multi-plot marine target

1.2 The problem of tracking trajectories of marine targets

The process of tracking trajectories includes three stages The first stage is used for target detecting target and the second is used for estimating its parameters, finally the tracking method is used corresponding to XLC1, XLC2 and XLC3 [4], [76], [84]

Level 1 Radar Signal Processing

(Point Coordinate Estimation)

- Target Location Estimation

- Target Feature Extraction

Linking Markers (LKDD) (Determine which mark belongs to

In this section, provides an analysis of clustering algorithms

According to the general definition [76], trajectory identification (TI) in trajectory processing (TP) is understood as the process of making decisions about the compatibility

of the data on the obtained markers with a certain trajectory at a specified observation period TI with trajectories appearing in TP when there are pseudo markers and/or there are concurrent neighboring trajectories in the same beat period so that the processed marker can simultaneously accommodate the above trajectories [65], [67], [74]

a) Single target b) multiple targets

Figure 1.9 Examples of situations where the correct link is required

TI algorithms include steps to generate possible connections between markers and trajectories Currently, the most popular TI algorithms are the Bayess and non-Bayess groups of algorithms With the group of non-Bayess algorithms, the hypo-dissertation

of production is built on the assumption of the authenticity of one of the sign plots and does not make any assumptions about the other sign plots For the Bayess group of algorithms, when estimating the dynamic state of an object, the algorithm would have

to take into account hypotheses not only in relation to the n-th time marker but in all

previous update cycles

1.3 The problem of analyzing radar images of marine target

To reduce the amount of computation without compromising the quality of the radar marks (RM) estimation problem, many authors binarize the radar images The advantages of the binary method are simple and fast calculation The disadvantage of this method is that the target can be missed when the maximum intensity value is smaller than the detection threshold [15], [23], [47], [64]

To solve problem of the binary method, Селезнева О.В [99] provided a new method, which decompose the radar image into three layers based on the energy level Figure 1.10 shows the advantages of using various threshold levels to detect marine targets With photos of 3 targets of different intensity: strong reflective targets, medium reflective targets, and weak reflective targets If a fixed threshold is used (corresponding

to the radar image binarization threshold), assuming the selected threshold is marked with continuous strokes, a weak reflection target may be missed With the same 3 targets

as above, when using multiple thresholds, assuming using 5 thresholds, it will reduce the probability of missing the target, especially for targets with weak reflexes

Ngưỡng cố định

Nhiều ngưỡng

In 2020, Viettel Group researched, designed, and manufactured the VRS-CSX medium-range marine multi-realm Observing the target image on Figure 1.12, the bright plots of the marine target rise and fall through successive cycles show that the reflected signal is greatly affected by sea turbulence However, the multi-plot radar image of the sea is still brought back to the point target providing RM for the filter, tracking the trajectory

Figure 1.12 Item image of the VRS-CSX radar station at Do Son C500 Station

In 2023, MTA will improve the NSC34 radar station, the handling of multi-plot targets also brings back to the center of the image, that is, it still considers marine targets as single-point glare targets

Figure 1.13 NSC34 radar display at C500-Do Son station

Figure 1.14 tracking trajectory of marine targets Observing on the screen of the improved Rangout radar (Radar Institute/Institute of Military Science and Technology), the trajectory after filtering does not follow the target's movement correctly due to the error of the RM estimation problem (Figure 1.14) Therefore, there are still disadvantages as follows:

- The coordinates of the target are usually estimated after each antenna scan Because during the period of 1 antenna scan, the marine target antenna did not travel the entire distance equal to the length of the hull (30350 m);

- The value and method of the large upward and downward velocities cause loss

of grip (Figure 1.13);

- The target trajectory is tracked; the same target appears many trajectorys with different directions of motion (Figure 1.14)

1.4 Research Orientation

To solve the problem of improving accuracy parameter measurement based on time radar image processing The problem of analyzing multi-plot marine target include:

real-1 Marine targets are usually in the class of multi-plot targets, when processing radar portraits of the same detected target, there will be many markers that are considered

"real" to be confirmed in the same observation cycleT qs

According to the classic numerical processing detection algorithms synthesized by the continuous testing method, the typical model is the l/ (m−k) or the CFAR algorithm with their modification So, this is a cause of estimation accuracy reducing by different levels Therefore, an efficient algorithm is needed to designed for improving the accuracy of target RM estimation for tracking trajectories for multi-plot marine targets

2 The formation of high-resolution radar images can add the attribute of the strength of

the reflected signal from the bright plots to act as the third coordinate parameter which is the quantitative parameter in the calculation problem to improve the accuracy of measuring and estimating the parameter of the marker point representing the marine target.

