A material physical property driven visible-infrared multi-modal detection interference method

Abstract

The application discloses a material physical property driven visible light-infrared multi-modal detection interference method. The method comprises the following steps: constructing a material attribute database containing a plurality of industrial materials; uniformly parameterizing coding is performed on an interference patch; based on the coding parameters, the physical parameters of the corresponding materials are obtained, and visible light interference images and infrared interference images are generated; the generated images are input into a visible light-infrared multi-modal target detection model, the interference performance corresponding to the coding parameters is calculated based on the target detection results output by the model; the coding parameters are jointly iteratively optimized to obtain an optimal material combination; the spatial structure of the interference patch is reconstructed, and physical interference on the visible light-infrared multi-modal target detection model is completed; and the application solves the technical problems of cross-modal physical modeling fragmentation, poor interference carrier stability and weak complex environment generalization ability in the existing visible light-infrared multi-modal physical interference method.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and more specifically, to a material physical property-driven method for detecting interference in visible and infrared multimodal modes. Background Technology

[0002] In recent years, visible light-infrared multimodal target detection technology has been widely used in fields such as UAV autonomous inspection, intelligent driving environment perception, and border security monitoring because it can fuse visible light texture details and infrared thermal radiation features, effectively making up for the shortcomings of single sensors in scene adaptability. It has significantly improved the accuracy of target perception in complex environments. However, deep learning models generally suffer from the problem of vulnerability to adversarial examples. Multimodal physical interference, as an important means of investigating security risks of intelligent perception systems and evaluating model robustness, is much more difficult to design than single-modal interference due to the difference in imaging mechanisms between visible light reflection and infrared radiation.

[0003] Existing multimodal physical interference methods mainly employ visible light texture optimization, infrared active temperature control, or cross-modal pixel joint optimization techniques, but still suffer from three core shortcomings: First, they lack a unified cross-modal modeling mechanism based on the intrinsic properties of materials, focusing only on constraints at the image layer or model output layer, failing to establish a complete correlation between "material properties - dual-modal imaging - detection features," which easily leads to dual-modal response mismatch; second, infrared interference depends on environmentally sensitive variables such as temperature, lacks a stable physical carrier, and exhibits large fluctuations in interference effects across different scenes; and third, they do not fully exploit the cross-modal modulation effects of intrinsic properties such as spectral reflectance and emissivity of materials, resulting in insufficient generalization ability in complex environments. Summary of the Invention

[0004] This invention provides a material physical property-driven visible-infrared multimodal detection interference method to at least solve at least one of the technical problems existing in the current visible-infrared multimodal physical interference methods, namely, fragmented cross-modal physical modeling, poor stability of the interference carrier, and weak generalization ability in complex environments.

[0005] According to one aspect of the present invention, a visible-infrared multimodal detection interference method driven by material physical properties is provided. This method may include: constructing a material property database containing various industrial materials; uniformly parameterizing and encoding interference patches attached to the surface of the target to be detected to obtain encoded parameters; querying the material property database based on the encoded parameters to obtain the physical parameters of the materials corresponding to each sub-region of the interference patch; synchronously driving the generation of visible light interference images and infrared interference images of the target to be detected based on the same set of physical parameters; inputting the generated visible light interference images and infrared interference images into the visible light-infrared multimodal target detection model to be interfered with; calculating the interference performance corresponding to the encoded parameters based on the target detection results output by the visible light-infrared multimodal target detection model; constructing a multi-objective optimization model with interference performance, visual concealment, and patch area as optimization objectives; jointly iteratively optimizing the encoded parameters to obtain the optimal material combination that satisfies the constraints; reconstructing the spatial structure of the interference patch based on the optimal material combination; determining the real material type corresponding to each region of the interference patch based on the material property database; fabricating a physical patch and attaching it to the surface of the target to be detected; synchronously interfering with the dual-modal imaging expression based on the optical reflection and thermal radiation characteristics of the material to complete the physical interference of the visible light-infrared multimodal target detection model.

[0006] Optionally, based on the same set of physical parameters, a visible light interference image and an infrared interference image of the target to be detected are generated synchronously, including: generating an interference patch texture based on visible light reflectance parameters and mapping it onto the surface of the target to form a visible light interference image; calculating the thermal radiation response of each region based on infrared emissivity parameters; and generating an infrared interference image by combining the temperature of the corresponding region on the surface of the target and the infrared imaging response function.

[0007] Optionally, the coding parameters are jointly iteratively optimized to obtain the optimal material combination that satisfies the constraints, including: An initial population of multiple candidate individuals is generated within a preset range. Based on the material distribution parameter index of each candidate individual, homologous physical parameters are extracted from the material property database. Visible light patches and infrared interference patches are rendered separately and mapped onto the surface of the target object to form a homologous dual-modal interference image for calculating the interference objective function, concealment objective function, and area objective function. Based on the interference objective function, concealment objective function, and area objective function of each candidate individual, non-dominated sorting is performed and non-dominated levels are defined. By calculating the crowding distance of candidate individuals at the same level, individuals with high non-dominated levels and sparse distribution are preferentially retained. The parent generation is used as the first generation. Continuous weighted recombination is performed on the parent generation's position and shape parameters, and discrete material number selection is performed on the parent generation's material distribution parameters for each partition to generate offspring. Random perturbation mutations are applied to the offspring generation's position and shape parameters, and the offspring generation's material number is discretely replaced within the candidate material library. When an offspring generation does not meet constraints related to position boundaries, patch area, material compatibility, or manufacturing processes, truncation or reselection is performed on the corresponding parameters to complete constraint repair. The parent and offspring generations are merged and re-sorted using non-dominated sorting and crowding distance screening to iteratively update the population until convergence conditions are met. The optimal material combination is then obtained by using the coding parameters with the best perturbation performance.

[0008] Optionally, the interference patch attached to the surface of the target being tested is uniformly parameterized and encoded, including: using position parameters to characterize the deployment area of ​​the interference patch on the surface of the target being tested; using shape parameters to characterize the outline of the interference patch and the division of each geometric sub-region; and using material distribution parameters to characterize the unique discrete material number corresponding to each sub-region in the material property database.

[0009] Optionally, the method further includes: when the encoding parameters change, directly indexing the material property database to obtain the physical parameters of the materials in each sub-region corresponding to the changed encoding parameters; using the obtained physical parameters as input to synchronously drive the generation of visible light interference images and infrared interference images.

