A metal-polymer target recognition and classification system based on infrared broadband spectrum

Abstract

The present invention discloses a metal-polymer target recognition and classification system based on infrared wide spectrum, which belongs to the field of target classification and recognition. It includes: a radiation brightness inversion module for obtaining a detected spectral vector with a target, respectively calculating the response gain vector and radiation bias vector of each band, weightedly calculating the response gain and radiation bias vector of the system, and inverting the radiation brightness vector of the target; a target recognition module for dividing the target's radiation brightness vector into three sub-vectors: shortwave band, medium wave band, and long wave band, respectively inputting them into a trained recognition network to obtain the recognition probability under each band; a fusion classification module for weighted fusion of the recognition probabilities under each band, comparing the fused probability with a set threshold, and if it exceeds, it is a metal material target; otherwise, it is a polymer material target. The present invention breaks through the limitations of interference target recognition, improves the anti-interference ability of the recognition network, and also improves the accuracy of target recognition.

Description

Technical Field

[0001] The present invention belongs to the field of infrared photoelectric imaging target classification and recognition, and more specifically, relates to a metal-polymer target recognition and classification system based on an infrared wide spectrum. Background Art

[0002] Any target on the Earth's surface or in the atmosphere generates spectral signatures derived from the interaction between its own properties and the environment during the process of reflecting, scattering, and transmitting solar radiation, as well as the target's own radiation. Effectively utilizing these signatures in the target's spectral information can effectively improve target detection and recognition performance. For target recognition, relevant research has typically used infrared images. However, for decoy targets, their shape, brightness, and posture in infrared images are essentially identical to those of real targets, significantly reducing the probability of recognition.

[0003] Polymer targets closely resemble real metal targets in terms of appearance, structure, surface coating color, and internal interference heat sources. For example, realistic vehicles and simulated vehicles in film and television dramas achieve the same effect as real vehicles, all made of inexpensive, lightweight polymer materials. However, these simulated vehicles are so similar in appearance that they can be confusing or misleading. Therefore, being able to quickly and effectively identify such targets has become an important research direction, a key requirement for surveillance, reconnaissance, and identification.

[0004] When it comes to identifying metal and polymer targets, domestic and international technicians can only use near-infrared (0.7-2.5 μm) spectra for identification. However, when using near-infrared spectral identification, the target's mid- and long-wavelength spectral information is lost. Moreover, for targets at room temperature, their energy is mainly concentrated in the mid- and long-wavelengths. Therefore, relying solely on near-infrared identification for identification is not accurate enough. Summary of the Invention

[0005] In view of the defects of the prior art, the purpose of the present invention is to provide a metal-polymer target recognition and classification system based on infrared broadband spectrum, aiming to solve the problem of insufficient accuracy of existing recognition methods based on near-infrared band spectrum.

[0006] To achieve the above objectives, the present invention provides a metal-polymer target recognition and classification system based on infrared broadband spectrum, the system comprising:

[0007] The radiance inversion module is used to obtain the detected spectral vector with the target, calculate the response gain vector and radiation bias vector of the shortwave band, mediumwave band, and longwave band respectively, and then weight the response gain vector and radiation bias vector of the system. The system radiance conversion formula is substituted to invert the radiance vector of the target.

[0008] The target recognition module is used to divide the target's radiation brightness vector into three sub-vectors: shortwave band, medium wave band, and long wave band. The sub-vectors are input into the trained recognition network to obtain the recognition probability in each band. The training sample is the radiation brightness sub-vector-target type, and the target type is metal or polymer material.

[0009] The fusion classification module is used to weight the recognition probabilities under each band and compare the fused probability with the set threshold. If it exceeds the set threshold, the target is a metal material target; otherwise, the target is a polymer material target.

