Remote sensing image target detection deception method and system

Abstract

The invention provides a remote sensing image target detection deception method and system, and the method comprises the following steps: obtaining the feature information of a to-be-detected target in a remote sensing image, and extracting the first N colors with the highest occurrence frequency; obtaining a remote sensing image background image, carrying out statistical analysis on colors of the remote sensing image background image, and determining and selecting first N colors with the highest occurrence frequency; performing image texture simulation to obtain a gray-scale map of the texture; obtaining a color pattern according to the obtained 2N colors and texture grey-scale maps; determining the arrangement position of the color pattern to form a deception image; and inputting the deception image into the algorithm model for detecting the target, performing detection and identification, if the target can be detected, returning to modify the perlin noise parameter, and if the target cannot be detected, completing target deception. The method is high in calculation speed and good in hiding effect.

Description

Technical Field

[0001] The invention belongs to the technical field of remote sensing image processing and computer vision, and in particular relates to a remote sensing image target detection deception method and system. Background Art

[0002] With the rapid development of remote sensing technology, ground-based aerial images are increasingly used in military reconnaissance, traffic monitoring, urban planning and other fields. However, this widespread application also brings the risk of target exposure. Especially in the military field, the enemy may use advanced image processing and target detection technology to identify our important vehicle targets from remote sensing images, thus posing a potential threat to us.

[0003] To address this risk, researchers have proposed a variety of methods to detect deception. Among them, although the deep learning-based methods have achieved certain results, they often require a large amount of computing resources and training data, and have poor real-time performance, making it difficult to meet the needs of practical applications. In addition, the complexity and opacity of deep learning models also limit their application in certain sensitive areas. Summary of the invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a remote sensing image target detection deception method and system. The solution of the present invention can solve the problems existing in the above-mentioned prior art.

[0005] The technical solution of the present invention:

[0006] According to a first aspect, a remote sensing image target detection deception method is provided, comprising the following steps:

[0007] Obtain the feature information of the target to be detected in the remote sensing image, extract the top N colors with the highest frequency of occurrence, and record them as tc1, tc2, ..., tcN;

[0008] The pixel value of the area where the target to be detected is located in the remote sensing image is set to 0, and the rest remains unchanged to obtain the remote sensing image background image. The color of the remote sensing image background image is statistically analyzed to determine the top N colors with the highest frequency of occurrence, which are recorded as bc1, bc2, ..., bcN;

[0009] Use Perlin noise to simulate image texture, obtain texture map, and get the grayscale map of texture;

[0010] Obtain a color pattern based on the obtained 2N color and texture grayscale images;

[0011] According to the position of the target to be detected in the remote sensing image, the complexity of the background environment and the distribution of other targets, combined with the requirements of the color pattern layout position, the layout position of the color pattern is determined so that the color pattern effectively covers the target to be detected and forms a deceptive image without interfering with the normal image interpretation;

[0012] The deception image is input into the algorithm model of the target detection for detection and identification. If the target can be detected, the perlin noise parameters are modified. If the target cannot be detected, the target deception is completed.

[0013] Furthermore, the characteristic information of the target to be detected in the remote sensing image includes the size and color of the target.

[0014] Furthermore, the method for obtaining the color pattern is:

[0015] The grayscale values ​​in the texture grayscale image are sorted from small to large and divided into M intervals at equal intervals, and the grayscale value in the ii-th interval is replaced with the ii-th selected color, and the selected colors are sorted as tc1, tc2, ..., tcN, bc1, bc2, ..., bcN.

[0016] Furthermore, the color pattern arrangement requirement is: the size of the color pattern does not exceed half the size of the target to be detected, and the pixel values ​​of the color pattern that meet the requirement are replaced with the pixel values ​​of the same size at the top of the target to be detected.