[0010] Optionally, the material property database assigns a unique discrete number to each industrial material and records its color information, visible light reflectance parameters, infrared emissivity parameters, and processing and manufacturing constraints. The material property database also serves as a material combination search space in the interference optimization process, allowing the interference optimization variables to be mapped from image pixels to physical parameters.

[0011] Optionally, before constructing a material property database containing multiple industrial materials, the method further includes: using material physical parameters as a unified correlation between visible light modes and infrared modes; wherein, the visible light mode is determined by the visible light reflectance characteristics of the material surface, and the infrared mode is determined by the thermal radiation capability of the material; using the same set of physical parameters to synchronously describe the visible light reflectance characteristics and infrared emissivity parameters, so that the imaging response of the interference patch in both modes originates from the same real material; and establishing an infrared radiation model based on the thermodynamic radiation law, defining the mapping relationship between the infrared image grayscale value and the infrared emissivity parameter and the material surface temperature.

[0012] Alternatively, the expression for the infrared radiation model is as follows:

[0013] in, Represents the grayscale value of an infrared image; Represents the infrared imaging response function; Indicates the infrared emissivity parameter; This represents the Stefan–Boltzmann constant; This indicates the temperature of the corresponding area on the surface of the target being tested.

[0014] The beneficial effects of this invention are: This invention proposes a material physical property-driven visible and infrared multimodal detection interference method. Using physical parameters as a unified correlation benchmark between visible and infrared modes, a material property database encompassing visible light reflectance and infrared emissivity parameters is constructed. Simultaneous dual-modal imaging is driven by the physical parameters of homologous materials, establishing a correlation mechanism of "material properties—dual-modal imaging—detection response." This fundamentally solves the problem of dual-modal response mismatch in existing interference methods. Simultaneously, it avoids the dependence of existing infrared interference on environmentally sensitive variables such as active heating and cooling, using the intrinsic emissivity of the material as the core control variable. This ensures the interference patch has a stable physical carrier unaffected by environmental fluctuations, significantly reducing the impact of external conditions such as temperature, humidity, light, and wind speed on the interference effect. Furthermore, by integrating a multi-objective optimization mechanism that considers interference performance, visual concealment, and patch area, the cross-modal control effect of material properties is fully utilized. The resulting interference patch can be directly deployed based on industrial materials and maintains stable interference effects under complex weather conditions, different observation angles, and distances. This provides highly reliable and generalizable technical support for the robustness evaluation and safety protection testing of multimodal target detection models. Attached Figure Description

[0015] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a schematic diagram of a material physical property-driven overall framework for detecting interference in visible and infrared multimodal modes according to an embodiment of the present invention; Figure 2 This is a flowchart of a visible-infrared multimodal interference detection method driven by material physical properties according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a multimodal target detection interference optimization process based on material properties according to an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the visible light-infrared multimodal interference effect in a dataset scenario according to an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the deployment and interference effect of physical interference patches according to an embodiment of the present invention. Detailed Implementation

[0016] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0017] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] Example According to an embodiment of the present invention, a material physical property-driven visible light and infrared multimodal detection interference method is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system containing at least one set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0019] Figure 1This is a schematic diagram of visible-infrared multimodal detection interference driven by material physical properties according to an embodiment of the present invention, as shown below. Figure 1 As shown, this invention constructs a multimodal adversarial sample intelligent generation and robust optimization system architecture; this architecture is divided into five functional levels from bottom to top: 101 Data and Knowledge Foundation Layer: As the data foundation for the entire process, it integrates four types of core input resources, including a dual-modal image sample library composed of visible light data and infrared data, a material knowledge base covering industrial materials such as brass, concrete, and various colored alumina, and an environmental parameter library such as temperature, humidity, and atmospheric occlusion, providing real data support for subsequent physical modeling, patch rendering, and robustness optimization.

[0020] 102 Unified Parameter Modeling Layer: This layer transforms physical laws into optimizable parameters and comprises three core modules: First, it uses thermodynamic laws to perform physical modeling, establishing a quantitative mapping relationship between infrared radiation intensity, material emissivity, and ambient temperature. Second, it employs a gene encoding method that includes three types of variables—position, shape, and material—to transform interference patches into machine-solvable parameter vectors. Third, it uses a parameterization module to perform discretization and continuous unified characterization of patch deployment position, geometry, and material distribution, forming a standardized parameter input format.

[0021] 103 Interference Patch Generation and Rendering Layer: Generates dual-modal interference samples based on homologous material parameters, including two types of rendering outputs: visible domain samples and infrared domain samples. Modal correlation modeling ensures that the imaging responses of the two modes originate from the same set of material physical properties. The patch generation module completes the spatial structure and material distribution mapping of the interference patch. Cross-modal consistency verification ensures the physical homology of visible light texture and infrared thermal radiation, avoiding dual-modal response mismatch.

[0022] 104 Optimization and Robustness Enhancement Layer: This layer achieves multi-objective iterative optimization and environmental generalization enhancement of the interference patch. The core adopts the NSGA-II multi-objective optimization algorithm, which sequentially performs selection, crossover, mutation, and elite retention operations to complete population evolution. It is equipped with a DAR dynamic scene resampling mechanism, which uses attack success rate, visual concealment, and patch area as core evaluation indicators. It dynamically adjusts the sampling weights based on multi-scene samples to continuously improve the cross-scene robustness of the interference patch.

[0023] 105 Output and Application Layer: Completes the physical implementation and scenario application of the optimization results, outputting Vis patch and Infra patch design schemes corresponding to visible light and infrared modes respectively; completes the production of physical patches and attachment to target surfaces through deployment modules; and can ultimately be applied to various scenarios such as intelligent driving, security monitoring, and target camouflage.

[0024] Figure 2This is a flowchart of a material physical property-driven visible-infrared multimodal interference detection method according to an embodiment of the present invention, such as... Figure 2 As shown, the method may include the following steps: Step S101: Construct a material property database containing various industrial materials. The material property database records the physical parameters of various industrial materials, including visible light reflectance parameters and infrared emissivity parameters.

[0025] In the technical solution provided by step S101 of the present invention, the constructed material property database covers a variety of industrial materials that can be engineered and used. Each industrial material is assigned a unique discrete number for subsequent indexing. The industrial material properties recorded in the material property database include, in addition to visible light reflectance parameters and infrared emissivity parameters, color information and visible light spectral reflectance properties of the corresponding materials. At the same time, the processing and manufacturing constraints of each type of material are additionally associated and stored. The above parameters together constitute a complete and optimizable description of a single material.

