Thermal activation bimodal confrontation garment and generation and control method thereof

Abstract

The invention discloses a thermal activation bimodal confrontation garment and a generation and control method thereof, belongs to the technical field of artificial intelligence safety and intelligent wearing, and aims to solve the problems that an existing confrontation patch is single in modal, conspicuous in appearance and lack of controllability. The garment comprises a fabric substrate layer, a confrontation pattern layer attached to the fabric substrate layer, a thermochromic layer covering the confrontation pattern layer and a flexible heating layer integrated in the garment. The thermochromic layer shields the confrontation pattern layer when the temperature is lower than a threshold value, the flexible heating layer is controlled to heat, so that when the temperature exceeds the threshold value, the thermochromic layer becomes transparent so as to expose the confrontation pattern layer, and meanwhile, the heating layer forms thermal textures in infrared imaging; the pattern of the confrontation pattern layer is generated through a two-stage algorithm including a shape optimization stage and a texture optimization stage so as to synchronously interfere with the visible light and infrared target detection model. According to the invention, on-demand activation of the confrontation effect, bimodal synchronous interference and good wearing concealment are realized.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence security and smart wearable technology, specifically relating to a thermally activated countermeasure garment capable of synchronously interfering with a dual-modal monitoring system of visible light and infrared, and a method for generating and controlling the countermeasures of the garment. Background Technology

[0002] In recent years, AI-driven surveillance systems (such as facial recognition and pedestrian detection) have been widely deployed in public safety, traffic management, and commercial security scenarios. Typical systems usually consist of visible light and infrared cameras, combined with edge / cloud computing and deep learning algorithms to achieve real-time detection and tracking of people or objects. However, the large-scale deployment of such systems has raised public concerns about privacy leaks and algorithm misuse. Due to the imperfections in the existing regulatory system, AI surveillance suffers from a serious lack of transparency and the risk of misuse in identity recognition and data collection, leading to a growing demand for proactive privacy protection technologies.

[0003] Among numerous privacy protection technologies, adversarial patches are a typical active defense method. Their principle is to attach algorithmically designed visual perturbation patterns to clothing or objects, causing AI detectors to misidentify or miss targets in the physical world. However, existing physical patches have significant drawbacks: first, most are designed only for a single mode of visible light or infrared, making it impossible to achieve simultaneous stealth and collaborative deception in composite monitoring systems deploying both visible light and infrared; second, traditional patches are often fixed textures with high saturation and high contrast, which, while interfering with detection models, easily attract human visual attention, making them unsuitable for everyday clothing scenarios and lacking in stealth and practicality. While existing research has attempted to achieve infrared deception using infrared heat sources or single-modal thermal patches, these solutions only work in the infrared domain, failing to address attacks under visible light and lacking control over visibility, making it difficult to meet the privacy protection needs of real-world scenarios. Summary of the Invention

[0004] (a) Technical problems to be solved The present invention aims to solve the technical problems in the prior art, such as the single mode of physical countermeasures patches, their conspicuous appearance and lack of concealment, and the lack of user-controllable on-demand activation mechanisms.

[0005] (II) Technical Solution To address the aforementioned technical problems, this invention proposes a thermally activated bimodal anti-interference clothing system and its generation and control method. The core idea is to combine digitally optimized bimodal anti-interference patterns with physically-domain thermochromic invisible materials and flexible heating technology to construct an intelligent garment with an ordinary appearance, but which can generate interference simultaneously in both visible and infrared dual modes through thermal activation. The specific technical solution is as follows: 1. Overall structural design The thermally activated bimodal combat garment comprises four layers arranged from top to bottom: a fabric base layer, a flexible heating layer, a combat patch layer, and a thermochromic layer. Figure 1 As shown. The four-layer structure works together to maintain the appearance of ordinary clothing at room temperature, and achieves dual-modal interference function after being activated by heating.