[0010] Preferably, the radiation brightness inversion module calculates the response gain vector and the radiation bias vector of the shortwave band, the mediumwave band, and the longwave band respectively by the following method:

[0011] (1) Obtain the Planck curves of different temperature targets and determine the peak point of the Planck curve corresponding to each temperature;

[0012] (2) Compare the wavelength corresponding to each peak point with the range of each wavelength band, and select the relatively high temperature peak point and low temperature peak point in each wavelength band;

[0013] (3) Obtain the radiance of the high-temperature peak point and the low-temperature peak point in each band;

[0014] (4) measuring the target spectral vector corresponding to the relatively high temperature and the target spectral vector corresponding to the low temperature in each band;

[0015] (5) Calculate the response gain vector and radiation bias vector of each band respectively:

[0016]

[0017]

[0018] Among them, u=1,2,3, respectively corresponding to the shortwave band, medium wave band, and long wave band, Gain(λ) u Represents the response gain vector of the u band, Offset(t) u Represents the response gain vector radiation bias vector of the U band, Represent the high temperature and low temperature target spectrum vectors in the u band respectively, They represent the radiation brightness of the high temperature peak point and the low temperature peak point in the u band respectively.

[0019] Preferably, the shortwave band is 1.7 μm-3 μm; the mediumwave band is 3 μm-5 μm; and the longwave band is 5 μm-14 μm.

[0020] Preferably, the radiation brightness inversion module calculates the response gain vector and radiation bias vector of the system by weighting in the following manner:

[0021] Gain(λ)=i·Gain(λ)1+j·Gain(λ)2+k·Gain(λ)3

[0022] Offset(λ)=i·Offset(λ)1+j·Offset(λ)2+k·Offset(λ)3

[0023] Among them, i+j+k=1, and the higher the surface temperature of the object, the greater the proportion of i, and vice versa, the smaller the proportion of i, Gain(λ) u Represents the response gain vector of the u band, Offset(t) u Represents the response gain vector radiation bias vector of the u band, u = 1, 2, 3, corresponding to the shortwave band, medium wave band, and long wave band respectively.

[0024] Preferably, the system radiance conversion formula is as follows:

[0025] DN(λ)=Gain(λ)·S input (λ)+Offset(λ)

[0026] Where DN(λ) is the spectrum vector detected at the target wavelength λ, Gain(λ) and Offset(λ) are the system response gain and system radiation offset at the wavelength λ, respectively; S input (λ) is the image-side radiance of the target.

[0027] Preferably, the system further comprises:

[0028] The radiance offset calculation module is used to average the inverted radiance vector to obtain the averaged intrinsic radiance. The radiance of the target is subtracted from the intrinsic radiance to obtain the radiance offset subvectors of targets in different bands. The subvectors are output to the trained recognition network. The training samples are radiance offset subvectors-target types.

[0029] It should be noted that the present invention preferably uses the above-mentioned radiation brightness offset calculation module to amplify the difference.

[0030] Preferably, the fusion classification module is divided into modes, and performs weighted fusion of the recognition probability under the shortwave band, the recognition probability under the mediumwave band, and the recognition probability under the longwave band:

[0031] In daylight conditions:

[0032] p day =a·p s (λ)+b·p m (λ)+c·p l (λ)

[0033] Among them, p day represents the probability after daytime fusion, p s (λ), p m (λ), p l (λ) represents the recognition probability in the shortwave band, the recognition probability in the mediumwave band, and the recognition probability in the longwave band, respectively; a, b, and c represent the coefficients of shortwave band recognition, medium wave band recognition, and long wave band recognition, respectively, a=0.6, b+c=0.4;

[0034] In night conditions:

[0035] p nig t =d·p s (λ)+e·p m (λ)+f·p l (λ)

[0036] Among them, p nig t represents the probability after night fusion, p s (λ), p m (λ), p l (λ) represents the recognition probability in the shortwave band, the recognition probability in the mediumwave band, and the recognition probability in the longwave band, respectively; d, e, and f represent the coefficients of shortwave band recognition, mediumwave band recognition, and longwave band recognition, respectively, d = 0.1, e + f = 0.9.