[0026] The material property database has a dual function: serving as the foundational material library for subsequent dual-modal physical rendering and parameter querying, allowing direct retrieval of the corresponding material's dual-modal parameters via coded indexes during optimization; and as the material combination search space in the interference optimization process, its core role is to transfer the optimization variables of traditional multimodal physical interference from image pixels / textures to real material properties, binding physical realism from the source of optimization. The database construction phase embeds engineering constraints, allowing only materials that meet actual processing conditions to enter the candidate library. Simultaneously, it predefines the total area limit of interference patches and the maximum quantity limit for a single material, ensuring that the results obtained from subsequent optimizations naturally meet engineering manufacturing requirements without the need for secondary adaptation.

[0027] After the material property database is established, in the subsequent parametric encoding and multi-objective optimization of the interference patch, each sub-region of the interference patch directly corresponds to a unique material number in the database rather than the image pixel value. This ensures that the entire optimization process always revolves around the properties of real materials. The final optimal material combination allows the interference patch to be completely expressed by the combination of real materials, directly mapping to the procurement, cutting, and splicing of physical industrial materials without the need for additional texture conversion or parameter fitting, greatly improving the engineering deployability of the solution.

[0028] Before constructing a material property database containing various industrial materials, this invention first completes cross-modal physical modeling, using material physical parameters as a unified correlation benchmark between visible light modes and infrared modes. This benchmark constitutes the theoretical core of all subsequent physical rendering and optimization.

[0029] Specifically, the visible light mode is determined by the visible light reflectance parameters of the material surface, and its imaging essence is the reflection behavior of external incident light (such as sunlight); the infrared mode is determined by the material's thermal radiation capability, and in the passive interference scenario of this invention, it specifically refers to the material's own infrared emission characteristics. During the modeling process, the same set of physical parameters is strictly used to simultaneously describe the aforementioned visible light reflectance parameters and infrared emissivity parameters, ensuring that the imaging response of the interference patch in both modes originates from the intrinsic properties of the same real material. This physically solidifies the homogeneous correlation between the two modes, fundamentally avoiding the response mismatch problem caused by separate optimization in existing technologies.

[0030] Based on this, an infrared radiation model was further established based on the thermodynamic radiation law. This model strictly follows the Stefan-Boltzmann law and clarifies the physical mapping relationship between the gray value of the infrared image and the emissivity of the material and the steady-state temperature of the environment, rather than relying on an artificially set temperature distribution.

[0031] Step S102: The interference patch attached to the surface of the target being tested is uniformly parameterized and encoded to obtain the encoding parameters, which include the position parameters, shape parameters and material distribution parameters of the interference patch.

[0032] In the technical solution provided by step S102 of the present invention, in order to achieve automatic optimization of the interference patch, a unified parameter encoding method is used to describe the interference patch, and its encoding can be represented as:

[0033] in, This represents the set of parameters for the interference patch, signifying a complete interference patch configuration scheme. This indicates the position parameters of the interference patch, which are used to determine the specific deployment area of ​​the interference patch on the surface of the target being detected; The geometric shape parameters of the interference patch are used to characterize the overall outline of the interference patch and the boundaries of each geometric sub-region; Material distribution parameters are used to describe the unique discrete material number corresponding to each geometric sub-region of the interference patch in the material property database. Each number corresponds one-to-one with the complete physical parameters of a certain industrial material in the database. The above unified coding not only describes the spatial structure of the interference patch, but also determines the real material type corresponding to each region, providing a unified parameter input for subsequent dual-modal physical rendering and interference performance evaluation.

[0034] After the unified coding is completed, each interference candidate corresponds to a complete material configuration scheme, which clearly includes three core information: the position, shape and material composition of the interference patch. When any dimension of the coding changes, the visible light reflectance parameters and infrared emissivity parameters of the corresponding material can be retrieved directly from the material property database. No additional conversion is required to drive the subsequent dual-modal physical rendering process, ensuring the consistency of the interference scheme from digital design to physical deployment at the coding level.

[0035] Step S103: Based on the coding parameters, query the material property database to obtain the physical parameters of the corresponding materials of each sub-region of the interference patch. Based on the same set of physical parameters, synchronously drive to generate visible light interference images and infrared interference images of the detected target.

[0036] In the technical solution provided by step S103 of the present invention, based on the material distribution parameters in the coding parameters... The material property database constructed in step S101 is queried to obtain the complete physical parameters of the materials corresponding to each sub-region of the interference patch, including visible light reflectance parameters, infrared emissivity parameters, color information, and spectral reflectance properties. The core of this step follows the design logic of synchronously driving dual-modal imaging with the same set of physical parameters: the imaging response of both the visible light mode and the infrared mode is determined by the physical parameters of the same set of materials, rather than designing perturbations independently for the two modes, thus ensuring the physical consistency of the dual-modal response from the rendering source.

[0037] The process of generating a visible light interference image is as follows: based on the queried material color and spectral reflectance characteristics, combined with the geometric information of the surface of the target being detected, a texture corresponding to the interference patch is generated, and this texture is mapped onto the surface of the target being detected by position parameters. The defined deployment area ultimately forms a visible light interference image containing interference features. Different material combinations can present differentiated textures and colors in visible light imaging.

[0038] Infrared interference images are generated based on the aforementioned Stefan-Boltzmann infrared radiation model and its corresponding infrared imaging response function. Unlike existing technologies that rely on active temperature control to change the target temperature distribution, this invention constructs the infrared response through the difference in infrared emissivity parameters of materials: at the same ambient temperature T, different materials exhibit different infrared emissivity parameters. Different thermal radiation amounts are generated due to their different characteristics. These are mapped to infrared image grayscale values ​​through the infrared imaging response function, so that the infrared imaging response naturally originates from the physical properties of the material rather than an artificially set temperature field. This step combines the temperature distribution of the surface of the target being detected with the above-mentioned infrared radiation model to calculate the thermal radiation response of each sub-region, and finally generates the corresponding infrared interference image. Different material combinations will form different thermal radiation distributions in the infrared image.

[0039] The infrared imaging response function is a mapping operator connecting the thermal radiation of a material with the grayscale value of an infrared image, denoted as . It covers the linear or piecewise linear correction rules for the signal transfer function of infrared cameras and the detector responsivity. It is only used to convert the thermal radiation calculated by the Stefan-Boltzmann law into grayscale values ​​that conform to the actual imaging law, without introducing additional temperature perturbations.

[0040] The infrared radiation model is established using the Stefan–Boltzmann law, and its expression is as follows:

[0041] in, Represents the grayscale value of an infrared image; Represents the infrared imaging response function; Indicates the emissivity of the material; This represents the Stefan–Boltzmann constant; This indicates the surface temperature of the material.