[0017] According to a second aspect, a remote sensing image target detection deception system is provided, including a feature information acquisition unit, a remote sensing image processing unit, a color pattern acquisition unit, a deception image acquisition unit and a detection unit, wherein the feature information acquisition unit acquires feature information of a target to be detected in a remote sensing image, extracts the top N colors with the highest frequency of occurrence, and sends the colors to the color pattern acquisition unit; the remote sensing image processing unit sets the pixel value of the area where the target to be detected in the remote sensing image is located to 0, and keeps the rest unchanged, obtains a remote sensing image background image, performs statistical analysis on the color of the remote sensing image background image, determines to select the top N colors with the highest frequency of occurrence, and sends the colors to the color pattern acquisition unit, and the color pattern acquisition unit uses Perlin noise for image texture analysis. The method comprises the following steps: performing a physical simulation to obtain a texture map, and obtaining a grayscale map of the texture, and combining the obtained 2N colors to obtain a color pattern, which is then sent to a deception image acquisition unit; the deception image acquisition unit determines the arrangement position of the color pattern according to the position of the target to be detected in the remote sensing image, the complexity of the background environment, and the distribution of other targets, combined with the requirements of the arrangement position of the color pattern, so that the color pattern effectively covers the target to be detected without interfering with the normal image interpretation, forms a deception image, and sends it to the detection unit; the detection unit has a built-in algorithm model for detecting the target, inputs the deception image into the algorithm model for detecting the target, performs detection and recognition, and if the target can be detected, returns to modify the Perlin noise parameters; if the target cannot be detected, the target deception is completed.

[0018] The beneficial effects of the present invention compared with the prior art are as follows:

[0019] The present invention does not need to use a deep learning method, the replaced color is close to the background, the number of replaced pixels is relatively small, the calculation speed is fast and the concealment effect is good. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The included drawings are used to provide a further understanding of the embodiments of the present invention, which constitute a part of the specification, are used to illustrate the embodiments of the present invention, and together with the text description, explain the principles of the present invention. Obviously, the drawings in the following description are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0021] Figure 1 A schematic diagram of the steps of a remote sensing image target detection deception method provided according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0022] It should be noted that, in the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0024] Unless otherwise specifically stated, the relative arrangement of the parts and steps described in these embodiments, numerical expressions and numerical values ​​do not limit the scope of the present invention. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the accompanying drawings are not drawn according to the actual proportional relationship. The technology, method and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but in appropriate cases, the technology, method and equipment should be regarded as a part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once a certain item is defined in an accompanying drawing, it does not need to be further discussed in subsequent drawings.

[0025] According to a first aspect of the present invention, an embodiment provides a remote sensing image target detection deception method, comprising the following steps:

[0026] Obtain the feature information of the target to be detected in the remote sensing image, extract the top N colors with the highest frequency of occurrence, and record them as tc1, tc2, ..., tcN;

[0027] The pixel value of the area where the target to be detected is located in the remote sensing image is set to 0, and the rest remains unchanged to obtain the remote sensing image background image. The color of the remote sensing image background image is statistically analyzed to determine the top N colors with the highest frequency of occurrence, which are recorded as bc1, bc2, ..., bcN;

[0028] Use Perlin noise to simulate image texture, obtain texture map, and get the grayscale map of texture;

[0029] Obtain a color pattern based on the obtained 2N color and texture grayscale images;

[0030] According to the position of the target to be detected in the remote sensing image, the complexity of the background environment and the distribution of other targets, combined with the requirements of the color pattern layout position, the layout position of the color pattern is determined so that the color pattern effectively covers the target to be detected and forms a deceptive image without interfering with the normal image interpretation;

[0031] The deception image is input into the algorithm model of the target detection for detection and identification. If the target can be detected, the perlin noise parameters are modified. If the target cannot be detected, the target deception is completed.

[0032] Through the above steps, the detection deception of the target to be detected in remote sensing can be completed without using deep learning methods for pattern design. The calculation is simple and fast, and the real-time performance is strong.

[0033] Further in one embodiment, the feature information of the target to be detected in the remote sensing image includes the size of the target and the color of the target.

[0034] Further in one embodiment, the method of obtaining the color pattern is:

[0035] Sort the grayscale values ​​in the texture grayscale image from small to large and divide them into M intervals at equal intervals, replace the grayscale value in the ii-th interval with the ii-th selected color, ii≤2N, and the selected colors are sorted as tc1, tc2, ..., tcN, bc1, bc2, ..., bcN.

[0036] Further in one embodiment, the color pattern arrangement requirement is: the size of the color pattern does not exceed half the size of the target to be detected, and the pixel values ​​of the color pattern that meet the requirement are replaced with the pixel values ​​of the same size at the top of the target to be detected.