1.4.2 Summarizing the cluster algorithm of the multi-plot marine target

The problem is that from the acquired radar image data, after processing, "multi-plot" will be formed It is necessary to determine which bright plots belong to which target, from which an RM for the target will be estimated

On the basis of synthesizing and analyzing relevant research results and especially taking into account the accuracy and real-time of processing, the research approach in the dissertation is the application of high-resolution multi-resolution image processing techniques combined with the algorithms of data clustering, each data cluster consists of bright plots belonging to a target Especially for cases where the number of clusters (or in other words, the number of targets) is unknown

1.4.3 Current methods for tracking trajectories of marine targets

Currently, Kalman filtration is still the most effective tool applied to problems of filtering and tracking trajectories In the dissertation, Kalman filtering is still used to ensure real-time provision of information and output data about trajectory The dissertation focuses on researching the application of useful information of "multi-plot" targets in LKDD processing to improve accuracy, while speeding up the processing process to provide real-time information for command and control

The accuracy of the filtering trajectory is mainly determined by the quality of the input data source for the TP filter, which is the representative position of the target However, in complicated situation when maneuvering targets to fight, there may still be a phenomenon

of filtering and mis-tracking the trajectory due to the physical distance between the two targets similar to the critical size of the filter and trajectory To solve the problem stated, the approach in this study is to apply the theory of maximizing the "multi-plot" property to handle the dispute situation mentioned above

On the basis of synthesizing the research results obtained, the algorithm for filtering, tracking and trajectory development will be synthesized, and the correctness will be tested

by simulation method using experimental measurement data sets

1.5 Methods of conducting and evaluating research results

1.5.1 Database of marine targets

The radar image data used in the dissertation was provided by the Marine Coast Watcher 100 radar [54], according to the ASTERIX CAT-240 data standard, CAT-10 using Wireshark software [17], [20], [24]

The data of the radar marine target is a matrix of A markers with the size of N rows,

1.5.2 Methodology and evaluation performed of the proposed method

The research diagram is shown in Figure 1.16, analyzing and evaluating the results based on actual radar image dataset The process of testing and evaluating method’s performance is provided in MATLAB environments

Preprocessing of multi-luminance target images, estimation of multi-luminance

coordinates

by image classification

Clustering of Luminance Points for Each Target (Proposed New Algorithm)

Representative marker estimation (DDD) and

target parameter extraction

- Filter, track single target multi-point

luminance

- Filter, track group multi-point luminance

(Proposed new algorithm)

Figure 1.16 Methods of conducting and evaluating research results

1.6 Conclusion of the first chapter

Chapter 1 studies the theoretical basis related to the problem of filtering and tracking the trajectory of targets at sea to build and form a problem to improve the quality of tracking of multi-plot marine targets for providing information in command and control

In the chapter, the general research contents for the dissertation are as follows:

1 A radar image processing method is proposed from the perspective of a multi-plot target that leads to the formation of a multi-plot set in an observation cycle

2 Applying cluster algorithms for grouping estimated plots into targets

3 Using the multi-plot characteristics of marine targets to improve the accuracy of estimating and extracting their characteristics such as: position (range, azimuth), direction, reflected area and energies

4 Researching a new method for filtering and tracking marine targets based on their estimated characteristics

5 Proposing a new method for improving accuracy of tracking marine targets in real situations

Some contents of the overview analysis published in the work [CT1] in the list of published scientific works

CHAPTER 2 MARINE TARGET CHARACTERISTIC ESTIMATION 2.1 Movement characteristics of marine target

2.1.1 Optimization of binary method

The binary method is known as iso-weighted algorithm Its principle is that the target

is detected when the total number of received signal greater than threshold value K

(positive integer)



1

,

M

i i

where x i is the TLN sequence; M is the number of pulses in the TLN burst and K is the threshold value

It is possible to perform the technique (2.2) using a counter and a number comparator

The system that implemented binary method (2.2) is also known such as "K/M" detector

For each value M - the number of TLN, there exists an optimal threshold number Kopt defined such as:

𝐾𝑜𝑝𝑡 = { 0,5𝑀

1,5√𝑀

Slow fluctuance

(2.3) Fast fluctuance

2.1.2 Algorithm for determining the optimal threshold for binary image radar

The dissertation used a modification of "Optimal Algorithm for Binary Quantum Signal Beam Detection" to determine the optimal threshold of binary radar image using image data in each scan cycle, which helps reduce time processig

The assumed target data is that the matrix A has a 3-dimensional space that includes information about the range, azimuth, and energy level nm Thus, from (2.1) , the radar image matrix are plated in many range and azimuth cells can be written as:

