A multi-band selection method for a colorimetric-based infrared detection system

Abstract

The application discloses a multi-band selection method of an infrared detection system based on a color ratio, and belongs to the fields of infrared detection and identification technology and infrared optical system design. Since an infrared detection and identification system of a single band obtains a small amount of information, a target cannot be detected by an infrared detection system or detection accuracy is reduced, multi-band detection and identification is a development trend, but a multi-band selection theory of the infrared detection system is not mature at present. In view of the above problem, the application provides a multi-band selection method of an infrared detection system based on a color ratio. Under the premise of fully considering the influence of detector noise on detection and identification, a target identification model based on a color ratio is provided, and a multi-band selection model of an infrared detector is derived on the basis. The method is suitable for multi-band design of a spectral spectrometer or a filter wheel spectrometer system.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of infrared detection and identification technology, infrared optical system design, and relates to a multi-band selection method of an infrared detection system based on color comparison. BACKGROUND

[0002] The infrared detection system has the advantages of light weight, small volume, low power consumption, strong environmental adaptability, strong anti-interference ability, high resolution, no electromagnetic wave radiation, and the ability to identify disguised targets to a certain extent. Due to these outstanding advantages of the infrared detection system, it has a very wide application in various fields.

[0003] Most of the current mainstream infrared detection systems are single-band detection systems in the middle or long wave band. Due to changes in the working environment of the infrared system, changes in the infrared characteristics of the target, and other reasons, the single-band infrared detection system obtains less information, especially the change of the target itself causes the peak infrared radiation band to move, so that the infrared detection system cannot detect the target or the detection accuracy decreases. Therefore, it is necessary to improve the existing single-band infrared detection system and increase the multi-band detection capability of the infrared system. With the development of detection technology, according to the radiation characteristics of the target, a multi-band detection system will be the development trend, so it is very important to select multiple bands in the infrared spectrum. A typical multi-band detection and identification system is shown in Figure 1 .

[0004] However, the current multi-band selection theory of the infrared detection system is not mature. Common methods include the Monte Carlo method, which has a very large amount of calculation, and other methods designed based on empirical data, which are greatly different from the actual situation and theoretical results due to the oversimplification of the model. For example, the typical band selection method based on the dual-color feature does not consider the influence of the detector noise. SUMMARY

[0005] The purpose of the present application is to solve the problems of the above-mentioned multi-band selection method of the infrared detection system, and to propose a multi-band selection method of an infrared detection system based on color comparison. This method fully considers the influence of the detector noise on the detection, proposes a color target identification model, and deduces a multi-band selection model of the infrared detector based on the model.

[0006] Therefore, the present application proposes a multi-band selection method of an infrared detection system based on color comparison, which specifically includes the following steps,

[0007] Step 1: Create a color target identification model;

[0008] Step 2: Create a multi-band selection model of the infrared detector;

[0009] Step 3: Calculate the optimal multi-band value based on the above two models.

[0010] Further, the specific creation method of the colorimetric target recognition model in step one is as follows,

[0011] According to the blackbody radiation formula, the ratio of the radiation of the target at two detection wave bands i and j is a certain value, where λ il ih , λ jl jh , and λ il ih il ih The ratio is different at different temperatures, and the temperature of the target can be deduced from the ratio. The ratio of the response signals of the two wave bands of the dual-waveband infrared detection system to the target, i.e., the dichromatic ratio, is defined as shown in equation (1), and only relates to the temperature of the target;

[0012]

[0013] where M(λ il , λ ih , T) and M(λ jl , λ jh , T) are the spectral radiance of the wave band i and the wave band j at the temperature T of the target, n i and n j are the detector noises of the wave band i and the wave band j, respectively; it is assumed that the noises follow Gaussian distribution n i ~ N(0, σ i ), n j ~ N(0, σ j ), and σ i and σ j are constants related to the noise level of the detector;

[0014] The equation (1) is expanded near the noise signal, and the first-order term is retained,

[0015]

[0016] After the detection system observes the target and the interference for a period of time, two measurement parameter curves, i.e., the target feature measurement parameter R tar and the interference feature measurement parameter R obj , are obtained; further, mathematical statistics are performed on them to obtain their respective probability distributions;

[0017] It is assumed that the above two probability distribution curves follow Gaussian distribution, and under the given constant false alarm condition, a constant false alarm threshold is used to distinguish the interference and the target; the dichromatic ratio is used as the feature, and the recognition model is further expressed as:

[0018] ​​​​​

[0019] wherein, and respectively represent target and clutter feature measurement parameter probability distribution center position, σ ij (T tar ) represents target feature measurement parameter statistical standard deviation, k is a constant related to constant false alarm, T tar is the temperature of the target, T obj is the temperature of the clutter;

[0020] Further, formula (3) is expressed as:

[0021]

[0022] Since k value is given, R ij (T tar ) represents target and clutter feature measurement parameter probability distribution, and temperature difference ΔT = |T tar -T obj | is the temperature difference of the target and the clutter, therefore the right side part of formula (4) is a constant.