[0037] According to the second aspect of the embodiment, a remote sensing image target detection deception system is provided, including a feature information acquisition unit, a remote sensing image processing unit, a color pattern acquisition unit, a deception image acquisition unit and a detection unit, wherein the feature information acquisition unit acquires feature information of a target to be detected in a remote sensing image, extracts the top N colors with the highest frequency of occurrence, and sends the colors to the color pattern acquisition unit; the remote sensing image processing unit sets the pixel value of the area where the target to be detected in the remote sensing image is located to 0, and keeps the rest unchanged, obtains a remote sensing image background map, performs statistical analysis on the colors of the remote sensing image background map, determines to select the top N colors with the highest frequency of occurrence, and sends the colors to the color pattern acquisition unit, and the color pattern acquisition unit uses Perlin noise for image texture modeling. The texture map is simulated, and the grayscale map of the texture is obtained, and the color pattern is obtained by combining the obtained 2N colors, and the color pattern is sent to the deception image acquisition unit; the deception image acquisition unit determines the layout position of the color pattern according to the position of the target to be detected in the remote sensing image, the complexity of the background environment and the distribution of other targets, combined with the requirements of the color pattern layout position, so that the color pattern effectively covers the target to be detected without interfering with the normal image interpretation, forms a deception image, and sends it to the detection unit; the detection unit has a built-in algorithm model for detecting the target, inputs the deception image into the algorithm model for detecting the target, performs detection and recognition, and if the target can be detected, returns to modify the perlin noise parameters, and if the target cannot be detected, the target deception is completed.

[0038] In order to have a further understanding of the remote sensing image target detection deception method provided by the present invention, a detailed description is given below with reference to specific examples and drawings.

[0039] In a specific embodiment, the target to be detected is a vehicle target, such as Figure 1 As shown, the method proposed in the present invention is mainly carried out in the following four steps when implemented:

[0040] Step 1: Vehicle target analysis and feature extraction

[0041] 1.1 Input preparation: Receive remote sensing image I as input, with a size of 800*600 pixels.

[0042] 1.2 Image segmentation: The remote sensing image I is processed using image segmentation techniques, such as threshold-based segmentation, edge detection combined with region growing, etc., to separate the vehicle target. This process outputs a vehicle target sub-image T with a size of 30*20 pixels to ensure the integrity and accuracy of the vehicle target.

[0043] 1.3 Color feature extraction: Perform color histogram statistics on the vehicle target sub-image T to generate a color histogram T_color_hist. According to this histogram, select the first N colors with the highest frequency, for example, N = 30, recorded as tc1, tc2, ..., tcN, these colors will be used for the design of color patterns.

[0044] Step 2: Background environment analysis and color selection

[0045] 2.1 Background extraction: The pixel values ​​of the area where the vehicle target is located in the remote sensing image I are set to 0 (or a specific mark value), and the rest remain unchanged, thereby obtaining a background sub-image B. The size of B is consistent with that of image I, 800*600 pixels.

[0046] 2.2 Background color feature analysis: Perform color histogram statistics on the background sub-image B to generate a color histogram B_color_hist. According to this histogram, select the first N colors with the highest frequency, for example, N = 30, recorded as bc1, bc2, ..., bcN, these colors will be used for the design of color patterns.

[0047] Step 3: Color pattern design and generation

[0048] 3.1 Texture design: The Perlin noise algorithm is used to generate the basic texture grayscale image text_base. Perlin noise parameters (such as frequency, amplitude, etc.) can be adjusted as needed to generate a variety of texture effects.

[0049] 3.2 Color mapping: Replace the grayscale values ​​in the base texture grayscale image text_base with the colors tc1, tc2, ..., tcN, bc1, bc2, ..., bcN selected in steps 1.3 and 2.2 according to linear mapping to generate the texture color image text_color. This pattern will be used for subsequent fusion operations.

[0050] The linear mapping of this step is to sort the grayscale values ​​in the base texture grayscale image text_base from small to large, and divide it into several intervals at equal intervals according to the minimum and maximum grayscale values, for example, 40 intervals. The grayscale value falling in the ii-th (i=1,2,...,40) interval is replaced by the ii-th selected color. The selected colors are sorted as tc1, tc2,..., tcN and bc1, bc2,..., bcN.

[0051] Step 4: Pattern fusion and effect verification

[0052] 4.1 Pattern fusion: The texture color map text_color is fused with the original remote sensing image I. The specific operation is to directly cover the text_color to the top of the vehicle target in I to obtain the deceptive image I_adv.

[0053] When performing fusion in this step, please note that the texture color image text_color should be scaled first, which should not exceed half the size of the vehicle target. The scaled image is recorded as text_color_scale. Then, the pixel values ​​of the top of the vehicle target that are equal to the size of the text_color_scale image are all replaced with the pixel values ​​of text_color_scale.

[0054] 4.2 Effect verification: Input the deceptive image I_adv into the vehicle detection model (such as YOLOv5) for detection and recognition. Record the detection results of the model to determine whether the model can be successfully deceived.