[0042] Throughout the entire dual-modal physical rendering process, the visible light interference image and the infrared interference image are generated by the same set of material parameters: when the material composition of the interference patch changes, the imaging results of the two modes will change synchronously, avoiding the shortcomings of existing technologies that require separate design of perturbations for the visible light and infrared modes, and ensuring that the interference patch always has a unified and traceable material physical source.

[0043] Step S104: The generated visible light interference image and infrared interference image are input together into the visible light-infrared multimodal target detection model to be interfered with. Based on the target detection results output by the visible light-infrared multimodal target detection model, the interference performance corresponding to the coding parameters is calculated.

[0044] Among them, the visible light-infrared multimodal target detection model is a mature existing technology that has been trained. It widely adopts the mainstream architecture of dual-stream parallel feature extraction + cross-modal feature fusion and can be adapted to various existing model structures such as YOLOv3, YOLOv5, YOLOv8, YOLOv11, and Faster R-CNN. It has been widely used in scenarios such as UAV autonomous inspection, intelligent driving environment perception, border security monitoring, and military target recognition. Its standard workflow is as follows: one branch receives visible light images and extracts visible light discrimination features such as color, texture, and edge; the other branch receives infrared images and extracts infrared discrimination features such as thermal radiation distribution and contour. Then, the two types of features are integrated through the cross-modal fusion module. Finally, the detection result is output, which includes three core information: target category, detection confidence, and target bounding box position. If there is a target miss (no target detection result is output) or a false detection (the background is mistakenly identified as the real target), it will also be directly reflected in the output.

[0045] In the technical solution provided by step S104 of the present invention, the visible light interference image and the infrared interference image paired with the same source are fed into the visible light-infrared multimodal target detection model to be interfered with as standard inputs to carry out forward inference, so as to obtain the actual detection result corresponding to the current coding parameter g. This result will serve as the core evaluation basis for subsequent interference optimization.

[0046] The interference performance is calculated based on the detection results output by the model, and the evaluation is carried out using only the final output of the model. There is no need to obtain the internal feature information of the model. Therefore, it can be adapted to visible light-infrared multimodal target detection models of any structure and has strong versatility. The evaluation logic of interference performance is to compare the difference in detection results before and after interference: if the target detection confidence after interference is significantly lower than the original non-interference scene, the target localization error (such as the boundary box intersection union ratio IoU is lower) increases, or there are cases of target missed detection, category misclassification, false detection, etc., then the current interference scheme is judged to have excellent performance.

[0047] Figure 3 This is a schematic diagram of a multimodal target detection interference optimization process based on material properties according to an embodiment of the present invention, as shown below. Figure 3As shown, the unified material coding module takes physical data as input, decomposes it into three types of parameters: position, shape, and material index, and generates multiple sets of candidate codes to form an initial population. It then initializes the configuration by combining this with a dynamic scene pool covering various natural environments (such as sunny days, snowy days, and nighttime) and their corresponding weights. The core optimization process employs a closed-loop iterative mechanism: in each generation of evolutionary cycles, new candidate solutions are generated through selection, crossover, and mutation operations. Subsequently, in the physical rendering stage, the candidate solutions are mapped to specific material parameters, generating adversarial examples. After comprehensive evaluation using multi-dimensional indicators (success rate, area, concealment, etc.), the fitness is fed back to the optimizer to guide the evolutionary direction of the next generation of the population. Simultaneously, the sampling weights of the scene pool are dynamically updated based on the evaluation results. This design achieves deep coupling between environmental perception and evolutionary algorithms, forming an intelligent closed loop of "coding-optimization-rendering-evaluation-re-optimization," significantly improving the generalization ability and generation efficiency of adversarial examples across multiple scenarios.

[0048] Step S105: With interference performance, visual concealment, and patch area as optimization objectives, a multi-objective optimization model is constructed, and the coding parameters are jointly iteratively optimized to obtain the optimal material combination that satisfies the constraints.

[0049] In the technical solution provided by step S105 of the present invention, in order to balance interference capability, visual concealment, and practical deployment feasibility, the present invention further establishes a constraint-aware multi-objective optimization model to jointly optimize the interference patch; its overall optimization objective is mathematically expressed as minimizing the objective vector, which is expressed as:

[0050] The model incorporates interference performance improvement, visual concealment enhancement, and manufacturing constraint satisfaction into the collaborative optimization process through a unified minimization framework, reflecting the design concept of constraint awareness. The objective function is called the interference patch, which evaluates the ability of the interference patch to perturb multimodal discriminative feature representation and cross-modal fusion results. A larger value indicates a better interference effect. The expression is:

[0051] in, This represents a set of scenes (covering different backgrounds, lighting, viewing angles, etc.). Representing a scene The sampling weights; , and These represent the confidence levels of the target category outputs by the visible light detection branch, infrared detection branch, and cross-modal fusion detection branch, respectively, after applying the interference patch. , and These represent the weights of the corresponding detection results. By minimizing the interference objective function, the target category confidence of the three detection branches is reduced simultaneously, thereby weakening the discriminative ability of the multimodal target detection model and achieving cross-modal interference effects.

[0052] The stealth objective function, used to constrain the color and texture differences between the interference patch and the target background, is expressed as follows:

[0053] in, Indicates the area covered by the interference patch. Its total number of pixels; Indicates the interference patch at coordinates The color vector at that location (obtained from the visible light rendering result of step S103). The reference color vector represents the target surface or adjacent background area; by minimizing the concealment objective function, the visual difference between the interference patch and the target surface or adjacent background is reduced, thereby improving the concealment of the interference patch under visible light imaging and avoiding identification by human visual inspection or simple visual algorithms.

[0054] Let be the area objective function, used to limit the area ratio of the interference patch, and its expression is:

[0055] in, The area covered by the interference patch (determined by shape parameters) (Calculated) Let the area be the region to be interfered with, and the area objective function satisfy:

[0056] in, This indicates the preset maximum patch area percentage; for example, to ensure that the size of the interference patch is controllable and meets engineering deployment requirements. The value can be 15, meaning the total area of ​​the patch does not exceed 15% of the total area of ​​the target area being detected. This is merely an example and does not represent a complete definition. The specific value is limited; by minimizing While meeting the maximum area ratio constraint, the patch size should be minimized as much as possible to reduce processing complexity and deployment costs.

[0057] A multi-objective optimization method based on the non-dominated sorting genetic algorithm (NSGA-II) is adopted to jointly optimize the three types of coding parameters. Each candidate individual corresponds to a complete set of interference patch configuration schemes, the expression of which is:

[0058] in, , , These represent the position parameters, shape parameters, and discrete material numbers corresponding to each partition of the i-th candidate individual.

