A method and apparatus for generating adversarial examples from multi-band UAV remote sensing images

Abstract

This invention proposes a method and apparatus for generating adversarial examples from multi-band UAV remote sensing images, belonging to the field of image processing technology. Based on multi-band UAV remote sensing images, this invention generates adversarial examples of the same shape across multiple bands, which can simultaneously counter multiple single-band intelligent detection methods. Because the generated single-band adversarial examples have consistent shapes, it also facilitates the physical domain implementation of the adversarial examples, which is beneficial for future widespread application.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of image processing, and particularly relates to a multi-band unmanned aerial vehicle remote sensing image adversarial sample generation method and device. BACKGROUND

[0002] With the wide application of remote sensing technology in various fields, the accuracy and security requirements of remote sensing image recognition are also increasingly high. In the experiment, different attack algorithms are used to deceive multiple high-precision RSI recognition models trained on multiple RSI datasets. In order to explore the characteristics of the adversarial examples of RSI scene classification, different scenes are created by testing two main attack algorithms (i.e. fast gradient sign method (FGSM) and basic iterative method (BIM)) trained on different RSI benchmark datasets to deceive convolutional neural network CNN (i.e. InceptionV1, ResNet and simple CNN). The misclassification of adversarial samples of RSI data is related to the similarity of the original class in the CNN feature space. Adversarial examples pose a threat to deep neural network remote sensing scene classification. Simply adding these perturbations on the original high-resolution remote sensing (HRRS) image can generate adversarial samples, which only have slight differences with the original samples. In order to comprehensively evaluate the influence of adversarial examples on remote sensing image (RSI) scene classification, 8 state-of-the-art classification DNNs are tested on 6 RSI benchmarks. In the experiment, four cutting-edge attack algorithms are used to study the influence of adversarial examples on RSI classification. Adversarial examples in remote sensing images are re-examined. Finally, the future research direction is proposed, aiming to establish a more secure machine learning model for adversarial examples in remote sensing images. The ubiquitous adversarial samples in remote sensing data are analyzed without any knowledge from the victim model. Specifically, a new black-box adversarial attack method, Mixup-Attack and its simple variant Mixcut-Attack, is proposed for remote sensing data.

[0003] Adding adversarial samples in data can cause deep neural networks to output unexpected results, thereby threatening the normal recognition and detection of remote sensing systems. The framework adopts a heat map-based explainability method to obtain the key regions of target recognition and generate irregularly shaped natural patches to reduce the interference area. Adversarial samples are generated under white box, and when transferred to attack a remote black box model, the success rate is often very low. Adversarial cloud attacks defend against remote sensing salient object detection. There have been many deep learning methods proposed for salient object detection in remote sensing images, and significant results have been achieved. Clouds are natural and common in remote sensing images, but there are few studies on remote sensing image adversarial attack and defense based on cloud camouflage.

[0004] The method for generating adversarial samples for unmanned aerial vehicle image target detection generally comprises: taking a control point on a vehicle vertically photographed by the unmanned aerial vehicle as a reference control point, marking the shape of a mask within the range of the reference control point; generating a mask according to the mask coordinates, and initializing a general adversarial patch with the mask; calculating a projection matrix by using the corresponding relationship of the control points in the original image and the target image, and performing projection transformation on the general adversarial patch and the mask; repositioning the general adversarial patch and the mask to a specified area of the vehicle by using the projection transformation to obtain an adversarial image; inputting the adversarial image into a target detection model, calculating a loss value by using an attack loss function, and optimizing the general adversarial sample by using a back propagation algorithm. The multi-feature collaborative adversarial attack multi-modal remote sensing image classification uses methods such as the fast gradient sign method (FGSM), the projection gradient descent method (PGD) and the Carlini and Wagner (C&W) attack method to attack the multi-modal satellite remote sensing target classification task, and good perturbation effects are obtained.

