Physical sample generation method for resisting spectral recognition
By extracting spectral features and performing difference analysis, the optimal hybrid filter material is selected to coat the target to be encoded, generating adversarial spectral recognition samples. This solves the problem of insufficient robustness of spectral images and enhances the security of spectral target detection.
Patent Information
- Application Number
- CN202511520246.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-23
- Publication Date
- 2026-01-09
AI Technical Summary
The lack of adversarial example generation methods for spectral images in existing technologies leads to insufficient robustness of spectral target detection networks and potential compromise of data privacy.
By acquiring spectral image data of the target to be encoded and the background, extracting spectral features and performing difference analysis, selecting material systems with different filtering effects, calculating the optimal mixed filtering material, coating it on the target to be encoded, making its spectrum close to the set background, and generating physical samples for adversarial spectral recognition.
It enables the generation of adversarial spectral recognition samples at the physical level, which can deceive machine learning models, reduce the risk of being attacked, and enhance the robustness of spectral target detection networks.
Smart Images
Figure CN121305263A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of spectral imaging and adversarial sample generation technology, in particular to a physical sample generation method for adversarial spectral recognition. BACKGROUND
[0002] Spectral imaging provides rich data in spectral and spatial dimensions, and is therefore widely used in various fields. Target detection technology based on spectral images has attracted increasing attention because it can provide more comprehensive information than traditional RGB images, thereby significantly improving the accuracy of target detection. However, this detection method can compromise data privacy. In addition, there is no established method to evaluate the robustness of this technology.
[0003] Adversarial sample generation technology is a subfield of computer science that aims to deceive machine learning models, particularly neural networks, by making small but meaningful changes to input data, causing them to produce incorrect outputs or misclassifications. Studying these techniques specifically for spectral images can improve data security while enhancing the robustness of spectral target detection networks. Adversarial sample generation technology is rapidly evolving. Since Szegedy et al. proposed adversarial samples, Goodfellow et al. explained the basic principles behind adversarial samples in 2014. Based on this research, a series of adversarial sample generation methods have emerged: 1. White-box attack: In these scenarios, the attacker has complete knowledge of the target model, including its structure and parameters, allowing them to generate adversarial samples directly on the target model. 2. Black-box attack: Conversely, the attacker lacks knowledge of the target model's internal structure and parameters and must rely on input-output correspondence to generate adversarial samples. 3. Targeted and untargeted attacks: Targeted attacks aim to induce specific misclassifications, while untargeted attacks seek to cause misclassifications without specifying a particular class.
[0004] Adversarial sample attacks have been more widely studied in the field of target detection, which aims to deceive or disrupt the normal operation of target detection systems, causing misclassifications or missed detections. However, current adversarial sample generation techniques for spectral images are still lacking, mainly due to the limitations of attacking image modalities. The main target recognition adversarial techniques are designed for RGB images, which are challenging to adapt to the increasingly widespread application of spectral imaging technology. SUMMARY
[0005] The present application aims to at least solve one of the above technical problems existing in the prior art.
[0006] To this end, the present application provides a physical sample generation method for adversarial spectral recognition.
[0007] The physical sample generation method for adversarial spectral recognition proposed by the present application comprises: Spectrum image data of a target to be coded and a set background are acquired, spectrum features of the target to be coded and the set background are extracted respectively, and difference analysis is performed; A material system with different filtering effects is selected, spectrum difference between the target to be coded and the set background is referred to, material ratio calculation is performed, optimal ratio coefficients of each material are determined, and optimal mixed filtering materials are obtained; the optimal mixed filtering materials are used to coat the target to be coded, so that the spectrum of the target to be coded is as close as possible to the spectrum of the set background. The target to be coded is coated based on the optimal mixed filtering materials; Target recognition is performed based on the target to be coded after coating, whether the recognition accuracy is reduced to an expected value is judged, and when the recognition accuracy reaches the expected value, the countermeasure design is completed, otherwise, the coding scheme is adjusted.
[0008] According to the physical sample generation method for spectrum identification countermeasure of the technical scheme of the present application, the following additional technical features can also be provided: In the above technical scheme, the set background is a background where the target to be coded is located or a set simulation target.
[0009] In the above technical scheme, the set background includes a plurality of objects or components; Each object or component has a corresponding optimal mixed filtering material, that is, each optimal mixed filtering material is used to coat the target to be coded, so that the spectrum of the target to be coded is as close as possible to the spectrum of the corresponding object or component.