[0037] In general, the above technical solutions conceived by the present invention have the following beneficial effects compared with the prior art:

[0038] This paper proposes a metal-polymer target recognition and classification system based on a wide infrared spectrum, effectively distinguishing metal and polymer targets. Spectral recognition effectively deciphers polymer targets and extracts features in key wavelengths (medium and long wavelengths). This overcomes the limitations of interfering target recognition, enhances the recognition network's anti-interference capabilities, and improves target recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a schematic diagram of a metal-polymer target recognition and classification system based on infrared broadband spectrum provided by the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0041] like Figure 1 As shown, the present invention provides a metal-polymer target recognition and classification system based on infrared broadband spectrum, the system comprising:

[0042] The radiation brightness inversion module is used to obtain the detected spectral vector with the target, calculate the response gain vector and radiation bias vector of the shortwave band, mediumwave band and longwave band respectively, and then weight the response gain vector and radiation bias vector of the system. The system radiation brightness conversion formula is substituted to invert the radiation brightness vector of the target.

[0043] The radiation brightness offset calculation module is used to average the inverted radiation brightness vector to obtain the averaged intrinsic radiation brightness, and to subtract the radiation brightness of the target from the intrinsic radiation brightness to obtain the radiation brightness offset sub-vectors of the target in different bands.

[0044] The target recognition module is used to divide the target's radiation brightness vector into three sub-vectors: shortwave band, medium wave band, and long wave band. The sub-vectors are input into the trained recognition network to obtain the recognition probability in each band. The training sample is the radiation brightness offset sub-vector-target type, and the target type is metal or polymer material.

[0045] The fusion classification module is used to weight the recognition probabilities under each band and compare the fused probability with the set threshold. If it exceeds the set threshold, the target is a metal material target; otherwise, the target is a polymer material target.

[0046] Radiosity inversion

[0047] The spectral signal obtained by the photoelectric detection equipment is inverted to obtain the radiation brightness of the target.

[0048] Metallic targets are composed of different materials (from the inside out) with different properties and temperatures, and their thermal infrared images exhibit texture. Polymer targets (composed of air and thin polymer materials) are homogeneous and are isothermal (except for localized heat sources), and their thermal infrared images lack texture information. Different substances have different specific heat capacities, which indicate an object's ability to absorb or dissipate heat. The greater the specific heat capacity of a metallic target, the greater its ability to absorb or dissipate heat, meaning it can exchange more energy with external materials. Due to its large mass, internal heat energy is transferred through its internal structure to the outer shell and radiated outward. Therefore, the total spectrum radiated by a real target includes not only the spectral energy reflected and absorbed by the external world, but also its own radiation spectrum. Polymer targets, on the other hand, lack distributed internal heat sources, and their radiated spectrum only includes the spectral energy reflected and absorbed. Therefore, metallic and polymeric targets differ due to their different components. Metallic targets have a high mass (density) and complex structure, particularly their complex internal metal structure, resulting in a large heat capacity and complex, spatially varying internal thermal and temperature fields, leading to complex and unique energy exchanges with the external world. This characteristic means that polymer materials can only have similar appearances but different spectral properties.

[0049] Taking advantage of the differences in spectral reflection, radiation, and refraction energy across different wavelength bands, a two-point correction method is used to invert the target spectrum using three bands (shortwave, mediumwave, and longwave). Based on the Planck curve, the shortwave band (1.7µm to 3µm) represents the peak concentration point for high-temperature (600°C to 1300°C) blackbodies; the mediumwave band (3µm to 5µm) represents the peak concentration point for medium-high-temperature (100°C to 600°C) blackbodies; and the longwave band (5µm to 14µm) represents the peak concentration point for low-temperature (below 100°C) blackbodies. Therefore, the spectral information collected by the device can be inverted using different temperature pairs. Two-point temperature calibration is based on the assumption that the system's radiation response is linear within its dynamic range. The spectrum of the currently measured object is corrected using a blackbody spectrum that is higher than and a blackbody spectrum that is lower than the current target. For example, if the current measured surface temperature of an object is 700°C, then two-point calibration is to correct the measured spectrum using a blackbody with a temperature higher than 700°C and a blackbody with a temperature lower than 700°C.

