An intelligent perception jamming method based on color trapping

Abstract

This invention discloses an intelligent perception interference method based on color decoys, comprising: image acquisition and preprocessing, color feature extraction and analysis, color-based adversarial patch design, adversarial patch generation and application, and effect evaluation and adjustment. The image acquisition and preprocessing part acquires and preprocesses data; the color feature extraction and analysis part extracts and analyzes multi-angle color features from the acquired images; the color-based adversarial patch design part designs adversarial patches of different sizes, shapes, and colors based on the determination of key locations; the adversarial patch generation and application part deploys the designed patches to key parts of the vehicle; and the effect evaluation and adjustment part uses a drone to re-photograph the vehicle after the adversarial patches are deployed, and sends the images to the detection model to evaluate the effectiveness of the color decoy adversarial patches, and adjusts the design scheme based on the detection results.

Description

Technical Field

[0001] This application relates to the technical field of deep learning intelligent perception interference algorithms, and in particular to an intelligent perception interference method based on color trapping. Background Technology

[0002] The research goal of intelligent perception perturbation algorithms is to create adversarial examples. These examples involve adding extremely minor perturbations or disturbances to original, uncontaminated images. These perturbations are almost imperceptible to human vision and therefore do not affect human recognition capabilities. However, these subtle changes can interfere with machine learning models, causing them to make incorrect judgments or predictions.

[0003] There are various intelligent perceptual interference methods for generating adversarial examples, each with its own characteristics. They can be mainly divided into three categories: gradient-based methods, optimization-based methods, and generative model-based methods.

[0004] First, gradient-based methods primarily rely on the model's gradient information to construct adversarial perturbations. This approach is direct and efficient because it directly utilizes the model's internal gradient changes to locate and generate perturbations that can mislead the model. However, it can also be limited by the model structure; for some complex models or structures, it may not be able to accurately capture enough gradient information to generate effective adversarial examples.

[0005] Secondly, optimization-based methods use an iterative optimization process to find perturbations that minimize the model's prediction error. This approach is more flexible because it does not depend on the specific structure of the model; instead, it iteratively adjusts the perturbations until it finds the minimum perturbation that can mislead the model. However, this method is relatively computationally expensive because it requires multiple iterations and optimization processes.

[0006] Finally, generative model-based methods utilize models such as Generative Adversarial Networks (GANs) to generate adversarial examples. This approach can generate diverse adversarial examples because models like GANs can learn the distribution of the original data and generate new data samples. However, the training process for this method is relatively complex, requiring substantial data and computational resources to train the generative model.

[0007] These three methods each have their own advantages and disadvantages, and are suitable for different scenarios and needs. However, they all basically follow the deep learning approach of updating based on the network model and gradient backpropagation. This approach allows for the generation of adversarial examples simply by designing new loss terms according to requirements. However, this approach also has obvious drawbacks, as deep neural network-based methods generally require a large amount of data, significant computing power, and considerable training time. Summary of the Invention

[0008] This application provides a color-based intelligent perception interference method with three advantages. First, through detailed analysis and optimization of colors, the adversarial patch can more effectively mislead the detection system, improving the interference success rate. Second, the high degree of color fusion makes the adversarial patch visually difficult to detect, improving the stealth of the interference attack. Third, this method does not rely on complex gradient learning, reducing implementation difficulty and computational cost.

[0009] Firstly, a color-based intelligent perception interference method is provided, including:

[0010] Image acquisition and preprocessing: Acquire target images taken from the airborne base station; perform image preprocessing, including color correction and noise reduction;

[0011] Color feature extraction and analysis: Using image processing techniques, extract color features from the image, including color distribution, hue, and saturation; analyze the extracted color features to find patterns that match the target color and background color;

[0012] Complete the design of a color-based adversarial patch: Based on the color feature analysis results, design a color scheme for the adversarial patch so that the patch visually blends with the original image;

[0013] Generate and apply adversarial patches: Generate adversarial patches based on the color scheme of the adversarial patches; paste the generated adversarial patches onto the target surface, ensuring that the patch colors are in harmony with the surrounding environment.