[0023] Further, in step two, the method for creating the multi-band selection model of the infrared detector is as follows,

[0024] It is obvious that optimal band selection is only related to the right side of the equation , and the optimal band selection hopes to be larger and larger;

[0025] For the noise level requirement of formula (2) σ i ,σ j << M(λ il ,λ ih ,T tar ),σ i ,σ j << M(λ jl ,λ jh ,T tar ), under this condition, the target feature measurement parameter statistical standard deviation is:

[0026]

[0027] Finally:

[0028]

[0029] wherein, when the temperature of the target is T tar , the blackbody spectral radiance integrated by band i and band j is M i = M(λ il ,λ ihT tar ) and M j = M(λ il , λ ih , T tar ) ;

[0030] When there are N wave ≥ 2 band optimization designs, the band optimization parameters The band optimization objective function is:

[0031]

[0032] The optimal solution is:

[0033] X * = argmax X {F(X)}, s.t. L(X) = 0 (8)

[0034] Where L(X) = 0 is the constraint equation set of band selection.

[0035] Further, the specific calculation method of step three is as follows,

[0036] When using spectral spectroscopy, the optimization parameter of formula (8) is expressed as Where the blackbody spectral radiance integrated by the band i is respectively:

[0037]

[0038] Assuming the number of false bands N wave = 3, the upper and lower limits of the band are λ max and λ min , respectively, and the three-band design optimization parameter contains two parameters (λ1, λ2) ; Let the noise level σ1 = σ2 = σ3, and the three-band optimization model is converted to, under a given target temperature T tar , the normalized objective function of formula (7) is obtained,

[0039]

[0040] Where M1 = M(λ min , λ1, T tar ), M2 = M(λ1, λ2, T tar ) and M3 = M(λ2, λ max , T tar ) ;

[0041] Under the maximum value of formula (10), the value of (λ1, λ2) is the optimal choice of spectral spectroscopy;

[0042] When using filter wheel spectrometer, assume the number of wave bands N wave =3, one of which is full wave band, the lower limit of which is set as λ 11 =λ min , and the upper limit is set as λ 2h =λ max , under a given target temperature T tar , three wave bands overlap, and the three wave band design optimization parameters include two parameters (λ1, λ2), let σ1=σ2=σ3, the three wave band optimization model is converted into, under a given target temperature T tar , the maximum value of the normalized target function formula (10), the value of (λ1, λ2), wherein M1=M(λ min , λ max , T tar ), M2=M(λ1, λ max , T tar ) and M3=M(λ min , λ2, T tar ).

[0043] The beneficial technical effects of the present application are as follows:

[0044] The method has smaller calculation amount than the Monte Carlo method; and since the influence of the detector noise on the detection and identification is fully considered, the multi-wave band selection of the infrared detection system is more accurate. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 is a typical multi-wave band detection and identification system.

[0046] Figure 2 is a typical target and interference feature measurement parameter curve.

[0047] Figure 3 is a typical target and interference feature measurement parameter probability distribution.

[0048] Figure 4 is the relationship between the three wave band parameters in the spectral spectrometer specific implementation of the present application, and the wave band parameters that need to be optimized.

[0049] Figure 5 is the relationship between the normalized optimized target function F(X) and the wave band parameters (λ1, λ2) in the spectral spectrometer specific implementation of the present application, under the target equivalent temperature 260K. The three wave band optimal design: 5.00-14.00 μm, 5.00-9.70 μm, 10.10-14.00 μm.

[0050] Figure 6For the spectral implementation of the present application, the target equivalent temperature is 330K, the normalized optimized objective function F(X) and the band parameters (λ1, λ2) are related. The three-band optimal design: 5.00-7.78 μm, 7.78-10.76 μm, 10.76-14.00 μm.

[0051] Figure 7 For the spectral implementation of the present application, the target equivalent temperature is 400K, the normalized optimized objective function F(X) and the band parameters (λ1, λ2) are related. The three-band optimal design: 5.00-7.19 μm, 7.19-10.14 μm, 10.14-14.00 μm.