[0050] The system radiation brightness conversion formula is as follows:

[0051] DN(λ)=Gain(λ)·S input (λ)+Offset(λ)

[0052] In the above formula, DN(λ) is the digital signal value measured by the photoelectric detection device at the target wavelength λ, Gain(λ) and Offset(λ) are the system response gain and system radiation bias at the wavelength λ respectively; S input(λ) is the image-side radiance of the target.

[0053] Gain(λ)=i·Gain(λ) 短波 +j·Gain(λ) 中波 +k·Gain(λ) 长波

[0054] Offset(λ)=i·Offset(λ) 短波 +j·Offset(λ) 中波 +k·Offset(λ) 长波

[0055] Among them, i+j+k=1.

[0056] In this embodiment, in the shortwave band, the system parameters, system response gain, and system radiation bias are obtained by measuring relatively high-temperature blackbody and low-temperature blackbody:

[0057]

[0058]

[0059] Among them, DN1 000 (λ), S 1000 (λ) are the digital signal value and radiation brightness of a 1000℃ blackbody; DN 800 (λ), S 800 (λ) are the digital signal value and radiation brightness of 800℃ black body respectively.

[0060] In this embodiment, in the medium wave band, the system parameters, system response gain and system radiation bias are obtained by measuring relatively high-temperature blackbody and low-temperature blackbody:

[0061]

[0062]

[0063] Among them, DN 500 (λ), S 500 (λ) are the digital signal value and radiation brightness of a 500℃ blackbody; DN 100 (λ), S 100 (λ) are the digital signal value and radiant brightness of a 100°C blackbody, respectively.

[0064] In this embodiment, in the long-wave band, the system-related parameters, system response gain, and system radiation bias are obtained by measuring relatively high-temperature blackbody and low-temperature blackbody:

[0065]

[0066]

[0067] Among them, DN 80 (λ), S 80 (λ) are the digital signal value and radiant brightness of a black body at 80℃; DN 10 (λ), S 10 (λ) are the digital signal value and radiant brightness of a 10℃ blackbody, respectively.

[0068] Radiance offset calculation

[0069] First, the target's radiance is averaged to obtain the averaged intrinsic radiance curve, and the target's radiance is subtracted from the intrinsic radiance to obtain the offset of the target's radiance.

[0070] In this embodiment, the offset of the target radiation brightness is a one-dimensional vector s(λ)={s(λ1), s(λ2), s(λ3), ..., s(λ n ), n = 1799}, where the offset of the target radiance in the shortwave band is The offset of the target radiance in the medium wave band The offset of the target radiance in the long-wave band

[0071] Target Recognition

[0072] The radiation brightness offset of metal and polymer targets is identified by band. Feature extraction is performed using a convolutional neural network. The network model consists of three one-dimensional convolutional layers, three activation layers, and two maximum pooling layers. The convolutional layer uses a convolution kernel of size 1×K and a step size of L. The recognition results under different bands are obtained: the recognition probability under the shortwave band is P s , the recognition probability in the medium wave band is P m , the recognition probability in the long wave band is P l .

[0073] Fusion classification

[0074] The current time information (daytime / nighttime) obtained from the photoelectric detection equipment is used to determine the corresponding weighting ratio (the target's energy expression in the shortwave, mediumwave, and longwave bands varies under different environmental conditions). Weighted fusion recognition is performed using the offset recognition results of the target's radiant brightness in the shortwave, mediumwave, and longwave bands.

[0075] In daylight conditions:

[0076] p day =a·p s (λ)+b·pm (λ)+c·p l (λ), where a=0.6, b+c=0.4;

[0077] In night conditions:

[0078] p nig t =d·p s (λ)+e·p m (λ)+f·p l (λ), where d = 0.1 and e + f = 0.9.