[0059] The steps to optimize the entire process are as follows: 1. Initial population generation: Generate an initial population containing multiple candidate individuals within a preset location range, shape range, and candidate material library; 2. Objective Function Calculation: For each candidate individual, the visible light reflectance and infrared emissivity of the corresponding material are extracted from the material property database based on its material number. Dual-modal physical rendering is then performed in step S103 to generate visible light interference images and infrared interference images. Subsequently, detection inference in step S104 is performed to calculate... , , Thus, the individual's , , Three objective function values; 3. Non-dominated ranking: When a candidate individual's evaluation value is not worse than another candidate individual in all three objective functions, and its evaluation value is better than another candidate individual in at least one objective function, the former candidate individual is determined to dominate the latter candidate individual; based on the dominance relationship, all candidate individuals are divided into multiple non-dominated levels; 4. Crowding distance calculation and selection: Crowding distance is calculated for candidate individuals at the same non-dominated level. Candidate individuals with higher non-dominated levels and larger crowding distances are retained first to maintain the diversity of material configuration schemes in the target space. Parent individuals are determined through selection operations based on non-dominated level and crowding distance. 5. Crossover: Perform crossover on the parent individuals—for continuous position parameters. With shape parameters Offspring parameters are generated through weighted recombination of continuous parameters; for discrete material distribution parameters... Each partition selects a material number from the material numbers corresponding to the two parent individuals to ensure that the child individuals maintain the discreteness of the material parameters; 6. Mutation Processing: Perform mutation operations on the offspring individuals generated by crossover—for continuous position and shape parameters, apply random perturbations under a preset mutation probability; for discrete material distribution parameters, select at least one patch partition under a preset mutation probability and replace its material number with other legitimate material numbers in the candidate material library to enrich population diversity; 7. Constraint Repair: Perform constraint repair on the mutated offspring individuals—when an offspring individual does not meet the position boundary constraints, area upper limit constraints, material compatibility constraints, or manufacturing constraints, truncate, scale, reselect, or replace the corresponding position parameters, shape parameters, or material numbers to ensure that all candidate individuals meet the engineering deployment requirements; 8. Population Update: Merge the current parent population with the generated offspring population, re-perform non-dominated sorting and crowding distance calculation on the merged population, and select a preset number of candidate individuals to form the next generation population based on the non-dominated level and crowding distance. 9. Iteration Termination: Repeat the above process of selection, crossover, mutation, constraint repair, and population update until the preset maximum number of iterations is reached or the Pareto solution set satisfies the preset convergence condition.

[0060] Step S106: Based on the optimal material combination, reconstruct the spatial structure of the interference patch, and determine the real material type corresponding to each region of the interference patch based on the material property database. Make a physical patch and attach it to the surface of the target to be detected. Rely on the optical reflection characteristics and thermal radiation characteristics of the material to synchronously interfere with the dual-modal imaging expression, and complete the physical interference of the visible light-infrared multimodal target detection model.

[0061] In the technical solution provided by step S106 of the present invention, the spatial physical structure of the interference patch is reconstructed based on the optimal shape parameters: the geometric parameters of the encoding layer are mapped to the two-dimensional plane / three-dimensional curved surface contour of the solid patch, and each material sub-region is divided according to the defined boundary to ensure that the partition structure of the solid patch is completely consistent with the rendering model in the digital design stage, and the total area of ​​the patch meets the area constraint preset in step S105, so as to avoid the patch size being too large and affecting the normal aerodynamic / hydraulic characteristics of the target or exposing the interference intention.

[0062] Based on the material property database constructed in step S101 of the optimal material distribution parameter index, complete physical parameters of the discrete material number corresponding to each sub-region are obtained, including visible light reflectance parameters, infrared emissivity parameters, color information, spectral reflectance properties and processing and manufacturing constraint parameters. This ensures that the intrinsic properties of the physical material are completely consistent with the parameters called in the digital optimization stage, thereby eliminating the attribute deviation between digital design and physical entity from the root and solving the problem of the gap between simulation and reality in existing physical adversarial examples.

[0063] Based on the reconstructed spatial structure and matching real materials, physical patches are fabricated using processes that comply with manufacturing constraints: for rigid materials (such as metal foil and ceramic sheets), precision laser cutting is used to process the contours of each section; for flexible materials (such as special coated fabrics and flexible films), die-cutting is used. When splicing materials from adjacent sections, material compatibility constraints must be met (such as matching thermal expansion coefficients and meeting the stability standards of the bonding interface), and the overall thickness of the patch must be controlled within a threshold much smaller than the feature size of the target being detected (for example, not exceeding 1 / 10 of the minimum feature size of the target; this is only an example and does not limit the specific value and range of the threshold for the feature size of the target being detected), to avoid the patch's own structure negatively affecting the target's operational characteristics. The material properties of each region of the fabricated physical patch are completely identical to the parameters called during the dual-modal physical rendering in step S103, ensuring the consistency between the simulated interference effect and the actual physical interference effect.

[0064] Based on the deployment area determined by the optimal position parameters, the physical interference patch is fixed to the surface of the target being tested using an attachment method adapted to the target surface material (such as structural adhesive bonding, magnetic fixation, riveting, etc., all of which must meet the processing and manufacturing constraints). During the attachment process, the positional deviation must be strictly controlled to not exceed a preset threshold (for example, set to not exceed 5% of the patch outline size; this is only an example and does not limit the specific value of the preset threshold). At the same time, it is ensured that the patch is completely attached to the target surface, without wrinkles or bubbles, to avoid unexpected shifts in visible light reflection characteristics or infrared radiation characteristics due to poor contact.

[0065] After the physical interference patch is attached, based on the constructed cross-modal physical model and infrared radiation model, the dual-modal imaging expression is simultaneously interfered with by the intrinsic optical reflection characteristics and thermal radiation characteristics of the material. Among them, the physical mechanism of infrared interference strictly follows the thermal radiation law driven by the material emissivity: without the need for active heating or cooling, the inherent emissivity difference of different materials is used to form a specific thermal radiation distribution at the same ambient temperature, thereby inducing the infrared detection branch to output low confidence results, fundamentally avoiding the defect of existing temperature control interference schemes being greatly affected by environmental fluctuations.