[0052] Figure 8 For the spectral implementation of the present application, the optimal band parameters under different target equivalent temperatures

[0053] Figure 9 For the filter wheel implementation of the present application, the relationship between the three band parameters and the band parameters that need to be optimized.

[0054] Figure 10 For the filter wheel implementation of the present application, the target equivalent temperature is 260K, the normalized optimized objective function F(X) and the band parameters (λ1, λ2) are related. The three-band optimal design: 5.00-14.00 μm, 5.00-9.70 μm, 10.10-14.00 μm.

[0055] Figure 11 For the filter implementation of the present application, the target equivalent temperature is 330K, the normalized optimized objective function F(X) and the band parameters (λ1, λ2) are related. The three-band optimal design: 5.00-14.00 μm, 5.00-8.81 μm, 9.25-14.00 μm.

[0056] Figure 12 For the filter implementation of the present application, the target equivalent temperature is 400K, the normalized optimized objective function F(X) and the band parameters (λ1, λ2) are related. The three-band optimal design: 5.00-14.00 μm, 5.00-8.21 μm, 8.63-14.00 μm.

[0057] Figure 13 For the filter implementation of the present application, the optimal band parameters under different target equivalent temperatures DETAILED DESCRIPTION

[0058] The technical scheme of the present application is: a multi-band selection method of an infrared detection system based on colorimetry, comprising two basic model components, a colorimetric target recognition model and an infrared detector multi-band selection model. The specific steps are as follows:

[0059] (1) Colorimetric target recognition model

[0060] According to the blackbody radiation model, assuming that the target temperature is T, the equivalent radiation area is A, and the influence of atmospheric transmittance is not considered, then the target incident to the entrance pupil of the detection system is:

[0061]

[0062] Wherein, represents the blackbody spectral radiance, λ l and λ h are the upper and lower limits of the band, n is the measurement noise, L is the distance from the target to the detection system, and c1 and c2 are the first and second radiation constants respectively.

[0063] The blackbody radiation follows the Planck formula, and the radiation ratio of two detection bands i and j (λ il ~ λ ih and λ jl ~ λ jh ) at a certain temperature is a certain value. The ratio is different at different temperatures, and the temperature of the target can be inversely deduced from the ratio. Even if the parameters of the two-band detection system are different, the influence of different detection system parameters can be equivalent to the noise signal, so the ratio of the response signals of the two bands of the target by the two-band infrared detection system, i.e. the dichromatic ratio, as shown in formula (2), is only related to the temperature of the target:

[0064]

[0065] Wherein, in order to simplify, it is assumed that the noise n i ~ N(0, σ i ), n j ~ N(0, σ j ), σ i and σ j are constants related to the noise level of the detector. Further, in order to facilitate subsequent discussion, formula (2) is modified as:

[0066]

[0067] Wherein, the noise difference between formula (2) and formula (3) is a constant, so the subsequent discussion is not distinguished.

[0068] Expand formula (3) near the noise signal and keep the first order term:

[0069]

[0070] After observing the target and interference objects for a period of time, the detection system will obtain two measurement parameter curves, namely the target measurement parameter R. tar And interference measurement parameters R obj (See Figure 2 Further mathematical statistics were performed on them to obtain their respective probability distributions (see...). Figure 3 ).

[0071] Assume the two probability distribution curves above follow a Gaussian distribution. Given a constant false alarm rate (CFAR), a CFAR threshold is used to distinguish between interfering objects and the target. This section uses the two-color ratio as a feature, and the recognition model is further expressed as:

[0072]

[0073] in, and σ represents the center location of the probability distribution of the characteristic measurement parameters of the target and the interfering object, respectively. ij (T tar ) represents the statistical standard deviation of the target feature measurement parameters, and k is a constant related to the constant false alarm rate, which is what we often call the threshold, usually 3 to 5.

[0074] Furthermore, equation (5) can be expressed as:

[0075]

[0076] Since the value of k is given, and the temperature difference ΔT = |T tar -T obj | represents the temperature difference between the actual target and the interfering object, therefore the right side of equation (6) is... It is a constant.

[0077] (2) Multi-band selection model for infrared detectors

[0078] The selection of the optimal band is obviously only related to the right side of the equation. Related, and the optimal band selection is desirable The bigger the better.