[0010] In the above technical scheme, the spectrum image data of the target to be coded and the set background are acquired, the spectrum features of the target to be coded and the set background are extracted respectively, and difference analysis is performed, including: The spectrum curves of the target to be coded and the set background are extracted respectively; The spectrum curve difference between the target to be coded and the set background is compared.
[0011] In the above technical scheme, the material system with different filtering effects is selected, including: Within a set spectrum band range, a plurality of spectrum modulation materials are selected to form a spectrum modulation material set; As an objective, the transmittance spectrum curves of the spectrum modulation materials are as unrelated as possible, a plurality of optimal spectrum modulation materials are selected from the spectrum modulation material set to form a material system.
[0012] In the above technical scheme, as an objective, the transmittance spectrum curves of the spectrum modulation materials are as unrelated as possible, a plurality of optimal spectrum modulation materials are selected from the spectrum modulation material set to form a material system, including: The K-Means algorithm is used to select several optimal spectral modulation materials from the set of spectral modulation materials to form a material system; wherein the spectral modulation angle is used as a distance measure to cluster the spectral curves of all spectral modulation materials in the set of spectral modulation materials, so as to maximize the similarity within the class and minimize the similarity between the classes, and several clustering centers are obtained through iteration, and the target number of optimal spectral modulation materials is selected based on the clustering centers to form the material system.
[0013] In the above technical solution, the reference to the target to be encoded and the set background spectrum difference is used to perform material ratio calculation, determine the optimal ratio coefficient of each material, and obtain the optimal mixed filter material, comprising: A material light transmission matrix is constructed, which is used to represent the light transmission rate of each material in each channel; A mixing coefficient vector is constructed, which is used to represent the mixing ratio of each material in the mixed filter material; The transmittance of the mixed filter material is multiplied by the spectrum of the target to be encoded to obtain the camouflage target spectrum; Based on the objective function, the camouflage target spectrum is linearly regressed with the set background spectrum to solve the optimal ratio coefficient of the mixed material, so that the camouflage target and the set background have the best linear relationship.
[0014] In the above technical solution, the objective function is:
[0015] Wherein, N represents the number of materials; C represents the number of channels; represents the ratio coefficient of the i-th material; represents the transmittance of the i-th material in the channel j; represents the spectral intensity of the set background spectrum in the channel j, represents the spectral intensity of the camouflage target spectrum in the channel j; represents the weight of the j-th channel, which is used to adjust the importance of the channel.
[0016] In the above technical solution, the spectral modulation materials are all organic dyes, which can be dissolved in organic solvents; The organic solvent includes photoresist, developer or ethanol; Each spectral modulation material in the material system is dissolved in the organic solvent according to the determined optimal ratio coefficient to obtain the mixed filter material; Based on the mixed filter material corresponding to each object or component in the set background, a mixed filter material group is formed.
[0017] In the above technical solution, the target to be encoded is coated based on the optimal mixed filter material, comprising: Design the spatial arrangement mode of each mixed filter material based on the mixed filter material group; wherein, different coating space combinations are performed on each mixed filter material; When the recognition accuracy is not reduced to the expected value, the adjustment of the coding scheme includes adjusting the mixed filter material ratio and / or coating space position.
[0018] In summary, due to the adoption of the above technical features, the beneficial effects of the present application are: The present application is based on spectral feature extraction and analysis of the target to be coded and background acquisition data, according to the physical target to be coded and its background (including several different objects), acquires a data set, extracts the spectral curve and feature vector of each substance and performs difference analysis; realizes spectral domain coding confrontation based on the filter material system, uses a material system with different filtering effects, refers to the spectral difference between the physical target to be coded and the background, calculates the ratio of each material, and determines the optimal ratio coefficient of each material; further realizes the joint optimization of space and spectrum and the coating of the material physical layer, respectively configures the optimal filter material for each object, forms a mixed filter material group (containing the same number of background) of the target to be coded and the background, and performs different spatial arrangement combinations and coating on the mixed filter material group; finally, based on the change of recognition rate, the spectral material ratio and the coating space position of each material in space are optimized, and the physical layer spectral recognition confrontation sample generation is realized. The method proposed in the present application realizes breakthrough progress in customized generation of spectral image physical confrontation samples, and the generated results can effectively deceive machine learning models and reduce the risk of being attacked.