[0066] In the visible light band, the visible light reflectance parameters of the material determine the reflectance radiance, generating texture features that are highly integrated with the target background in visible light imaging, inducing the visible light detection branch to output low-confidence or false detection results. In the infrared band, the infrared emissivity of the material, as an intrinsic constant, combined with the steady-state temperature of the target surface, is mapped to an infrared radiation response through the Stefan-Boltzmann law, generating a thermal radiation distribution that matches the target background in infrared imaging, inducing the infrared detection branch to output low-confidence or false detection results. Since both modal responses are synchronously driven by the same set of material physical parameters, the problem of dual-modal response mismatch in existing technologies is completely avoided, and there is no need to rely on active heating or cooling control.

[0067] After the physical patch is deployed, visible light-infrared dual-modal images containing the interference patch need to be collected in a real-world scenario and input into the visible light-infrared multimodal target detection model to be interfered with for actual measurement and verification. This confirms that the interference effect is consistent with the interference performance obtained from the simulation in step S105. Since the optimization variable is the intrinsic property of the material, it is not affected by fluctuations in external factors such as ambient temperature, light, and wind speed. It can still maintain a stable interference effect in complex environments such as high and low temperatures, rain, and dust. Furthermore, the physical patch can be reused repeatedly without the need to regenerate the disturbance pattern or equip it with temperature control equipment each time, which greatly reduces the engineering deployment cost.

[0068] The method described in this embodiment will be further described below.

[0069] As an optional embodiment, step S103, based on the same set of physical parameters, synchronously drives the generation of visible light interference images and infrared interference images of the target to be detected, including: generating interference patch textures based on visible light reflectance characteristic parameters and mapping them onto the surface of the target to be detected to form a visible light interference image; calculating the thermal radiation response of each region based on infrared emissivity parameters; and generating an infrared interference image by combining the temperature of the corresponding region on the surface of the target to be detected and the infrared imaging response function.

[0070] In this embodiment, the visible light interference image generation process is as follows: retrieve the visible light reflectance characteristics parameters (including spectral reflectance, color information, texture features, etc.) of each sub-region corresponding to the current encoding parameters from the material property database, combine the geometric information of the material surface (such as curvature and orientation) and the ambient lighting conditions (such as the incident angle of sunlight and spectral distribution), calculate the reflectance radiance of each coordinate point of the patch, and generate an interference patch texture that perfectly matches the real material properties; accurately map the texture to the surface of the target being detected according to the position and contour determined by the encoding parameters. The mapping process simultaneously considers the occlusion relationship of the target surface and the distribution of ambient shadows to ensure that the illumination and tone of the texture and the target background blend naturally to form a visible light interference image.

[0071] The process of generating infrared interference images is as follows: retrieve the infrared emissivity parameters of each sub-region corresponding to the current encoding parameters from the material property database. Calculating material surface coordinates based on the Stefan-Boltzmann law The amount of thermal radiation at a point is given by the formula:

[0072] in, Here, T is the Stefan-Boltzmann constant, and T is the temperature of the target surface being tested, eliminating the need for active heating or cooling regulation; due to the different materials The differences naturally lead to different thermal radiation distributions under the same temperature (T), eliminating the need for artificially set temperature disturbances and avoiding the instability problem of existing temperature control schemes from the source of physical calculations.

[0073] The physical thermal radiation is mapped to a grayscale image conforming to the imaging characteristics of an infrared camera. The infrared imaging response function involved has already been defined in the sovereign modeling stage. Here, we supplement its complete mapping logic: the calculated thermal radiation... Mapping the digital grayscale range of the infrared sensor yields the grayscale values ​​of the infrared image:

[0074] Will Substituting, we get:

[0075] in, and These represent the minimum and maximum irradiance corresponding to the infrared image, respectively. and These represent the minimum and maximum grayscale values ​​of the infrared image, respectively. This indicates that the mapping result is limited to the grayscale range (e.g., the range [0,255] corresponding to an 8-bit infrared image); the parameters of the infrared imaging response function (e.g., gain, bias, quantization bit depth) can be pre-stored when establishing the material property database in step S101 to adapt to different models of infrared cameras and improve the versatility of the simulation results.

[0076] As an optional embodiment, step S105 involves jointly iteratively optimizing the encoding parameters to obtain the optimal material combination that satisfies the constraints. This includes: generating an initial population consisting of multiple candidate individuals within a preset range; extracting homologous physical parameters from a material property database based on the material distribution parameter index of each candidate individual; rendering visible light patches and infrared interference patches respectively; mapping the visible light patches and infrared interference patches onto the surface of the target being detected to form a homologous dual-modal interference image to calculate the interference objective function, concealment objective function, and area objective function; performing non-dominated sorting and dividing non-dominated levels based on the interference objective function, concealment objective function, and area objective function of each candidate individual; and calculating the candidate at the same level... For each individual, individuals with high non-dominated ranking and sparse distribution are prioritized as parents based on their crowding distance. The position and shape parameters of the parents are continuously weighted and recombined, while the material distribution parameters of the parents are selected by discrete material numbering on a partition-by-partition basis to generate offspring. Random perturbation mutations are applied to the position and shape parameters of the offspring, and the material numbers of the offspring are discretely replaced and mutated within the candidate material library. When an offspring does not meet constraints related to position boundaries, patch area, material compatibility, or manufacturing processes, the corresponding parameters of the offspring are truncated or reselected to complete the constraint repair. Parents and offspring are merged and re-sorted by non-dominated ranking and crowding distance to iteratively update the population until convergence is met, at which point the coding parameters with the best perturbation performance are obtained as the optimal material combination.

[0077] In this embodiment, the steps for optimizing the entire process have been detailed in sovereign step S105 and will not be repeated here. The following focuses on the supplementary limiting details.

[0078] The initial population generation is pre-defined within the deployable area of ​​the target surface, excluding non-attachable areas such as the power compartment and sensors; the shape parameters are limited to the geometric contours that conform to processing techniques such as laser cutting and die cutting; the candidate material library directly calls the set of engineerable industrial materials screened in step S101 to ensure that the initial population naturally falls within the deployable space.

[0079] The dual-modal rendering process involves extracting common physical parameters from the material property database for each candidate individual based on its material distribution parameters. This means simultaneously extracting the visible light reflectance and infrared emissivity of the same set of materials, rendering visible light patches and infrared interference patches separately, and then mapping them onto the surface of the target being detected according to the position and shape parameters of the candidate individual to form a strictly paired common dual-modal interference image, thus avoiding dual-modal response mismatch from the rendering source.