[0014] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes:

[0015] Effect evaluation and adjustment: The target with the anti-countermeasures patch was photographed using an airborne ground-based imaging device. The captured images were then input into the target detection model for testing. Based on the test results, the interference effect of the anti-countermeasures patch was evaluated. Based on the evaluation results, the color scheme of the anti-countermeasures patch was adjusted to improve the interference efficiency.

[0016] In conjunction with the first aspect, in some implementations of the first aspect, image preprocessing includes at least one of the following:

[0017] Convert the image from the RGB color space to the HSV color space to separate hue, saturation, and brightness information;

[0018] A Gaussian filter is applied to smooth the image and remove noise;

[0019] Adjust the contrast and brightness of the image to ensure the accuracy and consistency of color information.

[0020] In conjunction with the first aspect, in some implementations of the first aspect, color feature extraction includes:

[0021] Divide the image into HSV channels: H, S, and V.

[0022] Calculate a color histogram for each channel and record the frequency of occurrence of each color component.

[0023] In conjunction with the first aspect, in some implementations of the first aspect, color feature analysis includes color feature calculation and color difference analysis;

[0024] Color feature calculation includes: calculating the average hue, average saturation, and average brightness values ​​of the image as a description of the overall color features; and calculating the variance and standard deviation of the color distribution to assess the degree of dispersion of color in the image.

[0025] Color difference analysis includes: analyzing the differences in hue and saturation between the target and the background; and identifying the key color components used to distinguish the target from the background.

[0026] In conjunction with the first aspect, in certain implementations of the first aspect, key color components used to distinguish the target from the background are determined, including:

[0027] Calculate the hue difference H between the target and the background in the HSV space respectively. dis And saturation difference S dis ,

[0028] Hue difference H dis H dis =|H objective_mean -H back_mean |,H objective_mean H is the average hue of the target. back_mean The average color tone of the background.

[0029] Saturation difference S dis :S dis =|S objective_mean -S back_mean |,S objective_mean S represents the average saturation of the target. back_mean The average saturation of the background;

[0030] By comparing hue differences H dis And saturation difference S dis The size of H determines the key color components for distinguishing the target from the background. dis >S dis If H, then hue is the key color component; if H dis dis Saturation is the key color component.

[0031] ​In conjunction with the first aspect, some implementations of the first aspect include color-based adversarial patching designs, such as:

[0032] Primary color selection: Based on the color difference analysis results, select a color that is similar in tone to the target color as the primary color for the adversarial patch;

[0033] Patch shape and size design: Based on the key feature areas of the target, design the shape and size of the countermeasure patch to ensure that the countermeasure patch can completely cover the key feature areas without affecting the target's normal driving and appearance;

[0034] Color blending: Using color mixing technology, the base color of the anti-patch is blended with the colors of the surrounding environment.

[0035] In conjunction with the first aspect, in some implementations of the first aspect, the determination of similar colors includes:

[0036] The color mean values ​​of the H, S, and V channels of the image sub-region where the target is located are calculated respectively to obtain the three-dimensional feature vector HSV. mean =[H objective_mean ,S objective_mean V objective_mean ], H objective_mean S represents the average hue of the target. objective_mean V represents the average saturation level of the target. objective_mean The average brightness of the target;

[0037] HSV mean The average value of the key color components in the image is replaced with the difference value of the key color components, resulting in... and the replacement As a base color.