[0019] Additional aspects and advantages of the present application will become apparent in the light of the following description and accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS
[0020] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which: Figure 1 is a flow chart of a physical sample generation method for spectral recognition confrontation according to an embodiment of the present application; Figure 2 is a comparison diagram of the spectral curve of the target to be coded and the background according to an embodiment of the present application; Figure 3 is a comparison diagram of the transmittance curve of several spectral modulation materials according to an embodiment of the present application; Figure 4 is a principle diagram for calculating the ratio coefficient of the mixed filter material according to an embodiment of the present application; Figure 5 is a comparison diagram of the spectral curve of the target to be coded 1 before and after coding according to an embodiment of the present application; Figure 6is a spectral curve comparison schematic diagram of a to-be-encoded target 2 before and after encoding in an embodiment of the present application; Figure 7 is a flowchart of verification of an adversarial effect in an embodiment of the present application; Figure 8 is a flowchart of application of a physical sample generation method for adversarial spectral recognition in an embodiment of the present application. DETAILED DESCRIPTION
[0021] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application will be further described in detail below with reference to the drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.
[0022] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, however, the present application can also be implemented in other different manners from those described herein, therefore, the protection scope of the present application is not limited by the specific embodiments disclosed below.
[0023] Some embodiments of the present application will be described below with reference to Figures 1 to 8 to provide a physical sample generation method for adversarial spectral recognition.
[0024] Some embodiments of the present application provide a physical sample generation method for adversarial spectral recognition.
[0025] As shown in Figure 1 , a first embodiment of the present application proposes a physical sample generation method for adversarial spectral recognition, including the following steps S1 to S4.
[0026] S1, obtaining spectral image data of a to-be-encoded target and a set background, respectively extracting spectral features of the to-be-encoded target and the set background and performing difference analysis.
[0027] Specifically, the to-be-encoded target can be any selected target, such as common objects in life, military models, etc. The set background is the background where the to-be-encoded target is located or a set simulation target. It can be understood that the set background includes a plurality of objects or components; taking a military model in an external environment background as an example, the background can include green grassland, yellow grassland, trees, sand soil, etc.
[0028] In some embodiments, step S1 includes: respectively extracting spectral curves of the to-be-encoded target and the set background; comparing the spectral curve differences of the to-be-encoded target and the set background. Specifically, when performing difference analysis, clustering analysis or principal component analysis algorithms based on machine learning can also be used.
[0029] In one embodiment of the present disclosure, two groups of objects to be coded and setting backgrounds are selected for the purpose of demonstration, as shown in FIG. 1. The object to be coded in the first group is a metal tank model (red solid line box), and the setting background is a plastic tank model (red dashed line box). The object to be coded in the second group is a real plant (blue solid line box), and the setting background is a plastic plant (blue dashed line box). Figure 2
[0030] The spectral curves of the two groups are extracted respectively, Figure 2 The red spectral curve is the spectral curve of the tank group (the first group), in which the red solid line represents the metal tank model, and the red dashed line represents the plastic tank model. The blue spectral curve is the spectral curve of the plant group (the second group), in which the blue solid line represents the real plant, and the blue dashed line represents the plastic plant.
[0031] The spectral curve difference between the object to be coded and the setting background in each group is analyzed. The present disclosure takes the ratio of the spectral curves of the object to be coded and the setting background as an example. Similarly, the blue ratio curve is the comparison result of the spectral curve difference between the object to be coded and the setting background in the plant group, and the red ratio curve is the comparison result of the spectral curve difference between the object to be coded and the setting background in the tank group.
[0032] It can be understood that the above comparison process is only a schematic representation. In actual application, since the setting background includes multiple objects and components, the spectral curve difference between the object to be coded and each object or component needs to be obtained through difference analysis.
[0033] S2, a material system with different filtering effects is selected, and the spectral difference between the object to be coded and the setting background is referred to for material ratio calculation to determine the optimal ratio coefficient of each material, so as to obtain an optimal mixed filtering material. The optimal mixed filtering material is used to coat the object to be coded, so that the spectrum of the object to be coded is as close as possible to the spectrum of the setting background.
