Fuzzy recognition of missile borne multi-line array infrared detection based on size calculating

Bing-shan Lei ,Jing Li ,Wei-na Hao ,Ke-ding Yan

a School of Mechanical and Electrical Engineering,Xi’an Technological University,Xi’an,710021,China

b School of Electronic and Information Engineering,Xi’an Technological University,Xi’an,710021,China

c Key Laboratory of Education Ministry for Modern Design&Rotor-Bearing System,Xi’an Jiaotong University,Xi’an,710049,China

Keywords: Multi-line array infrared detection Size calculating Custom threshold de-noising Fuzzy comprehensive discrimination algorithm

ABSTRACT In order to improve the infrared detection and discrimination ability of the smart munition to the dynamic armor target under the complex background,the multi-line array infrared detection system is established based on the combination of the single unit infrared detector.The surface dimension features of ground armored targets are identi fied by size calculating solution algorithm.The signal response value and the value of size calculating are identi fied by the method of fuzzy recognition to make the fuzzy classi fication judgment for armored target.According to the characteristics of the target signal,a custom threshold de-noising function is proposed to solve the problem of signal preprocessing.The multi-line array infrared detection can complete the scanning detection in a large area in a short time with the characteristics of smart munition in the steady-state scanning stage.The method solves the disadvantages of wide scanning interval and low detection probability of single unit infrared detection.By reducing the scanning interval,the number of random rendezvous in the infrared feature area of the upper surface is increased,the accuracy of the size calculating is guaranteed.The experiments results show that in the fuzzy recognition method,the size calculating is introduced as the feature operator,which can improve the recognition ability of the ground armor target with different shape size.

1.Introduction

Missile borne infrared detection technology is one of the important research directions of smart munition,which is responsible for the detection and recognition of the infrared band characteristics of armored targets[1].However with the improvement of anti-detection ability of modern armored vehicles,the perception ability of lidar,millimeter wave radar and other active detection methods for armored targets is greatly weakened[2-5].With the unknown detection and recognition model of armored vehicles,there is a multi-line array infrared detection and recognition method that need further investigation.The existing recognition model is dif ficult to solve the quantitative and qualitative calculation model of armored vehicle’s contour size.This problem restricts the development of missile borne multimode composite detection system in the information battle field.With the smart development trend of munition,it is necessary to carry out the research of multi-line array rotation scanning infrared detection system,which has unique technical advantages for the detection and identi fication of armored vehicles.

At present,the feature extraction of armored vehicle’s outline size by missile borne rotary scanning detection system mainly uses laser linear array radar to obtain the point cloud image of the target[6,7].The contour feature of the target is obtained by the edge point signals of the image,and then the classi fication and recognition of the target is realized by the point cloud feature matching algorithm.Although the laser linear array scanning detection can capture the size information of the target well,there are two problems in this detection method.1)Using point cloud data matching algorithm requires high-density point cloud data to obtain the feature points of the target,which requires a long time of data processing,increases the time of missile detection and recognition,thus affecting the viability of intelligent ammunition.2)The point cloud data of the target feature obtained by linear array laser is easy to be interfered by the false target which is similar to the feature of the real target,which leads to the false recognition.Moreover,the real size of the armored target can not be recognized accurately when the target is covered,which affects the anti-jamming ability of the smart munition.Therefore,in order to make up for the shortcomings of linear laser detection,a multi-line array infrared detection model based on contour solution is proposed to improve the fast detection and recognition ability of missile borne multimode composite detection.

Some of the smart munition in service use single unit infrared for target recognition[8,9].Due to the sparse scanning interval,there are missing targets or targets can not be detected quickly.The signal processing method of missile borne infrared detection is usually matched with the known template.Rui Guo[10]proposed a new direct classi fication method,the known feature template categories in the target feature library are matched with the target feature vectors after acquisition and processing respectively,and the category corresponding to the minimum value of all related values is taken as the target category.Yunpei Dang[11]proposed a detection structure,in which multiple detection single units and scanning directions are perpendicular to each other.This detection structure can quickly obtain the feature information of the target.However,on the missile platform,with the falling process,the change of height leads to the change of scanning interval,which affects the contour solution value.Li Zhu[12]proposed an infrared image target region extraction algorithm based on multi-modal feature image fusion.MH Yang[13]proposed a target feature extraction based on fuzzy recognition method.This paper proposes a multi-line array infrared detection system suitable for munition platform.The algorithm of size calculating is introduced to extract the target features.Finally,the dimension information of size calculating is introduced as the characteristic operator of classi fication decision in target recognition,so as to improve the recognition ability of armored vehicles.

