FR3070768A1 - METHOD FOR AUTOMATIC CLASSIFICATION OF A TARGET VESSEL AND RADAR SYSTEM THEREOF - Google Patents

Abstract

This method of automatic classification of a target ship (2) is characterized in that it is based on a sequence of energy profiles (Pi) delivered by a radar system (8) carried by an aircraft observing said target ship (2) , the sequence comprising at least one energy profile (Pi) indicating, for an observation direction (Z) of the radar system (8), a backscattered energy as a function of a distance from the radar system, and in that it consists of associating with the target ship (2) a class (Cj) selected from a group of predefined classes according to the values (pki) of a plurality of morphological parameters (pk) calculated on each profile of the sequence.

Description

METHOD FOR AUTOMATIC CLASSIFICATION OF A TARGET SHIP AND RADAR SYSTEM THEREOF

The present invention is in the field of maritime surveillance. It relates more particularly to the classification of ships.

Vessels provide self-declarative information that gives the identification, position, course and dimensions (length and width) of the vessel, using an automatic identification system, known by the acronym AIS ("Automated Identification System") ).

It is useful to be able to verify the information provided by a ship, in order to detect a suspicious ship, in a civilian or military context.

It is known to use the information provided by a radar in order to classify a target vessel, that is to say to be able to associate this target vessel with a class of vessel (such as the oil class, bulk carrier, container ship, fishing boat, etc.), in order to check the consistency of the information it provides with the corresponding information for the class with which it is associated.

Currently, the classification of a target vessel is carried out using radar images, for example SAR ("Synthetic Aperture Radar") or ISAR ("Inverse Synthetic Aperture Radar") of this target vessel.

These imaging modes require a long observation period of the targets taking into account the periods of their own movements (roll, pitch and yaw) of the latter.

It can therefore only be high-level information, obtained after the operator has decided to follow a previously detected vessel of interest. The classification is therefore made a posteriori.

Furthermore, a radar system is capable of providing an energy profile (or “Range Profile” in English), defined as a graph of the power reflected in an observation direction as a function of the distance from the radar system in the direction of observation.

However, until now, the classification of ships has not been based on energy profiles, because the nature of these signals, which is both very fluctuating and very dependent on the environment, is not favorable.

This great variability in the energy profile is due to the sensitivity to the angle of view; fluctuations in the power backscattered by a target; the environment, including sea clutter; variability between ships of the same class or between the same ship when loading different loads.

The invention therefore aims to overcome this problem, in particular by providing classification information for a target vessel without having to use a radar image and to perform complex calculations on this image.

The subject of the invention is therefore an automatic classification method of a target vessel, the method being based on a sequence of energy profiles delivered by a radar system carried by an aircraft observing said target vessel, the sequence comprising at least one energy profile indicating, for a direction of observation of the radar system, a backscattered energy as a function of a distance from the radar system, and in that it consists in associating with the target ship a class selected from a group of classes predefined according to the values a plurality of morphological parameters calculated on each profile of the sequence.

According to particular embodiments, the method comprises one or more of the following characteristics, taken in isolation or according to all technically possible combinations:

- the method comprises the steps consisting in, for each energy profile of the sequence, calculating the value of each of the morphological parameters of a set of predefined morphological parameters; for each morphological parameter, determine a score indicative of a correlation between the value calculated for said parameter and a reference value of said parameter for each class of the group of classes; for each class in the class group, aggregate the scores determined for each morphological parameter and for each energy profile of the sequence in a global score; and associating with said sequence the class for which the overall score is the highest.

- the reference value of a parameter for a class is chosen according to a viewing angle of the target ship by the radar system;

- the class for which the overall score is the highest is associated with said target ship only when said overall score is above a rejection threshold;

a morphological parameter of the plurality of morphological parameters is chosen from: a parameter corresponding to an equivalent radar surface; a parameter corresponding to a distribution of the equivalent radar surface along the area of the energy profile corresponding to the target ship; a parameter corresponding to an average slope along the area of the energy profile corresponding to the target vessel; a parameter corresponding to a proportion of the area of the energy profile corresponding to the target vessel which is flat; and, a parameter corresponding to a distance between peaks of the area of the energy profile corresponding to the target vessel;

- the method comprises an initial step consisting in rejecting an energy profile of the sequence when a dimension of the target vessel obtained from said energy profile is not consistent with the dimensions obtained from the other profiles of the sequence;

- an operator of the radar system defines a class of interest and the method comprises a step of generating an alert when the class associated with the target ship corresponds to the class of interest.

The invention also relates to a radar system capable of implementing an automatic classification process for a target ship as defined above.