[0034] In some embodiments, the selection of the material system with different filtering effects includes: In the setting spectral band range, a plurality of spectral modulation materials are selected to form a spectral modulation material set. The transmittance spectral curves of the spectral modulation materials are as unrelated as possible, and a plurality of optimal spectral modulation materials are selected from the spectral modulation material set to form the material system.
[0035] Specifically, according to the spectral curves of different materials, appropriate spectral modulation materials are selected, Figure 3 A contrast diagram of transmittance curves of different spectral modulation materials is shown. The material system selection scheme should contain the characteristics of each waveband in the identification interval (set spectral waveband range) as much as possible, and the transmittance spectral curve should be as unrelated as possible, so it has better spectral expression ability. Here, the modulation effect is measured by analyzing the absorption characteristics of the material for different waveband light. If a certain material has a more obvious absorption for a certain waveband light and the absorption rate is high, and the absorption rate for the waveband with no obvious absorption effect is low, it is considered to have a better modulation effect.
[0036] In one specific embodiment, the transmittance spectral curve of each spectral modulation material is as unrelated as possible, and a number of optimal spectral modulation materials are selected from the spectral modulation material set to form a material system, including: The K-Means algorithm is used to select a number of optimal spectral modulation materials from the spectral modulation material set to form a material system; wherein K=10, the spectral modulation angle (SAM) is used as the distance measure, the spectral curves of all spectral modulation materials in the spectral modulation material set are clustered, the intra-class similarity is maximized and the inter-class similarity is minimized, and 10 cluster centers are obtained by iteration. Based on the cluster centers, the target number (not greater than the number of cluster centers, usually equal to) of optimal spectral modulation materials is selected to form a material system. The number of optimal spectral modulation materials in the material system (numerically equal to K, i.e. the number of cluster centers) can be adjusted as needed.
[0037] In order to realize the generation of adversarial samples, the target to be encoded needs to be made more similar to the set background, so as to reduce the recognition rate of the spectral recognition network for the main target (the target to be encoded). Therefore, the optimal material ratio needs to be calculated based on the selected material system.
[0038] In some embodiments, the spectral difference between the target to be encoded and the set background is used to calculate the material ratio, determine the optimal ratio coefficient of each material, and obtain the optimal mixed filter material, including: A material light transmission matrix is constructed, which is used to represent the light transmission rate of each material in each channel; A mixing coefficient vector is constructed, which is used to represent the mixing proportion of each material in the mixed filter material; The transmittance of the mixed filter material is multiplied by the spectrum of the target to be encoded to obtain the camouflage target spectrum; Based on the objective function, the camouflage target spectrum is linearly regressed with the set background spectrum to solve the optimal ratio coefficient of the mixed material, so that the camouflage target and the set background have the best linear relationship.
[0039] Specifically, the objective function is:
[0040] where N denotes the number of material types; C denotes the number of channels; denotes the mixing coefficient of the i-th material; denotes the transmittance of the i-th material in channel j; denotes the spectral intensity of the set background spectrum in channel j, denotes the spectral intensity of the camouflage target spectrum in channel j; denotes the weight of the j-th channel, which is used to adjust the importance of the channel.
[0041] The objective function includes two parts: one part is used to measure the spectral similarity between the camouflage target and the set background, and the other part is used to ensure that the mixing coefficients of the mixed materials are non-negative. The goal of the optimization process is to minimize this objective function to obtain the optimal mixing coefficients of the mixed materials.
[0042] In one specific embodiment, as shown in Figure 4 , assuming that N different optimal spectral modulation materials are selected in the material system, the transmittance of each optimal spectral modulation material in each channel can be represented by an NxC material transmittance matrix, where N denotes the number of material types and C denotes the number of channels. At the same time, a 1xN mixing coefficient vector is introduced, which represents the mixing proportion coefficient of each optimal spectral modulation material, and these coefficients are to be calculated. Therefore, the transmittance of the mixed filter material in each channel can be represented by a 1xN mixing coefficient vector. In specific operation, we multiply the transmittance of this mixed filter material with the spectrum of the real target (the target to be encoded) to obtain a new camouflage target spectrum. Then, we perform linear regression on the camouflage target spectrum and the spectrum of an object or component in the set background to solve the optimal coefficients of the mixed filter material, so that the camouflage target and the background have the best linear relationship.
[0043] It can be understood that each object or component has a corresponding optimal mixed filter material, i.e., each optimal mixed filter material is used to coat the target to be encoded, so that the spectrum of the target to be encoded is as close as possible to the spectrum of the corresponding object or component.