[0006] The fabric base layer is made of polyester fiber or cotton blend fabric, which supports the upper flexible heating layer, anti-patch layer and thermochromic layer, while providing wearing comfort and flexibility, and playing a role in thermal insulation to prevent the temperature of the flexible heating layer from being directly transferred to the human body.

[0007] The flexible heating layer, approximately 1 mm thick, is composed of nickel alloy resistance wire embedded in a silicone pad, ensuring both thermal conductivity and flexibility while maintaining insulation. Powered by a temperature control module, the heating layer generates a temperature field of 30–60 °C locally on the clothing, thereby activating the color-changing reaction of the thermochromic layer and generating controllable thermal texture interference in infrared images.

[0008] The adversarial patch layer is located between the flexible heating layer and the thermochromic layer, carrying a colored polygonal adversarial pattern optimized by a deep learning algorithm. This adversarial pattern is obtained through two-stage training of shape optimization and texture optimization, and can simultaneously mislead the target detection model in both infrared and visible light modalities.

[0009] The thermochromic layer is a key visibility control layer, employing a microencapsulated thermochromic dye. Each microcapsule contains an electron donor, an acceptor, and an organic solvent. When the temperature is below the solvent's melting point, the system forms a π-conjugated structure and appears black, obscuring the underlying antagonistic patch layer. When the temperature rises to 30°C, the conjugated structure is destroyed, the dye becomes transparent, revealing the antagonistic pattern of the antagonistic patch layer. The color reversibly recovers upon cooling.

[0010] 2. Adverse Pattern Optimization Methods The optimization framework for adversarial patterns includes two stages: shape update and texture update, such as... Figure 2 As shown.

[0011] Figure 2(a) illustrates the shape update stage of the patch, whose main goal is to optimize the patch's geometry to maximally evade infrared detector recognition. Specifically, this stage starts with an initial polygon and systematically adjusts its key parameters during iterative optimization, including the number of vertices, vertex distribution, circumscribed radius, and the overall rotation angle of the polygon, thereby searching for more deceptive candidate geometries in a continuous shape space. This parameterization not only improves the controllability of the search process but also allows the patch shape to achieve greater structural diversity while maintaining physical realizability. For each candidate shape, we paste it onto all labeled pedestrian instances in the infrared image to generate corresponding infrared adversarial examples. Subsequently, these generated examples are input into the infrared detector for inference to evaluate the degree of interference the shape causes to the target detection task. When the detector cannot recognize the pedestrian target with the patch (i.e., the output confidence is below 0.5), we consider the instance a successful attack. Based on this, we further calculate the person-to-person attack success rate (ASR) of each candidate shape across the entire dataset to measure its overall attack effectiveness. To ensure that the shapes advancing to the next optimization stage exhibit statistically stable and consistent attack performance, we select the candidate shape with the best ASR performance and pass it to the texture optimization stage. This strategy effectively avoids randomness caused by differences between samples, enabling the subsequently learned textures to further improve cross-modal attack performance based on the optimal geometry.

[0012] Figure 2(b) illustrates the texture update stage of the patch. Instead of using the optimal geometry obtained in the previous stage as a fixed structure, we focus on optimizing the patch's color texture to effectively interfere with the visible light detector's discrimination process. To achieve this, we introduce learnable RGB texture parameters in this stage and progressively update its appearance using a gradient-driven optimization strategy, maximizing the patch's deceptive capability in the visible light domain. Considering the unavoidable environmental changes in real-world physical deployments (e.g., changes in shooting angle, fluctuations in lighting conditions, resolution changes due to camera lenses), we introduce an Expectation over Transformation (EOT) mechanism during texture learning to explicitly model these potential perturbations. Specifically, EOT approximates the visual appearance changes that the patch may exhibit under different observation conditions by applying a series of randomized transformations (including rotation, brightness and contrast adjustments, blurring, scaling, etc.) to the patch and its surrounding image in each iteration. This strategy not only improves the sample diversity during the training phase but also significantly enhances the patch's generalization ability in the real physical world, enabling it to maintain stable attack performance under various imaging conditions. After EOT randomization, the generated patches are pasted onto designated pedestrian targets in the RGB image to construct adversarial examples in the visible light domain. These patched images are then fed into a visible light detector, where the adversarial loss is calculated to evaluate the texture's attack effectiveness. The gradient of the loss function is further used to update the patched texture, gradually evolving it in a more misleading direction. Through continuous iterative optimization, the final texture minimizes the visible light detector's confidence in identifying the target while maintaining its geometric structure.