The invention and its advantages will be better understood on reading the detailed description which follows of a particular embodiment, given solely by way of nonlimiting example, this description being made with reference to the appended drawings in which:

- Figure 1 is a schematic representation from above of a radar system carried by an aircraft when observing a target ship;

- Figure 2 is an energy profile delivered by the radar system of Figure 1;

- Figure 3 is a representation in the form of a block diagram of the method according to the invention, implemented by the radar system of Figure 1; and,

FIG. 4 is an example of a function used in the method of FIG. 3.

FIG. 1 schematically represents a radar system 8, carried by a maritime surveillance aircraft (not shown in FIG. 1), observing a target ship 2, the search for which belongs to a class among a group of pre-established classes.

The length L and the width I of the target ship 2 are defined by the corresponding dimensions of a rectangle inside which the target ship 2 is inscribed.

When the target ship 2 is in motion, this movement is characterized by a speed vector V, which is parallel to the longitudinal axis of the ship.

An antenna 4 for transmitting / receiving electromagnetic radio waves, forming part of a radar system 8, is capable of transmitting electromagnetic radio waves towards the target ship 2 and of receiving waves reflected by any reflecting structure encountered along of the antenna pointing direction Z 4.

The geographical vertical being indicated by the direction Y, the direction of the speed vector V of the target ship 2 makes an angle Θ with the vertical plane defined by the directions Z and Y.

The antenna 4 is coupled to a radar device 6, the set of elements 4 and 6 forming the radar system 8.

In known manner, the radar system 8 is capable of determining the value of the speed vector V projected onto the direction Z. This makes it possible in particular to determine whether the target ship is viewed from the back or from the front.

In known manner, the radar system 8 is capable of determining, by Doppler effect, the angle of view Θ under which the target ship is observed, that is to say the angle between the course of the ship and the direction of pointing the antenna.

The radar device 6 comprises in particular transmitter / receiver modules 10, modules 12 for modulation / demodulation of the signals, a unit 14 for calculation and processing comprising a processor and capable of performing calculations when the radar device 6 is powered up, one or more memory units 16 capable of storing data and computer program code instructions.

Optionally, the radar device 6 also comprises or is connected to a man-machine interface 18, comprising a display screen and means of interaction with an operator.

In an alternative embodiment, a part of the modules of the device 6 is implemented independently, for example the man-machine interface 18, the memory units 16 and the calculation unit 14 are external to the device 6 and form part of a dedicated programmable device, for example a computer on the ground.

The radar system 8 is capable of acquiring an energy profile, or power profile, which is a digital signal which represents the power of the signal reflected in response to an emitted electromagnetic radio signal, as a function of the distance from a reference point. of the antenna 4. This signal is sampled by distance step, also called distance box.

FIG. 2 shows an example of a digital signal of acquired energy profile S _1t representative of the normalized power (ordinate axis) of the signal received by the antenna 4 as a function of the radial distance (abscissa axis), between a reference point of the antenna 4 and a spatial point of reflection located in the vertical plane defined by the directions Z and Y.

The digital signal Si has areas in which the amplitude is low (corresponding to noise or sea clutter) and areas in which the amplitude is high (corresponding to reflecting obstacles in the path of the electromagnetic waves emitted). The high power zones correspond to structural elements of the target ship 2. For example, metallic structural elements reflecting at the bow and stern of a ship, making it possible to indicate the ends of the ship along the Z axis observation and deduce the size of the target ship 2 in this direction.

FIG. 3 is a block diagram of the main steps of an automatic classification method of a target ship according to a preferred embodiment of the invention.

This method 100 is for example implemented by software capable of being executed by a processor of the calculation unit 14 of the radar system 8.

In step 105, a sequence of energy profiles is acquired.

The profiles are obtained successively, for a relative position between the radar system 8 and the target vessel 2 which is essentially identical.

The sequence comprises Np profiles Pi, with i integer between 1 and Np.

A sequence can contain between one and several tens of profiles.

The implementation of the method on a sequence of several profiles allows greater robustness than an implementation on a single profile, which by nature exhibits significant variability.

Advantageously, a filtering operation is applied to the digital signal of each profile Pi acquired to obtain a filtered digital signal.

In step 110, each profile Pi is processed to determine a dimension Li of the target ship 2. This dimension Li corresponds to the length L of the target ship when the direction Z of observation corresponds to the longitudinal axis of the ship and to the width I of the target vessel when direction Z corresponds to the transverse axis of the target vessel.

The dimension Li is for example obtained by retaining only the distance boxes of the profile Pi for which the reflected energy is greater than a predetermined energy threshold, then by determining the greatest of the distances between pairs of distance boxes selected.

A histogram is then constructed from the dimensions Li deduced from the profile Pi of the sequence.

This histogram is analyzed statistically to ensure that all of the profiles in the sequence have close dimensions. If the dimension Lm deduced from the profile Pm differs too much from the other dimensions, the profile Pm is eliminated from the sequence.