[0044] As shown in Figure 5 and Figure 6 , the spectral curve comparison results of the target to be encoded (a metal tank model and a real plant) before and after encoding and the set background (a plastic tank model and a plastic plant) are shown. Repeat the above process to calculate the optimal coefficients of the mixed filter material corresponding to each object or component in the target to be encoded and the set background. Figure 5 and Figure 6 , it can be seen that the spectral curve of the target after encoding is significantly close to the spectral curve of the set background.
[0045] S3, coating and coding the target to be coded based on the optimal mixed filter material.
[0046] After determining the optimal coefficient of the mixed filter material through step S2, the corresponding optimal mixed filter material needs to be configured, and then a mixed filter material group is formed, which includes all mixed filter materials corresponding to each object or component in the set background. Specifically, the spectral modulation material is an organic dye, which has excellent solubility in an organic solvent; the organic solvent includes photoresist, developer, or ethanol, etc.; each spectral modulation material in the material system is dissolved in the organic solvent according to the determined optimal proportioning coefficient to obtain the mixed filter material.
[0047] In some embodiments, based on the mixed filter material group obtained above, the coating and coding of the target to be coded based on the optimal mixed filter material includes: Based on the mixed filter material group, the spatial arrangement mode of each mixed filter material is designed; wherein, different coating space combinations are performed on each mixed filter material to realize joint optimization coding of the spatial domain and the spectral domain, and physical layer coating is performed. The spatial arrangement includes but is not limited to combined shape (lattice, strip, net, etc.), adjustment of material concentration, adjustment of position relationship, and the specific spatial arrangement mode can be configured or adjusted by human or randomly generated by machine, and the best solution is sought through subsequent feedback adjustment.
[0048] S4, target recognition based on the target to be coded after coating and coding, judging whether the recognition accuracy is reduced to the expected value, and completing the countermeasure design when the recognition accuracy reaches the expected value, otherwise, adjusting the coding scheme.
[0049] Specifically, the present disclosure uses a traditional spectral recognition module to respectively recognize the original target to be coded scene and the coded target scene, and observes the change of recognition rate. Among them, the spectral recognition module can use traditional spectral recognition algorithm, machine learning or deep learning, etc. Recognition method.
[0050] In one specific embodiment, as shown in Figure 7 After constructing the recognition model, the recognition model can be trained based on the unencoded scene, after training, the scene where the coded target is located is collected, and the trained recognition model is used for target recognition to analyze the influence of the coding method of the present disclosure on the recognition rate.
[0051] In Figure 7 the embodiment shown, the recognition model can accurately distinguish the original target to be coded and the set background, and frame them with dashed line and solid line respectively, while it is difficult to distinguish the target and the set background for the coded target, which proves that the current coding scheme meets the requirements and effectively reduces the recognition accuracy.
[0052] When the recognition accuracy is not reduced to the expected value, the adjustment coding scheme comprises adjusting the mixed filter material ratio and / or coating space position, i.e. further optimizing the spectral domain mixed material ratio scheme and the coating space position change scheme, and repeating the attempt of new space-spectrum joint optimization coding on this basis to achieve better coding effect.
[0053] Specifically, Figure 8 The actual application flowchart of the coding method of the present disclosure is shown, i.e. first, spectral measurement of the target to be coded and the background is carried out; then, feature distribution model establishment and material ratio calculation are carried out, i.e. analysis of spectral feature difference of the target to be coded and the background and calculation of optimal mixed filter material ratio coefficient, corresponding to the content of step S2; then, joint coding of space domain and spectral domain is carried out; the coding result is subjected to target recognition and recognition accuracy determination, and when the recognition accuracy is not reduced to the expected value, the space-spectrum coding mode is optimized, and recoding is carried out until the recognition accuracy is reduced to the expected value, and the countermeasure design is completed.
[0054] In this specification, the illustrative representations of the above-mentioned terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.
[0055] Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.