[0038] In conjunction with the first aspect, in some implementations of the first aspect, color blending processing includes:

[0039] Multiple sample points are selected from the area surrounding the adversarial patch, and the average value of these sample points in the HSV color space is calculated as the representative color of the background. The average color value of the sample points is denoted as B = [H]. b ,S b V b ], H b S represents the average hue of the sample points. b V represents the average saturation of the sample points. b This represents the average brightness of the sample points;

[0040] Color blending based on transparency; introducing a transparency parameter α, 0≤α≤1, and using a linear interpolation method to blend the base colors. Blend with background color B; the blended color is F = [H] f ,Sf V f ]for:

[0041] H f =(1-α)H b +αH objective_mean

[0042] S f =(1-α)S b +αS objective_mean

[0043] V f =(1-α)V b +αV objective_mean

[0044] H f S represents the hue values ​​of the fused sample points. f V represents the saturation value of the fused sample points. f This represents the brightness value of the sample points after fusion.

[0045] In a second aspect, an electronic device is provided for performing the method as described in any of the implementations of the first aspect above.

[0046] Compared with the prior art, the solution provided in this application has at least the following beneficial technical effects:

[0047] First, through detailed analysis and optimization of colors, adversarial patches can more effectively mislead detection systems, increasing the success rate of deception. Second, the high degree of color integration makes adversarial patches visually difficult to detect, improving the stealth of deception attacks. Third, this method does not rely on complex gradient learning, reducing implementation difficulty and computational cost. Attached Figure Description

[0048] Figure 1 This is a flowchart of an intelligent perception interference method based on color trapping. Detailed Implementation

[0049] The present application will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0050] This invention provides a color-based intelligent perception interference method. This method emphasizes the role of color in adversarial example design. By performing detailed color analysis on the entire image, it generates adversarial patches that highly blend with the background but mislead the detection system (adversarial patches are a special case of adversarial examples). The overall block diagram is shown below. Figure 1 As shown.

[0051] Step 1: Image acquisition and preprocessing.

[0052] Image acquisition refers to using airborne equipment such as drones or satellites to photograph ground vehicles at different times and under different lighting conditions, ensuring the diversity of image samples. The acquired images should contain clear vehicle outlines, colors, and background information, as well as rich color information.

[0053] Preprocessing may include at least one of the following: converting the image from the RGB color space to the HSV color space to better separate hue, saturation, and brightness information; applying a Gaussian filter to smooth the image, removing noise and irregularities in detail; and adjusting the image's contrast and brightness to ensure the accuracy and consistency of color information.

[0054] Step 2: Color feature extraction and analysis.

[0055] Color feature extraction can include color histogram extraction. Color histogram extraction involves dividing an image into multiple color channels (such as the H, S, and V channels of HSV). A color histogram is calculated for each channel, recording the frequency of each color component.

[0056] Color feature analysis can include color feature calculation and color difference analysis. Color feature calculation includes calculating the average hue, saturation, and brightness values ​​of an image as a description of its overall color characteristics; and calculating the variance and standard deviation of the color distribution to assess the dispersion of color in the image. Color difference analysis includes analyzing the color differences between the target vehicle and the background, particularly the differences in hue and saturation.

[0057] Determining which color components are most critical for distinguishing vehicles from the background is crucial. The present invention provides an implementation method for determining these critical color components as follows.

[0058] (1) Calculate the average values ​​of the hue (H) difference and saturation (S) difference between the target vehicle and the background in the HSV space.

[0059] Tone difference: H dis =|H objective_mean -H back_mean |,H objective_mean H represents the average hue of the target vehicle. back_mean The average color tone of the background.

[0060] Saturation difference: S dis =|S objective_mean -S back_mean |,S objective_mean S represents the average saturation level of the target vehicle. back_mean The average saturation value for the background.

[0061] (2) Determine the key color components. By comparing the magnitude of hue and saturation differences, determine the color components most critical for distinguishing the vehicle from the background. If H dis If S is large, then the hue is the key color component; if S dis If the saturation is high, then saturation is the key color component.

[0062] Step 3: Color-based adversarial patch design, including base color selection, patch shape and size design, and color blending.

[0063] The selection of a base color refers to choosing a color similar to the target vehicle's color as the base color for the anti-patch, based on the results of color difference analysis. The determination of similar colors is as follows.