[0080] Crossover and mutation operations are designed differently for continuous-discrete hybrid coding characteristics: continuous weighted recombination is performed on the position and shape parameters of the parent generation, and the parameters of the offspring generation are convex combinations of the corresponding parameters of the parent generation; discrete material number selection is performed on the material distribution parameters of the parent generation for each partition, and the number of each partition of the offspring generation is randomly selected from the numbers of the two corresponding partitions of the parent generation; random perturbation mutation is applied to the position and shape parameters of the offspring generation; and the material number of the offspring generation is discretely replaced and mutated in the candidate material library, and the number of the selected partition is replaced with other legal numbers in the library.

[0081] The specific triggering conditions for constraint repair are as follows: when the offspring does not meet the position boundary constraints, its position parameters are truncated; when the patch area constraints are violated, the shape parameters are scaled; when material compatibility constraints are triggered (such as mismatch in thermal expansion coefficients), the compatible number is reselected; when manufacturing constraints are violated (such as minimum manufacturing size), the processable number is replaced to ensure that the optimization results can be deployed directly without secondary adaptation.

[0082] Once the iteration meets the convergence condition, the coding parameters with the best interference performance are selected as the optimal material combination from the Pareto optimal solution set that satisfies all constraints, and output to step S106 for solid patch preparation.

[0083] As an optional embodiment, step S102 involves uniformly parameterizing and encoding the interference patch attached to the surface of the target being tested, including: using position parameters to characterize the deployment area of ​​the interference patch on the surface of the target being tested; using shape parameters to characterize the outline of the interference patch and the division of each geometric sub-region; and using material distribution parameters to characterize the unique discrete material number corresponding to each sub-region in the material property database.

[0084] In this embodiment, the deployment area of ​​the interference patch on the surface of the target being detected is characterized by positional parameters, i.e. The value is limited to the deployable area on the surface of the target being tested, as defined by processing and operational requirements (excluding areas such as the power compartment, sensors, and control surfaces that cannot be attached or affect the normal operation of the target). The position parameter can be represented by the anchor point coordinates or area index in the local coordinate system of the target surface, to ensure that the patch position can be accurately reproduced during subsequent optimization and physical deployment.

[0085] Shape parameters are used to characterize the contour of the interference patch and the division of each geometric sub-region, i.e. It not only defines the overall outer contour boundary of the interference patch (such as a rectangle, polygon, or irregular fitted contour), but also further clarifies the number, boundary, and topological relationship of several geometric sub-regions divided within the contour. This provides a spatial carrier for the subsequent material distribution parameter "assigning material numbers to each sub-region"—that is, each geometric sub-region corresponds to a unique discrete material number in the material property database in subsequent steps. Different sub-regions can be assigned the same or different material numbers, thereby realizing the optimization of material combination within the interference patch.

[0086] The material distribution parameters characterize the unique discrete material ID corresponding to each sub-region in the material property database, i.e. It is a one-dimensional vector or a two-dimensional mapping matrix, with the length / size consistent with the number of sub-regions defined by the shape parameters. Each element takes a unique discrete material number (not an image pixel value or RGB texture value) from the material property database constructed in step S101. Each number corresponds one-to-one with the complete physical parameters of the industrial material in the database, including visible light reflectance parameters, infrared emissivity parameters, color information, spectral reflectance properties, and processing and manufacturing constraints.

[0087] As an optional embodiment, the method further includes: when the encoding parameters change, directly indexing the material property database to obtain the physical parameters of the materials in each sub-region corresponding to the changed encoding parameters; using the obtained physical parameters as input to synchronously drive the generation of visible light interference images and infrared interference images.

[0088] In this embodiment, when the encoding parameters change due to crossover, mutation, or constraint repair operations during the optimization iteration process, the unique discrete material number of the corresponding sub-region in the material property database is directly indexed based on the changed material distribution parameters. The physical parameters such as visible light reflectance and infrared emissivity bound to the number are then retrieved. The indexing process has zero additional conversion overhead and requires no additional texture conversion or parameter fitting. The retrieved physical parameters are synchronously input into the visible light and infrared dual-modal rendering branches. No additional perturbation variables independent of material properties are set in the two branches, ensuring that the dual-modal response is synchronous and avoids the gap between simulation and reality and the dual-modal mismatch problem from the root.

[0089] Experimental section: To verify the effectiveness of the method of this invention, a visible-infrared multimodal test environment was constructed in a simulation platform, and various test scenarios were established, including different weather conditions, backgrounds, observation distances, and observation angles. Based on a visible-infrared multimodal target detection interference method driven by material physical properties, interference patch optimization was completed, and interference verification was conducted in both publicly available multimodal datasets and real physical scenarios.

[0090] As shown in Table 1, the present invention outperforms existing methods in terms of interference success rate, false negative rate and other interference evaluation indicators on publicly available multimodal datasets, and can achieve stable interference effect while maintaining high visual concealment. Experimental results show that the present invention uses real material properties as interference variables, enabling the interference patch to take into account dual-modal interference performance, engineering manufacturability and adaptability to complex environments, thus verifying the effectiveness and practical value of the proposed method.

[0091] Table 1: Comparison of interference success rates (%) on different multimodal datasets and detection models

[0092] in addition, Figure 4 This is a schematic diagram illustrating the visible-infrared multimodal interference effect in a dataset scenario according to an embodiment of the present invention, such as... Figure 4 As shown, the interference effect of this invention on a publicly available visible-infrared multimodal dataset is presented. After deploying the interference patch, the detection confidence of the target in both visible and infrared images decreases significantly, and the detection box is lost. This indicates that the present invention can effectively induce the multimodal target detection model to produce target misses or detection errors, verifying the interference effectiveness of the method in a digital testing environment.

[0093] Furthermore, Figure 5 This is a schematic diagram illustrating the deployment and interference effect of physical interference patches according to an embodiment of the present invention, as shown below. Figure 5 As shown, the interference effect of the present invention in a real physical scenario is illustrated. After the interference patch is fabricated based on the optimized material combination and physically deployed, the target can be stably interfered with in both visible and infrared modes. This demonstrates that the interference patch generated by the present invention not only maintains dual-mode interference capability but also has good physical deployment feasibility and engineering application value.