[0079] Generally, the noise level for the identification problem (4) is typically required to be σ. i ,σ j <<E(λ) il ,λ ih ,T tar ),σ i ,σ j <<E(λ) jl ,λ jh ,T tar), under which the target feature measurement parameter statistical standard deviation is:

[0080]

[0081] Finally:

[0082]

[0083] wherein, when the target temperature is T tar , the blackbody spectral radiance of waveband i and waveband j is M i = M(λ il , λ ih , T tar ) and M j = M(λ il , λ ih , T tar ) respectively. The above expression has an important feature, no matter whether waveband i and waveband j are compared or waveband j and waveband i are compared, the result of formula (8) does not change.

[0084] When there are N wave ≥ 2 wavebands optimized, the waveband optimization parameter is:

[0085]

[0086] The optimal solution is:

[0087] X * = argmax X {F(X)}, s.t. L(X) = 0 (10)

[0088] wherein, L(X) = 0 is a constraint equation set of waveband selection, which can generally be according to the optimization constraint equation of the spectral scheme.

[0089] The meaning of the optimization function F can be further explained as follows: under the condition that the temperature difference between the target and the interference is ΔT, if F·ΔT > k, the target and the interference radiation measurement signals can be distinguished; otherwise, they cannot be distinguished. Under the same condition, the greater the value of the objective function F, the better the recognition ability.

[0090] (3) Optimal waveband calculation embodiment

[0091] There are two kinds of spectral splitting schemes for multi-band infrared detector, one is spectral splitting by grating or prism, the other is spectral splitting by filter wheel. The former is spatial multiplexing, and the bands do not overlap. The latter is time multiplexing, and the bands can overlap. In order to illustrate more details of the present application, two specific examples are given below.

[0092] In the following two examples, some simplifications are made for the convenience of analysis, assuming that the typical equivalent temperature of the target is 260K-400K, and the noise of the detectors in different bands is the same. Moreover, the constraint equation L(X)=0 is hidden according to the spectral splitting scheme.

[0093] (a) Spectral splitting

[0094] When spectral splitting is used, the optimization parameters of formula (10) are expressed as where the blackbody spectral radiance integrated in the band i is respectively

[0095]

[0096] In specific examples, it is assumed that the number of bands N wave =3, and the upper and lower limits of the bands are λ max =14μm and λ min =5μm. According to the spectral splitting scheme, the three bands do not overlap, as shown in Figure 4 Therefore, the optimization parameters of the three bands contain two parameters (λ l , λ h ) (i.e. (λ1, λ2)). Generally, the detector noise is related to the type of detector and the working condition of the detector. In order to simplify, the present application assumes that σ1=σ2=σ3, and the three-band optimization model is converted to, under a given target temperature T tar , the value of (λ1, λ2) when formula (9) is the maximum value of the normalized target function.

[0097]

[0098] From the optimization results, the optimal parameters of multi-band are related to the equivalent temperature of the target. Using spectral splitting, the band is limited to the range of 5μm-14μm. When the equivalent temperature of the target is 260K, see Figure 5 The optimal design of the three bands is 5.00-8.67μm, 8.67-11.51μm, and 11.51-14.00μm. When the equivalent temperature of the target is 330K, see Figure 6, three optimal bands: 5.00-7.78 μm, 7.78-10.76 μm, 10.76-14.00 μm; when the target equivalent temperature is 400 K, see Figure 7 , three optimal bands: 5.00-7.78 μm, 7.78-10.76 μm, 10.76-14.00 μm; when the target equivalent temperature is 400 K, see Figure 8 , three optimal bands: 5.00-7.78 μm, 7.78-10.76 μm, 10.76-14.00 μm; when the target equivalent temperature is 400 K, see

[0099] (b) Filter wheel spectrometry

[0100] When filter wheel spectrometry is used, the number of bands is still assumed to be N wave = 3, and the upper and lower limits of the bands are λ max = 14 μm and λ min = 5 μm. In addition, if the limitation of the detector is not considered, one of the bands will use the full band, i.e., the lower limit of band 1 is set to λ 1l = λ min , and the upper limit is set to λ 2h = λ max . At a given target temperature T tar , the three bands overlap, as shown in Figure 9 The optimal parameters of the three-band design include two parameters (λ l , λ h ). Similar to the spectral spectrometry, let σ1 = σ2 = σ3, and the three-band optimization model is converted into the value of (λ1, λ2) at the maximum value of the normalized target function (equation (9)) at a given target temperature T tar . Wherein, M1 = M(λ min , λ max , T tar ), M2 = M(λ1, λ max , T tar ) and M3 = M(λ min , λ2, T tar ).