[0064] (1) Calculate the color mean values ​​of the H, S, and V channels of the image sub-region where the target vehicle is located, and concatenate the three mean values ​​into a three-dimensional feature vector, denoted as HSV. mean =[H objective_mean ,S objective_mean V objective_mean ], H objective_mean S represents the average hue of the target vehicle. objective_mean V represents the average saturation level of the target vehicle. objective_mean The average brightness of the target vehicle.

[0065] (2) HSV mean The average value of the key color components in the image is replaced with the difference value of the key color components, resulting in... If the key color component is hue, then H objective_mean Replace with H dis If the key color component is saturation, then S back_mean Replace with S dis ;

[0066] (3) Replace the As a base color.

[0067] Patch shape and size design refers to designing the shape and size of the anti-patch based on the key feature areas of the target vehicle (such as the roof, windows, etc.). This ensures that the anti-patch can completely cover the key feature areas without affecting the vehicle's normal operation or appearance.

[0068] Color blending refers to using color mixing techniques to blend the base color of an adversary patch with the colors of its surrounding environment. Transparency adjustments can be used to make the patch more harmonious with the background. The specific blending methods are as follows.

[0069] (1) Select multiple sample points from the area surrounding the adversarial patch, and calculate the average value of these sample points in the HSV color space as the representative color of the background. The average color value of the sample points is denoted as B = [H b ,S b V b ], H b S represents the average hue of the sample points. b V represents the average saturation of the sample points. b This represents the average brightness of the sample points.

[0070] (2) Color blending based on transparency. A transparency parameter α is introduced, 0 ≤ α ≤ 1, and the base color is blended using a linear interpolation method. Blend with background color B; the blended color is F = [H] f ,S f V f ]for:

[0071] H f =(1-α)H b +αH objective_mean

[0072] S f =(1-α)S b +αS objective_mean

[0073] V f =(1-α)V b +αV objective_mean

[0074] H f S represents the hue values ​​of the fused sample points. f V represents the saturation value of the fused sample points. f This represents the brightness value of the sample points after fusion.

[0075] Step 4: Generating and applying anti-patch measures.

[0076] Patch generation. Using image processing software (such as Photoshop, GIMP, etc.) or professional printing equipment, generate physical or digital patterns of the adversarial patch according to the designed color scheme and shape size. Ensure that the generated adversarial patch has sufficient accuracy and resolution to guarantee its effectiveness in practical applications.

[0077] Patch Application. Choose an appropriate adhesive material (such as tape, paint, etc.) to ensure the patch adheres firmly to the vehicle surface. Based on the designed shape and size, select an appropriate location on the vehicle surface and apply the patch.

[0078] Step 5: Evaluation and adjustment of results.

[0079] Testing and Evaluation. Images of the vehicles with the adversarial patches were re-captured using an airborne ground-based imaging device. These images were then fed into the vehicle detection model for testing, and the performance changes of the detection model were recorded.

[0080] Performance analysis. Analyze the changes in detection accuracy, false positive rate, and other metrics of the detection model before and after applying the adversarial patch. Evaluate the interference effect of the adversarial patch and determine its applicability in different scenarios and conditions.

[0081] Adjustments and optimizations. Based on the evaluation results, adjust factors such as the color, shape, size, and position of the anti-patch. Repeat steps 4 and 5 until a satisfactory interference effect is achieved.

[0082] The intelligent perception interference method based on color trapping provided by this invention has good physical meaning, strong interpretability, low computational complexity, simple implementation method, and good interference effect.

[0083] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope defined in the claims of the present invention.