2.Multi-line array infrared detection system and size calculating algorithm for munition platform

2.1.The detection principle of missile borne multi-line array infrared rotary scanning system

After the smart munition are separated from the cluster bombs,they enter the steady-state scanning stage at the end of the trajectory to scan and detect the ground targets.The separated smart munition is composed of deceleration and spin reduction device,infrared detection system and explosively formed penetrator.As shown in Fig.1,under the in fluence of aerodynamics,the guide rotor drives the projectile to do self-rotating motion when it falls.

Fig.1.Missile borne multi-line array infrared rotary scanning system model.

The speed of steady-state rotation isnand the steady-state speed isV.The multi-line array infrared detection system is composed of multiple single unit detectors,the number of detection units of the infrared detection system isK,the angle between the detection single units isγ=（2π/K）°.The detection angle between each detection single unit and plumb line direction isθ.When the scanning trajectory covers on the ground,and the scanning interval between the two adjacent detection elements is ΔR.When the scanning trajectory randomly intersects with the armored vehicle,the infrared signal response of the upper surface isS=（s1，s2，…，sn）,Wheresnis the response value of thenth infrared signal.

Through the steady-state scanning motion of the smart munition,the scanning trajectory with the number of detection single units ofKwill scan the upper surface of the target in turn.The scanning radius is decreases with the smart munition falling of constant speed.At timet1,the height of the detection system isH1.The rotate speedn.At this time,there will beS=（s1，s2，…，sK×n）scanning response signal in the same direction.Suppose that at timet1,the initial response signals1intersects the armored target.t2represents the time that the response signals2intersect the armored target.（1/K×n）s.In a period of time,there will beSK×nsignal and target random rendezvous.When the last response signal arrives,calculating the value ofS=（s1，s2，…，sK×n）and the size calculating of the target can be calculated.It is assumed that the starting height of steady-state scanning isH,which is provided by missile borne altimeter.When the movement of armored vehicle is not considered,the scanning interval of two adjacent response signals can be obtained as,

In the formula,N∈（1～K×n-1）.

The scanning path length of thes2response signal in time delay isΔt=（t2-t1）-1/K×ncan be obtained as,

In the formula,Hnindicates the height of the infrared detection system at timetn.When considering the movement of armored vehicles,it is assumed that the running speed of armored vehicles isVt,the scanning interval of two adjacent response signals can be expressed as,

whereΔR*is the scanning interval of the actual coverage area of the target.

2.2.An algorithm for size calculating of armored targets

When the Multi-line array infrared detection scans the target,there is a difference of response time between the starting point to the stopping point.It represents the order in which they detect the target in turn.According to the difference of the time sequence,we can calculate the rendezvous angle of the target in the infrared field of view.When the scanning trajectory random rendezvous with the armored target,there are two typical intersection modes of left oblique rendezvous and right oblique rendezvous between the armored target and the scanning direction.When the scanning trajectory rendezvous with the armored target,the rendezvous angle is generated.

Combining Eqs.(2)and(3),the rendezvous angle can be expressed as,

In the formula,Hnis the height of the infrared detection system at timetn,θis detection angle,1/K×nis inherent time difference.

Fig.2 shows the mathematics model of the width of the armored target.When the scanning trajectory covers the upper surface outline of armored target,it can be seen that the straight line BC represents the width information of armored target.

Fig.2.The mathematics model of the width.

The width dimension of armor target is parallel to the scanning trajectory direction.It can be seen from Fig.2,the real scanning trajectory is arc lengthAB⌒,which can be replaced by the length ofAB.Considering that the rendezvous angle of the target in the detection field of view isψ,it can be calculated that the width dimension parallel to the scanning track direction isBC,and the boundary length ofBCis recorded asC1.C1can be expressed as,

In the formula,Hnis the height of the infrared detection system at timetn,Δτnis the response time length of the signal during the rendezvous ofnscanning tracks and the target.That is the time difference between the end point and the start point.

Fig.3 shows the mathematics model of the length of armored target.When the scanning trajectory covers the upper surface outline of armored target,it can be seen that the straight lineDFrepresents the length information of armored target.

Fig.3.The mathematics model of the length.

The length dimension of armor target is vertical to the scanning trajectory direction.It can be seen from Fig.3,the length ofDFrepresents the boundary length information of the vertical scanning trajectory direction of armored target,which is expressed as length.The solution information of armor target length boundary is determined by the number of response signalsS=（s1，s2，…，sK×n）,length boundary isC2.C2can be expressed as,

3.Signal preprocessing and fuzzy recognition of infrared detection system

3.1.Signal preprocessing of infrared detection system

The interference of signal in complex battle field environment always exists,which makes the output signal of infrared detector impossible to keep stable all the time.Due to the sudden change of response voltage caused by interference,the target recognition system will produce false triggering,so it is necessary to shield the relevant interference signals in the environment.The amplitude of the output signal does not change with the size of the target.Three signal characteristics,waveform width,waveform energy and maximum rise slope are selected as sample eigenvalues.According to the signal characteristics of armored target,wavelet analysis method is selected to preprocess the signal[14-16].Firstly,the output signal of the sensor is transformed by multi-scale wavelet transform,from time domain to wavelet domain,then the wavelet coef ficients of each scale are extracted,and the wavelet coef ficients belonging to noise signal are removed or weakened as much as possible.Finally,the wavelet signal is reconstructed by inverse wavelet transform,which is the signal after de-noising[17].