By this analysis, we validate the possibility of a joint exploitation of the profiles contained in the sequence.

Step 120 then consists in calculating, for each profile Pi, the values of a plurality of discriminating morphological parameters.

Preferably, the plurality of parameters comprises seven parameters pk, with k integer between 1 and 7.

The pki values of seven parameters are thus calculated for each profile Pi: five basic parameters and two specific parameters.

The first basic parameter p1 is the radar equivalent surface (SER) of the target ship 2. This parameter is a datum delivered by the radar device 6. A fold value is obtained for each profile Pi.

The second basic parameter p2 is the true length of the target vessel, calculated from the dimension Li and the angle of view according to the relationship:

p2i = Li / | cos (0) | ;

The third and fourth basic parameters, p3 and p4 respectively, are parameters which indicate the distribution of the radar equivalent surface along the profile Pi studied.

For this calculation, the energy profile Pi is transformed into decibel (dB), that is to say expressed in relative power. A standardized P _nO rmi profile is thus obtained.

The normalized profile is then reduced, on the Li dimension, to sixteen distance boxes of identical size (each box comprising 1/16 ^th of the profile). We obtain a reduced profile P rediOn then applies analysis windows which overweight either the start of the normalized profile for p3 (WinDeb window) or the central part of the normalized profile for p4 (WinCen window).

The formulas for calculating these two parameters are as follows:

p3i =

100 x

Σ? = 1 Predi (l) X WinDeb (I)

Z & PrediQ) _A. _____ Zi ^ iPredKOxWinCena) ^p4 ' ^{= 100x} --- ς & ρ ^ κο - Where P _red i (0 is the sample I of the reduced profile P _re d · and WinDeb (l) and WinCen (l) are the samples I of analysis windows associated respectively with p3 and p4.

The fifth basic parameter, p5, is an average slope parameter of the profile studied.

We first calculate the intermediate quantity Pentei from the relation:

Pentei (l) ⁼ | PnormîQ Ί Step) PnormKOI

Where PnormKO is the sample I of the normalized profile P _norm i and Step is an integer corresponding to the difference chosen between the two samples compared. For example Step is taken equal to 10.

The parameter p5 is then defined for the profile Pi by:

P5i = £ x ΣΓ = ι min (2, Pentei (l)) p6i = greatest value of T such that: f

The first specific parameter p6 is based on the slope and translates the proportion of the profile which is flat. It allows for example to determine the length of the deck of the target ship 2.

This parameter is determined by:

I = to + T Pentei (l) <PENTECUM t = to where tO runs through all the indices of the profile samples and PENTECUM is a predefined threshold on the cumulative slope.

The second specific parameter p7 is a parameter which translates the distance between the first peak and the maximum of the following peaks along the profile Pi considered. This parameter is based on an algorithm for extracting significant peaks known to those skilled in the art. This parameter allows for example to discriminate the bulk carrier and container classes, which are the closest.

The next step 130 is a classification step proper.

In sub-step 132, for each profile Pi of the sequence, a note nkij is calculated to quantify the adequacy of the value pki of the parameter pk calculated in step 120 with respect to a reference value pkj of this same parameter pk for a class Cj.

A class Cj is part of a group of predefined classes whose characteristics are stored in the memory of the radar system 8. If Ne is the number of class of the group, j is an integer which varies between 1 and Ne.

The adequacy is preferably evaluated according to the adequacy function fkj represented in FIG. 4. With such a function, the value of pki makes it possible to directly obtain the value of nkij.

Advantageously, for each class Cj, the characteristic variables Min, Max, σ, nO of the adequacy function fkj depend on the angle of view Θ. For example, two adequacy functions are defined for each class depending on whether the ship is viewed from the back or from the front.

Generally, the adequacy function is determined by learning. The analysis of the profiles of ships which we know belong to a particular class makes it possible to position the values of the characteristic parameters of the or each adequacy function of this class. These are the matching functions that are then used operationally.

Sub-step 132 is iterated over the integer k for each of the seven parameters.

Sub-step 134 then consists in calculating a metric Mij of distance between the profile Pi and the class Cj, by determining the product of the notes nkij obtained for the k parameters of the profile Pi with respect to the class Cj. This metric is indicative of the correlation of the profile studied with the class considered. For example Mij results from the product of the seven notes nkij, for k between 1 and 7.

In sub-step 136, an overall score Nj is calculated. This sub-step aims to determine an overall score reflecting the correlation of the sequence of profiles with each class Cj of the group of classes to allow decision-making at the level of the sequence.

For this, we have the following table which contains the metrics of all the profiles of the sequence for each of the classes of the class group.