[0079] Through the above weighting, the final recognition results of metal material targets and polymer material targets are obtained.

[0080] Compare the fused probability with the set threshold. If it exceeds the set threshold, the target is a metal material target; otherwise, the target is a polymer material target.

[0081] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A metal-polymer target recognition and classification system based on infrared broadband spectrum, characterized in that: The system includes: The radiance inversion module is used to obtain the detected spectral vector with the target, calculate the response gain vector and radiation bias vector of the shortwave band, mediumwave band, and longwave band respectively, and then weight the response gain vector and radiation bias vector of the system. The system radiance conversion formula is substituted to invert the radiance vector of the target. The target recognition module is used to divide the target's radiation brightness vector into three sub-vectors: shortwave band, medium wave band, and long wave band. The sub-vectors are input into the trained recognition network to obtain the recognition probability in each band. The training sample is the radiation brightness sub-vector-target type, and the target type is metal or polymer material. The fusion classification module is used to weight the recognition probabilities under each band and compare the fused probability with the set threshold. If it exceeds the set threshold, the target is a metal material target; otherwise, the target is a polymer material target; The radiation brightness inversion module calculates the response gain vector and radiation bias vector of the shortwave band, mediumwave band, and longwave band respectively by the following methods: (1) Obtain the Planck curves of different temperature targets and determine the peak point of the Planck curve corresponding to each temperature; (2) Compare the wavelength corresponding to each peak point with the range of each wavelength band, and select the relatively high temperature peak point and low temperature peak point in each wavelength band; (3) Obtain the radiance of the high-temperature peak point and the low-temperature peak point in each band; (4) Measure the target spectral vector corresponding to the relatively high temperature and the target spectral vector corresponding to the low temperature in each band; (5) Calculate the response gain vector and radiation bias vector of each band respectively: in, = , corresponding to the shortwave band, mediumwave band, and longwave band respectively. express The response gain vector of the band, express The response gain vector of the band is the radiation bias vector, Respectively The high and low temperature target spectrum vectors of the band, Respectively The radiation brightness of the high temperature peak point and low temperature peak point of the band.

2. The system according to claim 1, wherein The shortwave band is 1.7 μm-3 μm; the mediumwave band is 3 μm-5 μm; and the longwave band is 5 μm-14 μm.

3. The system according to claim 1, wherein: The radiation brightness inversion module calculates the system's response gain vector and radiation bias vector by weighting in the following way: in, , and the higher the surface temperature of the object, The larger the proportion, the smaller the The smaller the proportion, express The response gain vector of the band, express The response gain vector of the band is the radiation bias vector, = , corresponding to the shortwave band, mediumwave band, and longwave band respectively.

4. The system according to claim 1, wherein: The system radiation brightness conversion formula is as follows: in, The target wavelength is The spectrum vector detected at The wavelengths are System response gain and system radiation bias at ; is the image-side radiance of the target.

5. The system according to claim 1, wherein: The system further comprises: The radiance offset calculation module is used to average the inverted radiance vector to obtain the averaged intrinsic radiance. The radiance of the target is subtracted from the intrinsic radiance to obtain the radiance offset subvectors of targets in different bands. The subvectors are output to the trained recognition network. The training samples are radiance offset subvectors-target types.

6. The system according to any one of claims 1 to 5, characterized in that The fusion classification module is divided into modes, and the recognition probability under the shortwave band, the recognition probability under the mediumwave band and the recognition probability under the longwave band are weighted and fused: In daylight conditions: in, represents the probability after fusion during the day, They represent the recognition probability in the shortwave band, the recognition probability in the mediumwave band, and the recognition probability in the longwave band, respectively. Respectively represent the coefficients of shortwave band identification, mediumwave band identification, and longwave band identification, ; In night conditions: in, represents the probability after night fusion, They represent the recognition probability in the shortwave band, the recognition probability in the mediumwave band, and the recognition probability in the longwave band respectively; Respectively represent the coefficients of shortwave band identification, mediumwave band identification, and longwave band identification, .

Images