T N ; N is the total number of pixels with a positive value

2.1.3 Estimation of radar target representative coordinates

The mathematical basis for estimating target RM from binary radar target images is based on the problem of finding the center of gravity of a cylinder bounded by the curved surface f(x,y) and the base D

The center C x y of the flat plate of D (Figure 2.2.b) with area S is calculated D( D, D)according to the formula as follows:

a) Describe the center of the planar plate D and the

Figure 2.2 Estimation of the center of the cube and plane D

2.1.4 Representative coordinates estimation based on binary method

a) Performance evaluation of the optimized binary method

To evaluate the effectiveness of the formula (2.5), the fast ascension target is used The performance of this method for this target (MT1) is evaluated over 16 radar scan cycles

Figure 2.4 Calculation of Radar Image Binary Threshold MT1

Analyzing the data of the MT1 radar image, for the threshold (K=0.5T – the threshold for

the slow ascending target) in the 4th, 9th, 12th, 13th, 14th, 15th and 16th scan cycles, the

number of pixels cut off accounts for nearly 50% of the total number of pixels (Figure 2.4 Calculation of Radar Image Binary Threshold MT1Fig 2.4a) For the optimal threshold, the number of pixels cut off was not much during all 16 scan cycles (Figure 2.4 Calculation of Radar Image Binary Threshold MT1Fig 2.4b) The number of pixels cut gradually corresponds to large binary thresholds, corresponding to the average energy level

T If the radar image binaries threshold is increased, the number of cropped pixels also

increases Thus, the information about the radar image will be lost, the image data after binary

is only the pixel with a large energy level (glare), these glare plots will increase and fall greatly, which will lead to a very large error in the radar image center estimation problem For large targets at sea, radar images have many bright plots that rise and fall rapidly through scan cycles On the basis of analysis and calculation, the dissertation proposes

to select the optimal threshold K opt = 1,5 T to binary the target image radar for the purpose of background noise compression, while ensuring that the image information is kept intact and improve the quality of the RM estimation problem of radar target images

b) Results of estimation of the representative coordinate

Above Figure 2.11 describes the binary results and estimates RM MT1 with optimal thresholds in 3 scan cycles 1, 2 and 3 The results show that the RM of the target over the scan cycles depends on the number of pixels distributed by the image after binary

Figure 2.11 MT1 Target Momentum Estimation Results Above Figure 2.12 is the result of calculating the average velocity and instantaneous velocity (normalized, taking the maximum velocity in cycles equal to 1) of the target

through the scan cycles The results show that for the binary threshold according to the

average velocity KOtsu and the instantaneous velocity change rapidly through the cycles,

this is not consistent with the fact that when the selected target moves slowly, the direction

of movement almost does not change through the cycles For the binary threshold according

to the Kopt average velocity and instantaneous velocity of the item The increase in

expenditure is not much and is relatively stable, which is in line with the chosen goal This once again confirms that the optimal threshold of radar image binaries is suitable for the problem of processing large radar images, which fluctuates rapidly through cycles

a) Average velocity b) Instantaneous velocity

Figure 2.13 MT1 Velocity Estimation

2.2 Estimation of the representative coordinates based on radar image decomposition

In order to improve the quality of target RM estimation at sea, the dissertation proposes a real-time radar image processing plan for each scan cycle

In order to estimate RM based on radar image layering, the dissertation focuses on the following main problems: layering algorithm according to the level of reflected signal from marine targets, identification of radar image local maximum regionss, estimation

of the center of glare and RM on the basis of radar image local maximum regionss

2.2.1 Optimizing number of layers

In order to properly layer the target image according to the pixel intensity level, it is necessary to evaluate the overall distribution of the intensity level of the entire image To evaluate the distribution of the intensity level of the image, a histogram is usually constructed

In order to construct a histogram, it is necessary to determine the number and size of histogram bars There are many rules for determining the number of bars of a k-histogram such as: Sturges rule, Rice rule, Doane rule, Scott standard reference rule, Freedman-Diaconis

- Square-root Rule: The number of bars of the

K histogram is determined by the formula: k n

k

k

i

k n

+ The number of layers selected when

building a chart from regular data is calculated

by the formula:

2

k= log n+1. (2.14)

- Scott Rule: The width of each h-column is

3, 49 ,

n



- Freedman–Diaconis Rule: The width of each

H column is calculated according to the formula: 3

IQR(x)

n

- Doane Rule: The number of bars in a

(2.17)

(2.18) After making a histogram of each target group such as fishing ships, cargo ships, and