[0094] In this embodiment of the invention, a material property database containing various industrial materials is constructed; interference patches attached to the surface of the target being detected are uniformly parameterized and encoded to obtain encoding parameters; based on the encoding parameters, the material property database is queried to obtain the physical parameters of the materials corresponding to each sub-region of the interference patch; based on the same set of physical parameters, a visible light interference image and an infrared interference image of the target being detected are generated synchronously; the generated visible light interference image and infrared interference image are jointly input into the visible light-infrared multimodal target detection model to be interfered with; based on the target detection results output by the visible light-infrared multimodal target detection model, the interference performance corresponding to the encoding parameters is calculated; with interference performance, visual concealment, and patch area as optimization objectives, a multi-objective optimization model is constructed to perform joint optimization on the encoding parameters. Through iterative optimization, the optimal material combination that satisfies the constraints is obtained. Based on the optimal material combination, the spatial structure of the interference patch is reconstructed, and the real material type corresponding to each region of the interference patch is determined based on the material property database. A physical patch is then fabricated and attached to the surface of the target being detected. By relying on the optical reflection and thermal radiation characteristics of the material, the dual-modal imaging representation is synchronously interfered with, thus completing the physical interference of the visible light-infrared multimodal target detection model. This solves at least one of the following technical problems: response mismatch caused by separate optimization of the dual modes, poor environmental stability and high cost of infrared temperature control schemes, and the gap between simulation and reality caused by the difficulty in materializing pixel-level perturbations. It achieves at least one of the following technical effects: synchronous interference from the same source of dual modes, no need for active temperature control, consistency between simulation and physical properties, and strong engineering deployability.

[0095] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0096] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0097] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between units or modules may be electrical or other forms.

[0098] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0099] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a first processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0100] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A material physical property-driven visible-infrared multimodal interference detection method, characterized in that, include: A material property database containing various industrial materials is constructed. The material property database records the physical parameters of various industrial materials, including visible light reflectance parameters and infrared emissivity parameters. The interference patch attached to the surface of the target being tested is uniformly parameterized and encoded to obtain encoding parameters, which include the position parameters, shape parameters and material distribution parameters of the interference patch; Based on the encoding parameters, the material property database is queried to obtain the physical parameters of the material corresponding to each sub-region of the interference patch. Based on the same set of physical parameters, the visible light interference image and infrared interference image of the detected target are generated synchronously. The generated visible light interference image and the infrared interference image are input together into the visible light-infrared multimodal target detection model to be interfered with. Based on the target detection results output by the visible light-infrared multimodal target detection model, the interference performance corresponding to the coding parameters is calculated. Using the interference performance, visual concealment, and patch area as optimization objectives, a multi-objective optimization model is constructed to jointly iteratively optimize the coding parameters and obtain the optimal material combination that satisfies the constraints. Based on the optimal material combination, the spatial structure of the interference patch is reconstructed, and the real material type corresponding to each region of the interference patch is determined based on the material property database. A physical patch is then made and attached to the surface of the target being detected. By relying on the optical reflection and thermal radiation characteristics of the material, the dual-modal imaging expression is simultaneously interfered with, thus completing the physical interference of the visible light-infrared multimodal target detection model.

2. The method according to claim 1, characterized in that, The method of synchronously driving based on the same set of physical parameters to generate visible light interference images and infrared interference images of the detected target includes: Based on the visible light reflectance parameters, an interference patch texture is generated and mapped onto the surface of the target being detected to form the visible light interference image; Based on the infrared emissivity parameters, the thermal radiation response of each region is calculated; The infrared interference image is generated by combining the temperature of the corresponding area on the surface of the detected target with the infrared imaging response function.

3. The method according to claim 1, characterized in that, The joint iterative optimization of the encoding parameters to obtain the optimal material combination that satisfies the constraints includes: An initial population consisting of multiple candidate individuals is generated within a preset range; Based on the material distribution parameter index of each candidate individual, the same physical parameters are extracted from the material property database. Visible light patches and infrared interference patches are rendered respectively, and the visible light patches and infrared interference patches are mapped onto the surface of the target being detected to form a same dual-modal interference image to calculate the interference objective function, the concealment objective function and the area objective function. Based on the interference objective function, concealment objective function and area objective function of each candidate individual, non-dominated sorting is performed and non-dominated levels are divided. By calculating the crowding distance of candidate individuals in the same level, individuals with high non-dominated levels and sparse distribution are preferentially retained as parents. The position and shape parameters of the parent generation are continuously weighted and recombined, and the material distribution parameters of the parent generation are discretely selected by partition to generate the offspring generation. Random perturbation mutations are applied to the position and shape parameters of the offspring, and the material numbers of the offspring are discretely replaced and mutated in the candidate material library; When the offspring does not meet the constraints of location boundary, patch area, material compatibility or manufacturing process, the corresponding parameters of the offspring are truncated or reselected to complete the constraint repair. The parent and offspring generations are merged and non-dominated sorting and crowding distance screening are performed again to iteratively update the population until the convergence condition is met. The coding parameters with the best interference performance are then obtained as the optimal material combination.

4. The method according to claim 1, characterized in that, The process of uniformly parameterizing and encoding the interference patches attached to the surface of the target being detected includes: The location parameters are used to characterize the deployment area of ​​the interference patch on the surface of the target being detected; The shape parameters are used to characterize the outline of the interference patch and the division of each geometric sub-region; The material distribution parameters are used to characterize the unique discrete material number corresponding to each sub-region in the material property database.

5. The method according to claim 1, characterized in that, The method further includes: When the coding parameters change, the material property database is directly indexed to obtain the physical parameters of the materials in each sub-region corresponding to the changed coding parameters. The acquired physical parameters are used as input to synchronously drive the generation of the visible light interference image and the infrared interference image.

6. The method according to claim 1, characterized in that, The material property database assigns a unique discrete number to each industrial material and records its color information, visible light reflectance parameters, infrared emissivity parameters, and processing and manufacturing constraints. The material property database also serves as the material combination search space in the interference optimization process, allowing the interference optimization variables to be mapped from image pixels to physical parameters.

7. The method according to claim 1, characterized in that, Before constructing a material property database containing various industrial materials, the method further includes: Material physical parameters are used as a unified correlation between visible light modes and infrared modes; wherein, the visible light modes are determined by the visible light reflection characteristics of the material surface, and the infrared modes are determined by the thermal radiation capability of the material; The visible light reflectance parameters and the infrared emissivity parameters are described synchronously using the same set of physical parameters, so that the imaging response of the interference patch in both modes originates from the same real material. An infrared radiation model is established based on the thermodynamic law of radiation, defining the mapping relationship between the grayscale value of the infrared image and the infrared emissivity parameter and the surface temperature of the material.

8. The method according to claim 7, characterized in that, The expression for the infrared radiation model is as follows: in, Represents the grayscale value of an infrared image; Represents the infrared imaging response function; Indicates the infrared emissivity parameter; This represents the Stefan–Boltzmann constant; This indicates the temperature of the corresponding area on the surface of the target being tested.

Images