[0055] 4.3 Iterative optimization: If the model can still correctly detect the vehicle target, return to step 3.1, adjust the parameters of Perlin noise (such as increasing the frequency, changing the amplitude, etc.), regenerate the texture grayscale image text_base, and repeat steps 3.2 to 4.2 until a satisfactory deception effect is achieved (that is, the model cannot correctly detect the vehicle target).

[0056] End: When the model cannot correctly detect the vehicle target in the deceptive image I_adv, the algorithm ends and outputs the final deceptive image and its related parameters.

[0057] By implementing the above four steps, the present invention can realize a remote sensing image vehicle target detection deception method based on color interference. The method does not need to use deep learning methods for pattern design, has fast calculation speed and good concealment effect, and has important application value.

[0058] In summary, the remote sensing image target detection deception method provided by the present invention has at least the following advantages over the prior art:

[0059] The calculation pattern provided by the present invention has a usage time of seconds, while the depth calculation pattern used in the prior art has a usage time of grades or hours. The present invention has a short calculation time, a fast calculation speed, and a good concealment effect.

[0060] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A remote sensing image target detection deception method, characterized in that: The following steps are involved: Obtain the feature information of the target to be detected in the remote sensing image, extract the top N colors with the highest frequency of occurrence, and record them as tc1, tc2, ..., tcN; The pixel value of the area where the target to be detected is located in the remote sensing image is set to 0, and the rest remains unchanged to obtain the remote sensing image background image. The color of the remote sensing image background image is statistically analyzed to determine the top N colors with the highest frequency of occurrence, which are recorded as bc1, bc2, ..., bcN; Use Perlin noise to simulate image texture, obtain texture map, and get the grayscale map of texture; Obtain a color pattern based on the obtained 2N color and texture grayscale images; According to the position of the target to be detected in the remote sensing image, the complexity of the background environment and the distribution of other targets, combined with the requirements of the color pattern layout position, the layout position of the color pattern is determined so that the color pattern effectively covers the target to be detected and forms a deceptive image without interfering with the normal image interpretation; The deception image is input into the algorithm model of the target detection for detection and identification. If the target can be detected, the perlin noise parameters are modified. If the target cannot be detected, the target deception is completed.

2. A remote sensing image target detection deception method according to claim 1, characterized in that: The characteristic information of the target to be detected in the remote sensing image includes the size and color of the target.

3. A remote sensing image target detection deception method according to claim 2, characterized in that: The method for obtaining the color pattern is: The grayscale values ​​in the texture grayscale image are sorted from small to large and divided into M intervals at equal intervals, and the grayscale value in the ii-th interval is replaced with the ii-th selected color, and the selected colors are sorted as tc1, tc2, ..., tcN, bc1, bc2, ..., bcN.

4. A remote sensing image target detection deception method according to claim 1, characterized in that: The requirements for the arrangement of the color pattern are: the size of the color pattern shall not exceed half the size of the target to be detected, and the pixel values ​​of the color pattern that meet the requirements shall be replaced with the pixel values ​​of the same size at the top of the target to be detected.

5. A remote sensing image target detection deception system implementing any one of claims 1 to 4, characterized in that: The invention comprises a feature information acquisition unit, a remote sensing image processing unit, a color pattern acquisition unit, a deception image acquisition unit and a detection unit. The feature information acquisition unit acquires feature information of a target to be detected in a remote sensing image, extracts the top N colors with the highest frequency of occurrence, and sends the colors to the color pattern acquisition unit. The remote sensing image processing unit sets the pixel value of the area where the target to be detected in the remote sensing image is located to 0, and keeps the rest unchanged, obtains a remote sensing image background map, performs statistical analysis on the color of the remote sensing image background map, determines to select the top N colors with the highest frequency of occurrence, and sends the colors to the color pattern acquisition unit. The color pattern acquisition unit uses Perlin noise to simulate the image texture, obtains a texture map, and obtains to the grayscale image of the texture, and combined with the obtained 2N colors to obtain a color pattern, which is sent to a deception image acquisition unit; the deception image acquisition unit determines the layout position of the color pattern according to the position of the target to be detected in the remote sensing image, the complexity of the background environment and the distribution of other targets, combined with the requirements of the color pattern layout position, so that the color pattern effectively covers the target to be detected without interfering with the normal image interpretation, forms a deception image, and sends it to the detection unit; the detection unit has a built-in algorithm model for detecting the target, inputs the deception image into the algorithm model for detecting the target, performs detection and identification, and if the target can be detected, returns to modify the perlin noise parameters, and if the target cannot be detected, completes the target deception.

Images