[0013] 3. Garment preparation and control methods To prepare the thermochromic layer, 5 mL of thermochromic dye was first diluted with cyclohexanone at a volume ratio of 1:3, stirred thoroughly, and then sonicated for 10 seconds. The resulting solution was then transferred to an airbrush container and sprayed onto a fabric substrate. After coating, the dark thermochromic coating obscured the underlying fabric pattern, but became nearly transparent upon heating. In this study, a microencapsulated dye with a color-changing temperature of approximately 30 °C was selected. The color-changing temperature threshold of the microencapsulated dye could be controlled by adjusting the solvent composition.

[0014] To achieve stable and controllable infrared imaging attacks, we integrated flexible silicone-based heating elements into the fabric structure to reproduce the patch geometry obtained from the algorithm optimization. Figure 3(a) Demonstrates various patch shapes generated through a shape optimization stage, and corresponding heating elements with different contours were fabricated. The core of these heating elements consists of nickel alloy resistance wires wound around a glass fiber skeleton. This material features stable resistance, uniform thermal conductivity, and good bending performance, generating a continuous and uniform heat distribution at a power density of up to 0.4 W / cm². By controlling the layout of the resistance wires, we can accurately map the geometric perturbations of the patch to the infrared domain, allowing the optimized shape to exhibit a specific temperature distribution pattern in thermal imaging, thereby effectively interfering with the infrared detector's extraction of structural thermal features of the human body contour. To improve safety and mechanical stability during actual deployment, the heating element is coated with a composite silicone rubber insulation layer. This material not only possesses excellent flexibility and high-temperature resistance, but also has a dielectric breakdown strength of 20-50 kV / mm, effectively isolating the human body from the internal heating element and meeting the safety standards for wearable devices. Simultaneously, the silicone rubber layer has excellent thermal diffusion characteristics, further smoothing local temperature gradients and making the anti-texture more stable and continuous, thereby enhancing the perturbation consistency of the patch in the infrared mode.

[0015] Figure 3 (b) A digital temperature controller for the heating elements is demonstrated. Each heating element is powered by a portable DC power supply and adjusted in real time via a digital temperature controller. The temperature control range is 20-70°C, and it can adaptively adjust according to ambient temperature, subject's body surface temperature, and test scenario, thereby ensuring that the temperature contrast in the infrared image remains within the operating range that can effectively interfere with the detector. Furthermore, the total thickness of the heating elements is approximately 1 mm, allowing them to be easily embedded between the double-layered fabric structure of ordinary clothing without affecting the subject's comfort or causing significant visual abrupt changes in the visible light image, ensuring the coordinated attack capability of the visible light-infrared dual-modality system.

[0016] Figure 3 (c) Demonstrates the effectiveness of our patch clothing under infrared and visible light camera imaging. When the heating element temperature is set around 30°C (close to human body surface temperature), subtle but stable differences in heat distribution form on the surface, resulting in significant artificial thermal textures in infrared imaging. These textures neither deviate excessively from the natural temperature distribution of the human body and appear abrupt, nor do they effectively disrupt the structural thermal features relied upon by infrared detectors, thus significantly reducing the confidence in target detection. Extensive indoor and outdoor experiments demonstrate that this design can maintain stable thermal imaging perturbations under various environmental conditions, providing reliable hardware support for physical countermeasures attacks in the infrared modality.