Generally,it can be found that the wavelet coef ficients of effective signals are higher than those of noise after wavelet decomposition.Based on the critical threshold,the effective signal is a wavelet coef ficient larger than the threshold,which is reserved.The signal is a wavelet coef ficient less than this threshold,which is considered to be caused by noise and eliminated.

The hard threshold denoising function is expressed as,

In the formula,wj，kand^wi，jis the wavelet coef fi cients before and after signal denoising.The value of thresholdλisσ2lg（N）,σ=median/0.6745 is the estimation of noise level.

The soft threshold denoising function is expressed as,

The schematic diagram of these two threshold functions is shown in Fig.4.The traditional hard threshold has discontinuities at±λ,and the existence of discontinuities easily leads to signal oscillation[18].Although the traditional soft threshold has no discontinuity and the signal is smooth,there is always a fixed deviation between the soft threshold and the asymptote[19].This will lead to a deviation between the reconstructed signal and the real signal that can never be eliminated,so the denoised signal can not correspond to the real signal.

Fig.4.Comparison between hard threshold and soft threshold.

Based on the characteristics of hard threshold function and soft threshold function,a custom threshold de-noising function is proposed and can be expressed as,

In the formulaμ=1-e-α（|wj，k|-λ）2,λ=aandbis a real constant greater than 1,μmakes the hard threshold function and the new function sign（wj，k）×is linear combination.When→±λ,then→0.That is,is not disconnected at the position of segment point=±λ.Whenis increased,is in fi nitely close to the real wavelet coef ficient.When→∞,thenμ→0 and.

Fig.5 shows the schematic diagram of the custom threshold function.The signal reconstruction of the custom threshold function has no signal oscillation when the hard threshold function is de-noising.At the same time,it also reduces the deviation from the real wavelet coef ficients to a greater extent,so as to eliminate the phenomenon of fixed deviation.

Fig.5.The custom threshold function.

3.2.Fuzzy recognition of infrared detection system

The fuzzy classi fication algorithm is selected because the target characteristic signal detected by the detector shows certain fuzziness.When the detection system is at a certain detection height,there is no clear boundary among waveform width,maximum rise slope and waveform energy,which is affected by the azimuth and offset of the detection system.In addition,there is background environment interference,which makes the soft boundary represented by fuzzy function more reliable than geometric boundary.

Assuming that there are three kinds of target’s characteristic databases in different height of detection system.At a certain detection height,the mean vector and standard deviation vector of classitargets are expressed asM（i）respectively.The feature of target to be identi fied is recorded asFXand its expression is

Assume this target belongs to classitargets,due to the existence of various factors,FXcan not be equal to the mean vectorM（i）of classitargets.Instead,it should be concentrated aroundM（i）in the form of Gaussian distribution.Each component ofFXshould be concentrated near its ownmij,but it may also be nearmkj（k≠j）.As a result of this state,the uncertainty of target recognition is brought.

Taking trapezoidal membership function as the sample,the membership degreeμijbelonging to the class i targets and can be expressed as,

In the formula,vijrepresents the elements in the standard deviation vectorV（i）,μijis between 0 and 1.The membership degree within the 1 time standard deviation is 1.When the deviation is 1-2 times of standard deviation,the membership degree decreases linearly.

4.Calculation and experiment analysis

4.1.Calculation analysis of size calculating

In order to verify the algorithm of multi-line array infrared detection,3 m×1.5 m heating box is used as the simulation target heat source andΔR*=0.6m.The structure diagram of the detection system is shown in Fig.6.The three unit detection devices scan the ground target at an angle of 120°.The detection system is placed on the test tower,and the detection is carried out at different height of the tower.

Fig.6.Structure of multi-line array infrared detection system.

When the rendezvous angle of the target in the field of view of the detection system isψ=90°,ψ=45°andψ=0°.In view of these three rendezvous angles,correlation analysis is made.When ψ=90°,the scanning trajectory direction of the infrared detector is vertical to the long edge of the target contour.The number of scanning trajectory passing through the target is five signals,as shown in Fig.7.

Fig.7.Signal response when the rendezvous angle is 90°.

Fig.8.Signal response when the rendezvous angle is 45°.

Whenψ=45°,the scanning trajectory direction of infrared detector rendezvous both the long side and the short side of the target.Part of the scanning trajectory passing through the target is just the diagonal of the target contour.The number of scanning trajectory passing through the target is six signals,as shown in Fig.8.