[0101] From the optimization results, the optimal parameters of the multi-band are related to the equivalent temperature of the target. When filter wheel spectrometry is used, the bands are limited to the range of 5 μm to 14 μm, and full-band detection is considered to be fully used, when the target equivalent temperature is 260 K, see Figure 10 , three optimal bands: 5.00-7.78 μm, 7.78-10.76 μm, 10.76-14.00 μm; when the target equivalent temperature is 400 K, see Figure 11, three wave band optimal design: 5.00~14.00um, 5.00~8.81um, 9.25~14.00um; when the target equivalent temperature is 400K, see Figure 12 , three wave band optimal design: 5.00~14.00um, 5.00~8.21um, 8.63~14.00um. For the target with equivalent temperature of 260K~400K, see Figure 13 , the center temperature should be selected as the most parameter, i.e. three wave band optimal design: 5.00~14.00um, 5.00~8.81um, 9.25~14.00um.

[0102] Those skilled in the art can understand that the above description is only the preferred embodiment of the present application, and is not intended to limit the present application, and any modification, equivalent replacement and improvement within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A multi-band selection method for an infrared detection system based on colorimetry, characterized in that, Step one, creating a colorimetric target recognition model; Step two, creating an infrared detector multi-band selection model; Step three, calculating the value of the optimal multi-band based on the above two models; In step one, the specific creation method of the colorimetric target recognition model is as follows, According to the blackbody radiation formula, the ratio of the radiation of the target at two detection wave bands and at a certain temperature is a certain value, wherein, , , ; the ratio is different at different temperatures, and the temperature of the target can be inversely deduced from the ratio; the ratio of the response signals of the dual-waveband infrared detection system to the target at two wave bands, i.e. the dichromatic ratio, is only related to the temperature of the target, as shown in formula (1);​ (1) where and are the waveband and waveband spectral radiance at the temperature of the target , and are the waveband and waveband detector noise; assume that the noise is Gaussian distributed , , and are constants related to the level of detector noise; Expand formula (1) near the noise signal and retain the first-order term, (2) After observing the target and interference objects for a period of time, the detection system obtains two measurement parameter curves, namely the target feature measurement parameters. and interfering object characteristic measurement parameters Further mathematical statistics were performed on them to obtain their respective probability distributions; Assuming that the above two probability distribution curves follow Gaussian distribution, under the given constant false alarm condition, the constant false alarm threshold is used to distinguish the interference and the target; using the dual-color ratio as the feature, the recognition model is further expressed as: (3) wherein, and respectively denote target and clutter feature measurement parameter probability distribution center positions, denotes target feature measurement parameter statistical standard deviation, is a constant related to constant false alarm, is a temperature of the target, is a temperature of the clutter. Further, formula (3) is expressed as: (4) Since the values are given, denote the target and clutter feature measurement parameter probability distributions, and the temperature difference is the temperature difference between the target and clutter, so the right side of equation (4) is a constant.

2. The multi-band selection method for a colorimetric-based infrared detection system according to claim 1, wherein, In step two, the creation method of the infrared detector multi-band selection model is as follows, The optimal band selection is obviously related to and the optimal band selection is expected to be as large as possible; Noise level requirements for formula (2) Under this condition, the target feature measurement parameter statistical standard deviation is: (5) Finally: (6) wherein, when the target temperature is the waveband and the waveband have integrated blackbody spectral radiance of and respectively. When there is The waveband optimization parameter The waveband optimization objective function is: (7) The optimal solution is: ， (8) wherein is a set of constraint equations for the wavelength band selection.

3. The multi-band selection method for a colorimetric-based infrared detection system according to claim 1, wherein, The specific calculation method of step three is as follows, When using spectral spectroscopy, the optimization parameter of formula (8) is expressed as ， where the wavelength band The integrated blackbody spectral radiance is (9) Number of pseudo-wavebands , the upper and lower limits of the wavebands are and , the three-waveband design optimization parameters contain two parameters ; let the noise level , the three-waveband optimization model is converted into, under a given target temperature , the normalized objective function of formula (7) is obtained, (10) wherein , and ; The maximum of the formula (10) The value of is the optimal choice for spectral analysis; When using filter wheel splitting, assume the number of wavebands , one of which uses the full waveband, the lower limit of which is set to , and the upper limit of which is set to , at a given target temperature , three wavebands overlap, and the three-waveband design optimization parameters include two parameters , let , the three-waveband optimization model is converted to, at a given target temperature , the maximum value of the normalized objective function formula (10) is , the value of , and .

Images