[0017] (III) Beneficial Effects 1. This invention, through a four-layer collaborative structure and a two-stage algorithm optimization, enables clothing to simultaneously interfere with visible light and infrared monitoring systems after activation, thus overcoming the shortcomings of existing single-modal attacks. It is suitable for complex monitoring scenarios and significantly improves privacy protection capabilities.

[0018] 2. In this invention, the thermochromic layer is black at room temperature, and the appearance of the clothing is consistent with ordinary clothing, which can be integrated into daily wearing scenarios, avoiding the problem of traditional anti-pattern patches being conspicuous; the anti-pattern can be made visible or invisible on demand through temperature control, and the process of making or invisible is reversible, allowing users to flexibly switch protection modes according to the environment.

[0019] 3. In this invention, the adversarial pattern is optimized in two stages: shape and texture, and an expectation conversion mechanism is introduced to simulate various observation conditions in real-world scenarios, enabling the patch to maintain stable attack performance under different environments and exhibiting excellent physical robustness.

[0020] 4. In this invention, the flexible heating layer adopts a low-power design and is isolated from the human body through insulating materials, ensuring the safety and comfort of wearing it. Attached Figure Description

[0021] Figure 1 Design schematic diagram for the anti-clothing system; Figure 2 Optimize the framework to counter patching algorithms; Figure 3 This is a schematic diagram of a garment control method. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0023] like Figure 1 As shown, the thermally activated bimodal anti-color clothing system of the present invention mainly comprises a four-layer structure: from the outside to the inside (or from top to bottom), a thermochromic layer, an anti-color pattern layer, a flexible heating layer, and a fabric base layer. The thermochromic layer is directly exposed to the environment, and microcapsule thermochromic dyes with a color-changing temperature of approximately 30°C are applied to the fabric using a spraying method. The anti-color pattern layer is printed or woven onto the base layer, and its pattern is a color polygonal texture optimized by an algorithm. The flexible heating layer is located below the anti-color pattern layer, and is composed of nickel alloy resistance wires embedded in a silicone pad, with a thickness of approximately 1 mm. It is connected to a portable power supply and a digital temperature controller via wires. The fabric base layer is made of conventional fabric.

[0024] The system works as follows: At room temperature, the thermochromic layer is dark, obscuring the underlying anti-pattern layer and maintaining the garment's normal appearance. When the anti-pattern function needs to be activated, the flexible heating layer is activated via a temperature controller. The heat generated by the heating layer raises the temperature of the thermochromic layer to above 30°C, causing its color to fade and become transparent, thus revealing the colored texture of the anti-pattern layer and interfering with the visible light camera. Simultaneously, the flexible heating layer itself appears as a hot zone in the infrared thermal imager that matches the shape of the patch, such as... Figure 3 (c) As shown on the right, this abnormal heat distribution interferes with the infrared detector's recognition of the normal thermal outline of the human body. After the heating is turned off, the temperature drops, the thermochromic layer returns to its dark color, and the countermeasure effect is hidden.

[0025] Pattern generation for adversarial pattern layers is key. See also Figure 2 The generation process is divided into two stages. First, in the shape optimization stage (corresponding to...) Figure 2 (a) The algorithm initializes a polygon and generates multiple candidate shapes by adjusting parameters such as its vertices. These shapes are then pasted onto pedestrian targets in the infrared training image, input into the infrared detector, and the overall attack success rate of each shape is calculated. The shape with the highest success rate is selected to proceed to the next stage. Next, in the texture optimization stage (corresponding to...) Figure 2 (b) Using the optimal shape obtained in the previous stage as a fixed mask, its internal RGB texture parameters are initialized. In each optimization iteration, the expectation transformation technique is used to randomly transform the texture and background image (such as rotation and brightness change) to simulate changes in the physical world. Then, the transformed pattern is pasted onto the visible light training image, input into the visible light detector to calculate the loss, and the texture parameters are updated through backpropagation. Finally, a color texture that conforms to the optimal shape and can effectively attack the visible light detector is obtained.