Whenψ=0°,the scanning trajectory direction of the infrared detector is parallel to the long edge of the target contour.The number of scanning trajectory passing through the target is three signals,as shown in Fig.9.

Fig.9.Signal response when the rendezvous angle is 0°.

Based on simulation signal,and obtain the signal response data and parametersC1andC2calculated by contour solution under different field of view angles,and record them in Table 1.

Analyze the data in Table 1,and get the error rate ofC1andC2as shown in Fig.10.

Table 1Size calculating values of different rendezvous angles.

Fig.10.Error rate of size calculating values.

AsC1is directly calculated from the pulse width of the rendezvous signal of the field of view and the target,it is a direct response to the target size,so its error rate is below 5%.C2is a parameter obtained by combining the density of scanning interval with the corresponding spatial geometric relationship,and its error rate is closely related to the density of scanning interval.C2is below 28.14%.

4.2.Experimental analysis of fuzzy comprehensive discrimination algorithm

The three characteristics of the response signal after de-noising are used as the basic characteristics:the waveform width,the waveform energy and the maximum rise slope.At the same time,the target dimension informationC1andC2obtained by size calculating are introduced as supplementary eigenvector operators.When the weight distribution of feature quantity is different,the recognition rate is different.Fig.11 shows three different simulated heat source targets(A,B and C)are taken as the detection target,and their dimensions are 3 m×4 m,3 m×6 m and 4 m×7 m respectively.

Fig.11.Three simulated targets of different size.

The experiments are carried out every 10 m in the range of 60~30 m.The average value of the data in each height range is calculated by mean function as the clustering center of each category.The data is shown in Table 2.

Table 2 Waveform eigenvalue template data.

B is the real target to be identi fied and A and C are the false targets to be interfered.According to the waveform template data of the target,under the condition of different detection height,the three characteristic operators and five characteristic operators are compared and analyzed.The fuzzy classi fication algorithm is used to analyze the comprehensive membership degree of the three targets.

When the weights of three characteristic operators are equal,the membership weights of waveform width,maximum rise slope and waveform energy are 0.3,0.3 and 0.3 respectively.The detection system is located at different detection heights,and the comprehensive membership degree changes with the number of samples,as shown in Fig.12.

As shown in Fig.A,the height range of 40~30 m,B real targets can be distinguished.However,as shown in Fig.B,the height range of 60~50 m,with the increase of height,B true target and false target cannot be distinguished well.

WhenC1andC2are introduced as supplementary eigenvector operators,the weights of the two eigenvectors are equal and the complementary operators account for 40%of the comprehensive weight ratio.The other three characteristic operators for 60%of the comprehensive weight ratio,and the other three features have the same weight.The weight of membership grade is 0.2,0.2,0.2,0.2 and 0.2 respectively,as shown in Fig.13.

Fig.12.Membership grade of three characteristic operators.

Fig.13.Membership grade of five characteristic operators.

Fig.14.Membership grade of five characteristic operators with different membership weight.

TheC1represents the signal response of the target size,which has a higher weight.TheC2is combined from the number of rendezvous and the spatial relationship.C1:C2is set to 7:3.At the same detection height,the signal pulse width has a great in fluence on target feature recognition.The membership weight of waveform width,waveform energy,maximum rise slope,C1andC2are 0.24,0.18,0.18,0.28 and 0.12 respectively.The change of the membership grade with the number of samples at different detection heights is obtained,as shown in Fig.14.

Increasing the membership weight ofC1and waveform width can reduce the membership grade of false target C(4 m×7 m).With the introduction ofC1andC2supplementary eigenvector operators,B can be distinguished from A and C in the range of 40~30 m and 60~50 m.

5.Conclusions

In this paper,the target size feature obtained by the size calculation method is applied as the feature operator.Through the fuzzy recognition method,the signal response feature and the size calculation of armored target are judged by fuzzy classi fication.In the aspect of signal preprocessing,according to the characteristics of the detection system,a custom threshold de-noising function is proposed.This filtering method avoids the disadvantages of hard threshold and soft threshold and has a better denoising ability.Finally,the validity of the model is veri fied by simulation and experiments,and the error rate ofC1is below 5%,and that ofC2is below 28.14%.When the membership weight ofC1and waveform width is increased,the weight of membership is 0.24,0.18,0.18,0.28 and 0.12 respectively.The five feature quantities can distinguish the true target well in different height.

Declaration of competing interest

We declare that we do not have any commercial or associative interest that represents a con flict of interest in connection with the work submitted.

Acknowledgment

This work was supported by the National Natural Science Foundation of China(No.11804263)and the Program for Innovative Science and Research Team of Xi’an Technological University.