[0005] Currently, the existing technology generates adversarial samples based on single-band or RGB visible light unmanned aerial vehicle remote sensing image data, and does not generate adversarial samples based on multi-band (multi-spectral, hyper-spectral, multi-spectral band, cross-spectral) unmanned aerial vehicle image data; although the existing target identification classification technology of satellite remote sensing generates adversarial samples, there is no intelligent adversarial for multi-band unmanned aerial vehicle remote sensing image data for multi-band target detection tasks. SUMMARY

[0006] The present application focuses on the intelligent adversarial of the multi-band unmanned aerial vehicle remote sensing detection target detection task, takes multi-band unmanned aerial vehicle remote sensing aerial image data as the data source, generates multi-band adversarial samples of the same shape by constructing a multi-loss function joint optimization and optimization algorithm, attacks the multi-band image intelligent tasker for multiple times, achieves the best attack effect, and finally obtains the adversarial sample for multi-band unmanned aerial vehicle remote sensing detection, solves the problem of simultaneously attacking the unmanned aerial vehicle remote sensing detection system mounted with multiple different band detectors.

[0007] The first aspect of the present application discloses a method for generating adversarial samples of multi-band unmanned aerial vehicle remote sensing images. The method comprises:

[0008] Step S1, multi-band images of a protected target are collected by a multi-band unmanned aerial vehicle, and single-band images are aligned to align the positions of the protected target in different band images;

[0009] Step S2, based on the multi-band images collected by the multi-band unmanned aerial vehicle, a loss function is calculated by using a machine learning method, and adversarial samples of single-band images are generated by continuous iteration;

[0010] Step S3: Use constraint functions to jointly optimize the loss function so that the success rate of the adversarial examples of multi-band images against the multi-band detection system on the multi-band UAV reaches the target value, and obtain the adversarial examples of multi-band images in this state.

[0011] Step S4: Combining stealth material technology, the adversarial samples of the obtained multi-band images are applied to the surface of the protected target without affecting the original function of the protected target, thereby countering the multi-band detection system on the multi-band UAV.

[0012] According to the method of the first aspect of the present invention, in step S1, the multi-band UAV is a multispectral version of the UAV, the protected target is a pedestrian and / or vehicle, and the multi-band image is a visible light image, a near-infrared image, a mid-infrared image, and a far-infrared image.

[0013] According to the method of the first aspect of the present invention, in step S2, based on the multi-band images acquired by the multi-band UAV, a loss function f is calculated using a machine learning method based on gradient descent or optimization algorithm. 可见光 f 近红外 f 中红外 f 远红外 Adversarial examples x for single-band images are generated through continuous iteration. 可见光 x 近红外 x 中红外 x 远红外 .

[0014] According to the method of the first aspect of the present invention, in step S3, the loss function f is adjusted using the constraint function F. 可见光 f 近红外 f 中红外 f 远红外 Joint optimization is performed to make the adversarial examples of single-band images similar in shape and non-interfering with each other. After multiple optimizations, the success rate of the adversarial examples of multi-band images against the multi-band detection system on the multi-band UAV reaches the target value, and the adversarial examples X of the multi-band images in this state are obtained; wherein, the constraint function F is the minimum value constraint function.

[0015] A second aspect of this invention discloses an adversarial example generation device for multi-band UAV remote sensing images. The device includes:

[0016] The first processing module is configured to: acquire multi-band images of the protected target collected by a multi-band UAV, and perform alignment processing on the single-band images so that the position of the protected target is aligned in different band images;

[0017] The second processing module is configured to: calculate a loss function using machine learning methods based on the multi-band images acquired by the multi-band UAV, and generate adversarial examples of single-band images through continuous iteration;

[0018] The third processing module is configured to: jointly optimize the loss function using constraint functions, so that the success rate of the adversarial examples of multi-band images against the multi-band detection system on the multi-band UAV reaches the target value, and obtain the adversarial examples of multi-band images in this state.

[0019] The fourth processing module is configured to: combine stealth material technology to apply adversarial samples of the obtained multi-band images to the surface of the protected target without affecting the original function of the protected target, thereby countering the multi-band detection system on the multi-band UAV.

[0020] According to the apparatus of the second aspect of the present invention, the multi-band UAV is a multispectral version of the UAV, the protected target is a pedestrian and / or vehicle, and the multi-band image is a visible light image, a near-infrared image, a mid-infrared image, and a far-infrared image.