Class 1 Class 2 Class 3 Class j Class Ne Profile 1 Mu Profile 2 Profile 3 Profile i Mij Profile Np M | \ | PNC

The calculation of the overall score Nj is for example obtained with the following operator: Π ^Νρ M Ni = ______________

Where j is the class index, i is the profile index and Np is the number of profiles in the sequence.

With this operator, we see that it may be necessary to refocus the values of Mij between 0.2 and 0.8 in order to limit the sensitivity of the overall score to extreme Mij values.

In sub-step 138, the class Cj retained for the target ship 2 is the one which has the highest overall score Nj.

Advantageously, if this highest overall score Nj remains below a rejection threshold, the class Cj is not associated with the target ship 2, and no classification is retained.

In step 140, the classification used for the sequence of profiles is displayed on the operator's screen, for example superimposed on the radar image otherwise obtained and displayed on the screen.

Advantageously, the operator defines in advance one or more interest classes from the group of classes. In step 140, if the class associated with the target ship 2 belongs to one or other of these interest classes, an alarm is generated.

Alternatively, an alarm is issued when the class associated with the target ship 2 does not correspond to the identification data provided by the target ship 2 itself.

The classification by analysis of an energy profile as obtained by the implementation of the process presented above, is low-level information which is much faster to determine and can be applied, at a given time, on the set of radar tracks so as to provide the operator with a first level of classification.

This makes it possible to perform the classification on all the objects detected immediately after their detection and allows alerts to be raised only on criteria requested by the operator.

Thus, despite the variability of the radar signals, the energy profiles contain discriminating information which makes it possible to infer the class to which a target ship belongs.

The morphological parameters identified above have the property of being little sensitive to the high variability of the profiles. A person skilled in the art can develop other morphological parameters from the concepts presented above.

The method according to the invention allows the exploitation of an energy profile preferably obtained in "Static Range Profile" mode, that is to say that the angle of view is known.

It has been noted that the process just presented allows an optimal classification for the angles of view relative to the longitudinal axis of the ship between 7 and 45 °; -7 ° and -45 °; 135 ° and 173 ° and -135 ° and -173 °.

The process also allows an optimal classification of vessels with a length greater than 100 m.

Claims (8)

1. - Method (100) for automatic classification of a target ship (2), characterized in that it is based on a sequence of energy profiles (Pi) delivered by a radar system (8) carried by an aircraft observing said target vessel (2), the sequence comprising at least one energy profile (Pi) indicating, for a direction of observation (Z) of the radar system (8), a backscattered energy as a function of a distance from the radar system, and by what it consists of associating with the target ship (2) a class (Cj) selected from a group of predefined classes as a function of the values (pki) of a plurality of morphological parameters (pk) calculated on each profile of the sequence. 2. - Method according to claim 1, comprising the steps consisting in: - for each energy profile (Pi) of the sequence, calculate (120) the value (pki) of each of the morphological parameters (pk) of a set of predefined morphological parameters; - for each morphological parameter, determine (132) a score (nkij) indicative of a correlation between the value (pki) calculated for said parameter (pk) and a reference value of said parameter (pk) for each class (Cj) of the class group; - for each class (Cj) of the class group, aggregate (134, 136) the scores (nkij) determined for each morphological parameter and for each energy profile (Pi) of the sequence in a global score (Nj); and, - Associate (138) with said sequence the class for which the overall score (Nj) is the highest. 3. - Method according to claim 2, wherein the reference value of a parameter (pk) for a class (Cj) is chosen according to a viewing angle of the target ship (2) by the radar system (8 ). 4. - Method according to claim 2 or claim 3, wherein the class (Cj) for which the overall score (Nj) is the highest is associated with said target vessel (2) only when said overall score is above a threshold of rejection. 5. - Method according to any one of the preceding claims, in which a morphological parameter (pk) of the plurality of morphological parameters is chosen from: - a parameter corresponding to an equivalent radar surface; - a parameter corresponding to a distribution of the equivalent radar surface along the area of the energy profile corresponding to the target vessel; - a parameter corresponding to an average slope along the area of the energy profile corresponding to the target vessel; - a parameter corresponding to a proportion of the area of the energy profile corresponding to the target vessel which is flat; and, - a parameter corresponding to a distance between peaks in the area of the energy profile corresponding to the target vessel. 6. - Method according to any one of the preceding claims, comprising an initial step (110) consisting in rejecting an energy profile of the sequence when a dimension of the target vessel obtained from said energy profile is not consistent with the dimensions obtained from the other profiles of the sequence. 7. - Method according to any one of the preceding claims, in which an operator of the radar system defines a class of interest and the method comprises a step (140) of generating an alert when the class associated with the target ship corresponds to the class of interest. 8. - Radar system (8), characterized in that it is capable of implementing an automatic classification process of a target ship (2) according to any one of claims 1 to 7.