Claims

1. A smart perception interference method based on color trapping, characterized in that, include: Image acquisition and preprocessing: Acquire target images taken from the ground by an airborne base station; Image preprocessing includes color correction and noise reduction; Color feature extraction and analysis: Using image processing techniques, extract color features from images, including color distribution, hue, and saturation; Analyze the extracted color features to find patterns that match the target color and background color; Complete the design of color-based adversarial patches: Based on the color feature analysis results, design a color scheme for the adversarial patches so that the patches visually blend with the original image; Generate and apply adversarial patches: Generate adversarial patches based on the color scheme of the adversarial patches; The generated adversarial patch is pasted onto the target surface, and the patch's color is coordinated with the surrounding environment; Color feature extraction includes: dividing the image into H, S, and V channels of HSV; calculating a color histogram for each channel and recording the frequency of occurrence of each color component; Color feature analysis includes color feature calculation and color difference analysis; Color feature calculation includes: calculating the average hue, average saturation, and average brightness values ​​of the image as a description of the overall color features; and calculating the variance and standard deviation of the color distribution to assess the degree of dispersion of color in the image. Color difference analysis includes: analyzing the differences in hue and saturation between the target and the background; and identifying the key color components used to distinguish the target from the background. Identify the key color components used to distinguish the target from the background, including: Calculate the hue difference H between the target and the background in the HSV space respectively. dis And saturation difference S dis , Hue difference H dis H dis =|H objective_mean -H back_mean |,H objective_mean H is the average hue of the target. back_mean The average color tone of the background. Saturation difference S dis :S dis =|S objective_mean -S back_mean |,S objective_mean S represents the average saturation level of the target. back_mean The average saturation of the background; By comparing hue differences H dis And saturation difference S dis The size of H determines the key color components for distinguishing the target from the background. dis >S dis If H, then hue is the key color component; if H dis dis Saturation is the key color component;​ Color-based adversarial patch design includes: Primary color selection: Based on the color difference analysis results, select a color that is similar in tone to the target color as the primary color for the adversarial patch; Patch shape and size design: Based on the key feature areas of the target, design the shape and size of the countermeasure patch to ensure that the countermeasure patch can completely cover the key feature areas without affecting the target's normal driving and appearance; Color blending: Using color mixing technology, the base color of the anti-patch is blended with the colors of the surrounding environment; The determination of similar colors includes: The color mean values ​​of the H, S, and V channels of the image sub-region where the target is located are calculated respectively to obtain the three-dimensional feature vector HSV. mean =[H objective_mean ,S objective_mean V objective_mean ], H objective_mean S represents the average hue of the target. objective_mean V represents the average saturation level of the target. objective_mean The average brightness of the target; HSV mean The average value of the key color components in the image is replaced with the difference value of the key color components, resulting in... and the replacement As a base color.

2. The method according to claim 1, characterized in that, The method further includes: Effect evaluation and adjustment: The target with the anti-countermeasures patch was photographed using an airborne ground-based imaging device. The captured images were then input into the target detection model for testing. Based on the test results, the interference effect of the anti-countermeasures patch was evaluated. Based on the evaluation results, the color scheme of the anti-countermeasures patch was adjusted to improve the interference efficiency.

3. The method according to claim 1, characterized in that, Image preprocessing includes at least one of the following: Convert the image from the RGB color space to the HSV color space to separate hue, saturation, and brightness information; A Gaussian filter is applied to smooth the image and remove noise; Adjust the contrast and brightness of the image to ensure the accuracy and consistency of color information.

4. The method according to claim 1, characterized in that, Color blending processing includes: Multiple sample points are selected from the area surrounding the adversarial patch, and the average value of these sample points in the HSV color space is calculated as the representative color of the background. The average color value of the sample points is denoted as B = [H]. b ,S b V b ], H b S represents the average hue of the sample points. b V represents the average saturation of the sample points. b This represents the average brightness of the sample points. Color blending based on transparency; introducing a transparency parameter α, 0≤α≤1, and using a linear interpolation method to blend the base colors. Blend with background color B; the blended color is F = [H] f ,S f V f ]for: H f =(1-α)H b +αH objective_mean S f =(1-α)S b +αS objective_mean V f =(1-α)V b +αV objective_mean H f S represents the hue values ​​of the fused sample points. f V represents the saturation value of the fused sample points. f This represents the brightness value of the sample points after fusion.

5. An electronic device, characterized in that, The electronic device is used to perform the method as described in any one of claims 1 to 4.

Images