[0021] According to the apparatus of a second aspect of the present invention, the second processing module is specifically configured to: calculate a loss function f based on the multi-band images acquired by the multi-band UAV, using a machine learning method based on gradient descent or optimization algorithms. 可见光 f 近红外 f 中红外 f 远红外 Adversarial examples x for single-band images are generated through continuous iteration. 可见光 x 近红外 x 中红外 x 远红外 .

[0022] According to the apparatus of the second aspect of the present invention, the third processing module is specifically configured to: apply the constraint function F to the loss function f 可见光 f 近红外 f 中红外 f 远红外 Joint optimization is performed to make the adversarial examples of single-band images similar in shape and non-interfering with each other. After multiple optimizations, the success rate of the adversarial examples of multi-band images against the multi-band detection system on the multi-band UAV reaches the target value, and the adversarial examples X of the multi-band images in this state are obtained; wherein, the constraint function F is the minimum value constraint function.

[0023] A third aspect of this invention discloses an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the adversarial example generation method for multi-band UAV remote sensing images described in the first aspect of this invention.

[0024] A fourth aspect of this invention discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the adversarial example generation method for multi-band UAV remote sensing images described in the first aspect of this invention.

[0025] In summary, the technical solution proposed in this invention can solve the problem of identifying friendly targets by multi-band UAV intelligent detection systems, effectively countering multi-band intelligent detection, and protecting important friendly targets. Based on multi-band UAV remote sensing images, the generated multi-band adversarial samples with the same shape can simultaneously counter multiple single-band intelligent detections. Because the generated single-band adversarial samples have consistent shapes, it also facilitates the physical domain implementation of the adversarial samples, which is beneficial for future promotion and application. Attached Figure Description

[0026] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0027] Figure 1 This is a flowchart illustrating the method for generating adversarial examples from multi-band UAV remote sensing images according to an embodiment of the present invention.

[0028] Figure 2(a) and 2(b) This is a schematic diagram of a multi-band unmanned aerial vehicle according to an embodiment of the present invention.

[0029] Figures 3(a) to 3(e) This is a multi-band UAV remote sensing image according to an embodiment of the present invention.

[0030] Figure 4 This is a schematic diagram of the differential evolution method according to an embodiment of the present invention.

[0031] Figures 5(a) to 5(e) This is a schematic diagram of an adversarial sample image from a multi-band UAV remote sensing system according to an embodiment of the present invention.

[0032] Figure 6 This is a structural diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0034] The first aspect of this invention discloses a method for generating adversarial examples from multi-band UAV remote sensing images. The method includes:

[0035] Step S1: Use a multi-band UAV to acquire multi-band images of the protected target and perform alignment processing on the single-band images so that the protected target is aligned in different band images.

[0036] Step S2: Based on the multi-band images acquired by the multi-band UAV, calculate the loss function using machine learning methods, and generate adversarial examples for single-band images through continuous iteration;

[0037] Step S3: Use constraint functions to jointly optimize the loss function so that the success rate of the adversarial examples of multi-band images against the multi-band detection system on the multi-band UAV reaches the target value, and obtain the adversarial examples of multi-band images in this state.

[0038] Step S4: Combining stealth material technology, the adversarial samples of the obtained multi-band images are applied to the surface of the protected target without affecting the original function of the protected target, thereby countering the multi-band detection system on the multi-band UAV.

[0039] According to the method of the first aspect of the present invention, in step S1, the multi-band UAV is a multispectral version of the UAV, the protected target is a pedestrian and / or vehicle, and the multi-band image is a visible light image, a near-infrared image, a mid-infrared image, and a far-infrared image.

[0040] According to the method of the first aspect of the present invention, in step S2, based on the multi-band images acquired by the multi-band UAV, a loss function f is calculated using a machine learning method based on gradient descent or optimization algorithm. 可见光 f 近红外 f 中红外 f 远红外 Adversarial examples x for single-band images are generated through continuous iteration. 可见光 x 近红外 x 中红外 x 远红外 .

[0041] According to the method of the first aspect of the present invention, in step S3, the loss function f is adjusted using the constraint function F. 可见光 f 近红外 f 中红外 f 远红外 Joint optimization is performed to make the adversarial examples of single-band images similar in shape and non-interfering with each other. After multiple optimizations, the success rate of the adversarial examples of multi-band images against the multi-band detection system on the multi-band UAV reaches the target value, and the adversarial examples X of the multi-band images in this state are obtained; wherein, the constraint function F is the minimum value constraint function.