Claims
1. A method of generating physical samples to counter spectral identification, characterized in that, The application relates to an anti-counterfeiting method and device. The application comprises the following steps: Spectrum image data of a target to be coded and a set background are acquired, spectrum features of the target to be coded and the set background are extracted respectively, and difference analysis is performed; A material system with different filtering effects is selected, material ratio calculation is performed according to spectrum difference between the target to be coded and the set background, optimal ratio coefficients of each material are determined, and optimal mixed filtering materials are obtained; the optimal mixed filtering materials are used for coating the target to be coded, so that the spectrum of the target to be coded is as close as possible to the spectrum of the set background; The target to be coded is coated based on the optimal mixed filtering materials; 2. The method of claim 1, wherein, Target recognition is performed based on the target to be coded after coating, whether the recognition accuracy is reduced to an expected value is judged, and when the recognition accuracy reaches the expected value, the anti-counterfeiting design is completed; otherwise, the coding scheme is adjusted.
3. The method of claim 2, wherein the method further comprises: The set background is a background where the target to be coded is located or a set simulation target. The set background comprises multiple objects or components; 4. The method of claim 3, wherein, Each object or component has corresponding optimal mixed filtering materials, that is, each optimal mixed filtering material is used for coating the target to be coded, so that the spectrum of the target to be coded is as close as possible to the spectrum of the corresponding object or component. The spectrum image data of the target to be coded and the set background are acquired, the spectrum features of the target to be coded and the set background are extracted respectively, and difference analysis is performed, which comprises the following steps: Spectrum curves of the target to be coded and the set background are extracted respectively; 5. The method of claim 4, wherein, Difference between the spectrum curves of the target to be coded and the set background is compared. The material system with different filtering effects comprises the following steps: Within a set spectrum band range, a plurality of spectrum modulation materials are selected to form a spectrum modulation material set; 6. The method of claim 5, wherein the method further comprises: Optimal spectrum modulation materials are selected from the spectrum modulation material set to form a material system, so that the transmittance spectrum curves of the spectrum modulation materials are as irrelevant as possible. The optimal spectrum modulation materials are selected from the spectrum modulation material set to form a material system, so that the transmittance spectrum curves of the spectrum modulation materials are as irrelevant as possible, which comprises the following steps:
7. The method of claim 6, wherein the method further comprises: K-Means algorithm is used to select optimal spectrum modulation materials from the spectrum modulation material set to form a material system; wherein, the spectrum modulation angle is used as a distance measurement, the spectrum curves of all the spectrum modulation materials in the spectrum modulation material set are clustered, the similarity within a class is maximized, and the similarity between classes is minimized, a plurality of clustering centers are obtained through iteration, and a target number of optimal spectrum modulation materials are selected based on the clustering centers to form a material system. The material ratio calculation is performed according to the spectrum difference between the target to be coded and the set background, the optimal ratio coefficients of each material are determined, and the optimal mixed filtering materials are obtained, which comprises the following steps: A material transmittance matrix is constructed, the material transmittance matrix is used to represent the transmittance of each material in each channel; A mixing coefficient vector is constructed, the mixing coefficient vector is used to represent the mixing proportion of each material in the mixed filtering material; The transmittance of the mixed filtering material is multiplied by the spectrum of the target to be coded to obtain a camouflage target spectrum; 8. The method of claim 7, wherein the method further comprises: Based on an objective function, the camouflage target spectrum is linearly regressed with the set background spectrum to solve the optimal ratio coefficients of the mixed material, so that the camouflage target and the set background have the best linear relationship. The objective function is: Wherein, N represents the material type; C represents the number of channels; represents the proportioning coefficient of the i-th material; represents the transmittance of the i-th material in the channel j; represents the spectral intensity of the set background spectrum in the channel j, represents the spectral intensity of the camouflage target spectrum in the channel j; represents the j-th channel weight, used to adjust the importance of the channel.
9. The method of claim 8, wherein, The spectral modulation materials are all organic dyes, and can be dissolved in organic solvents; The organic solvents include photoresists, developing solutions or ethanol; Each of the spectral modulation materials in the material system is dissolved in an organic solvent according to a determined optimal proportioning coefficient to obtain a mixed filter material; A mixed filter material group is formed based on the mixed filter material corresponding to each object or component in the set background.
10. The method of claim 8, wherein the method further comprises: The coating coding of the target to be coded based on the optimal mixed filter material includes: Based on the mixed filter material group, a spatial arrangement mode of each mixed filter material is designed; wherein each mixed filter material is combined in different coating spaces. When the recognition accuracy is not reduced to the expected value, the adjustment of the coding scheme includes adjusting the proportioning of the mixed filter material and / or the coating space position.