[0026] The realization of flexible heating layer, such as Figure 3 As shown. Figure 3 (a) Shows flexible heating elements with different contours customized based on various patch shapes generated by an algorithm. At its core is a nickel alloy resistance wire, wrapped in an outer layer of silicone rubber insulation. Figure 3 (b) Shows the digital thermostat and power supply used to control the heating element. Figure 3 (c) The imaging effects of the clothing under visible light and infrared cameras were compared in the active state. The clothing showed a colored anti-interference pattern under visible light and a corresponding thermal shape under infrared, which verified the dual-modal interference effect.

[0027] In summary, this invention creatively realizes a practical and controllable bimodal physical countermeasures wearable solution through multi-material fusion and algorithmic collaboration. Those skilled in the art can make various modifications and variations based on the concept of this invention, but all such modifications and variations should be included within the protection scope of this invention.

Claims

1. A thermally activated dual-modal combat garment, characterized in that... The structure comprises a fabric base layer, a flexible heating layer, an anti-counterfeiting patch layer, and a thermochromic layer, arranged sequentially from top to bottom. The fabric base layer supports the upper structure and provides thermal insulation. The flexible heating layer generates specific thermal textures and heats the thermochromic layer. The anti-counterfeiting patch layer carries anti-counterfeiting patterns optimized by a two-stage algorithm. The thermochromic layer enables the controllable display and concealment of the anti-counterfeiting patterns. Through the synergistic effect of the four layers, synchronous interference with visible light and infrared monitoring systems is achieved.

2. The thermally activated bimodal combat garment according to claim 1, characterized in that... The thermochromic layer uses microcapsule-encapsulated thermochromic dyes. The microcapsules contain electron donors, acceptors, and organic solvents. It is black when the temperature is below the solvent melting point and becomes transparent when the temperature reaches 30°C.

3. The thermally activated bimodal combat garment according to claim 1, characterized in that... The adversarial pattern of the adversarial patch layer is generated through two-stage training: shape optimization and texture optimization. Shape optimization is for infrared detectors, and texture optimization is for visible light detectors.

4. The thermally activated dual-modal combat garment according to claim 3, characterized in that... The shape optimization stage starts with the initial polygon, adjusting the number of vertices, vertex distribution, circumscribed radius, and rotation angle. The candidate shape with the best attack success rate is evaluated and screened using an infrared detector.

5. The thermally activated bimodal combat garment according to claim 3, characterized in that... The texture optimization stage introduces an expectation transformation mechanism, which applies rotation, brightness adjustment, blurring, and scale transformations to optimize texture color through gradient-driven optimization.

6. The thermally activated bimodal combat garment according to claim 1, characterized in that... The flexible heating layer is made of nickel alloy resistance wire embedded in a silicone pad, with a thickness of 1mm, a maximum power density of 0.4W / cm², and a temperature control range of 20-70℃.

7. The thermally activated bimodal combat garment according to claim 1, characterized in that... The base layer of the fabric is made of polyester fiber or cotton blend, which provides both comfort and flexibility.

8. A method for generating a thermally activated bimodal combat garment, characterized in that... This includes garment structure preparation and adversarial pattern optimization. Garment structure preparation involves integrating four layers of structure sequentially into the garment, while adversarial pattern optimization is achieved through a two-stage training process involving shape and texture.

9. The generation method according to claim 8, characterized in that... The preparation steps of the thermochromic layer are as follows: dilute 5 mL of thermochromic dye with cyclohexanone at a volume ratio of 1:3, stir evenly, sonicate for 10 seconds, and then spray it onto the fabric substrate layer through an airbrush.

10. A control method for a thermally activated dual-modal countermeasures garment, characterized in that... The flexible heating layer is powered by a portable DC power supply, and the temperature is adjusted in real time by a digital temperature controller, enabling the thermochromic layer to switch between visible and invisible states, and producing stable thermal textures.

Images