[0042] Specific example 1 (such as...) Figure 1 (As shown)

[0043] Step S1: Use a multi-band UAV to acquire multi-band images (such as visible light, near, mid, and far infrared images) of the protected target (e.g., pedestrians, vehicles, etc.), and perform alignment processing on the single-band images to align the positions of the protected target in different band images.

[0044] Step S2: Based on the remote sensing images acquired by multi-band UAVs, calculate the loss function (f) using scientific machine learning methods such as gradient descent or optimization algorithms. 可见光 f 近红外 f 中红外 f 远红外 (etc.), continuously iterating to generate single-band image adversarial examples (x 可见光 x 近红外 x 中红远红外 wait);

[0045] Step S3: Apply constraint function F (e.g., minimum value constraint function, etc.) to the loss function (f) 可见光 f 近红外 f 中红外 f 远红外 (etc.) are jointly optimized to make the single-band adversarial samples similar in shape and do not interfere with each other. After multiple optimization iterations, the success rate of the multi-band image adversarial sample attacking the multi-band intelligent detection system is maximized, and finally the multi-band adversarial sample X is obtained.

[0046] Step S4: Combining stealth material technology, the obtained multi-band countermeasure sample X is applied to the surface of the protected target without affecting the original function of the protected target, thereby realizing a multi-band detection intelligent system to counter UAV remote sensing.

[0047] Specific Example 2

[0048] Using the DJI Mavic 3 Multispectral Edition drone, the drone's remote sensing images include visible light RGB images, green (G), red (R), red-edge (RE), and near-infrared (NIR) single-band images, such as...Figure 2(a) and 2(b) As shown.

[0049] Utilizing multi-angle, multi-band UAV imagery data acquisition targeting stealth targets, such as Figures 3(a) to 3(e) The images shown are visible light RGB images, green / G images, near-infrared / NIR images, red / R images, and red-edge / RE images, respectively. The model vehicle is a stealth target that needs protection, and the images are multi-band target images collected by a multi-band UAV.

[0050] like Figure 4 As shown, based on remote sensing images acquired by multi-band UAVs, the loss function (f) is calculated using the differential evolution optimization algorithm. 可见光RGB f 绿 / G波 f 近红外 / NIR f 红 / R f 红边 / RE ), continuously iterating to generate single-band image adversarial examples (x 可见光RGB x 绿 / G波 x 近红外NIR x 红 / R x 红边 / RE Using the minimum constraint function F to adjust the loss function (f) 可见光RGB f 绿 / G波 f 近红外 / NIR f 红 / R f 红边 / RE Joint optimization is performed to make the single-band adversarial samples similar in shape and non-interfering with each other. After multiple optimization iterations, the success rate of multi-band image adversarial samples attacking multi-band intelligent detection systems is maximized, and finally multi-band adversarial sample X is obtained.

[0051] By jointly optimizing multi-band data using differential evolution, a multi-band adversarial sample X with multiple single-band samples of the same shape capable of simultaneously countering multi-band detection is obtained. The results are as follows: Figures 5(a) to 5(e) The images shown are adversarial examples of visible light RGB images, green / G images, near-infrared / NIR images, red / R images, and red-edge / RE images, which can successfully deceive five multi-band intelligent detection systems.

[0052] A second aspect of this invention discloses an adversarial example generation device for multi-band UAV remote sensing images. The device includes:

[0053] The first processing module is configured to: acquire multi-band images of the protected target collected by a multi-band UAV, and perform alignment processing on the single-band images so that the position of the protected target is aligned in different band images;

[0054] The second processing module is configured to: calculate a loss function using machine learning methods based on the multi-band images acquired by the multi-band UAV, and generate adversarial examples of single-band images through continuous iteration;

[0055] The third processing module is configured to: jointly optimize the loss function using constraint functions, so that the success rate of the adversarial examples of multi-band images against the multi-band detection system on the multi-band UAV reaches the target value, and obtain the adversarial examples of multi-band images in this state.

[0056] The fourth processing module is configured to: combine stealth material technology to apply adversarial samples of the obtained multi-band images to the surface of the protected target without affecting the original function of the protected target, thereby countering the multi-band detection system on the multi-band UAV.

[0057] According to the apparatus of the second aspect of the present invention, the multi-band UAV is a multispectral version of the UAV, the protected target is a pedestrian and / or vehicle, and the multi-band image is a visible light image, a near-infrared image, a mid-infrared image, and a far-infrared image.

[0058] According to the apparatus of a second aspect of the present invention, the second processing module is specifically configured to: calculate a loss function f based on the multi-band images acquired by the multi-band UAV, using a machine learning method based on gradient descent or optimization algorithms. 可见光 f 近红外 f 中红外 f 远红外 Adversarial examples x for single-band images are generated through continuous iteration. 可见光 x 近红外 x 中红外 x 远红外 .

[0059] According to the apparatus of the second aspect of the present invention, the third processing module is specifically configured to: apply the constraint function F to the loss function f 可见光 f 近红外 f 中红外 f 远红外 Joint optimization is performed to make the adversarial examples of single-band images similar in shape and non-interfering with each other. After multiple optimizations, the success rate of the adversarial examples of multi-band images against the multi-band detection system on the multi-band UAV reaches the target value, and the adversarial examples X of the multi-band images in this state are obtained; wherein, the constraint function F is the minimum value constraint function.

[0060] A third aspect of this invention discloses an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the adversarial example generation method for multi-band UAV remote sensing images described in the first aspect of this invention.

[0061] Figure 6 A structural diagram of an electronic device according to an embodiment of the present invention is shown below. Figure 6 As shown, the electronic device includes a processor, memory, communication interface, display screen, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, Near Field Communication (NFC), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.

[0062] Those skilled in the art will understand that Figure 6 The structure shown is merely a structural diagram of the part related to the technical solution of this disclosure and does not constitute a limitation on the electronic device to which the solution of this application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0063] A fourth aspect of this invention discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the adversarial example generation method for multi-band UAV remote sensing images described in the first aspect of this invention.

[0064] In summary, the technical solution proposed in this invention can solve the problem of identifying friendly targets by multi-band UAV intelligent detection systems, effectively countering multi-band intelligent detection, and protecting important friendly targets. Based on multi-band UAV remote sensing images, the generated multi-band adversarial samples with the same shape can simultaneously counter multiple single-band intelligent detections. Because the generated single-band adversarial samples have consistent shapes, it also facilitates the physical domain implementation of the adversarial samples, which is beneficial for future promotion and application.

[0065] Please note that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. The above embodiments only illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be pointed out that for those skilled in the art, several modifications and improvements can be made without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for generating adversarial samples of multi-band unmanned aerial vehicle remote sensing images, characterized in that, The method comprises: Step S1, acquiring multi-band images of a protected target by a multi-band unmanned aerial vehicle, and aligning single-band images so that the position of the protected target in different band images is aligned; Step S2, calculating a loss function by a machine learning method based on the multi-band images acquired by the multi-band unmanned aerial vehicle, and generating adversarial samples of the single-band images by continuous iteration; Step S3, jointly optimizing the loss function by a constraint function so that the attack success rate of the adversarial samples of the multi-band images on a multi-band detection system on the multi-band unmanned aerial vehicle reaches a target value, and the adversarial samples of the multi-band images in this state are obtained; Step S4, applying the obtained adversarial samples of the multi-band images to the surface of the protected target by combining stealth material technology, without affecting the original function of the protected target, so as to counter the multi-band detection system on the multi-band unmanned aerial vehicle.

2. The method of claim 1, wherein, In the step S1, the multi-band unmanned aerial vehicle is a multi-spectral version unmanned aerial vehicle, the protected target is a pedestrian and / or a vehicle, and the multi-band images are visible light images, near-infrared images, mid-infrared images, and far-infrared images.

3. The method of claim 2, wherein, In the step S2, based on the multi-band image collected by the multi-band unmanned aerial vehicle, a machine learning method based on gradient descent method or optimization algorithm is used to calculate a loss function f 可见光 , 近红外 , 中红外 , 远红外 Through continuous iteration, the adversarial sample x 可见光 , x 近红外 , x 中红外 , x 远红外 of the single-band image is generated.

4. The multi-band unmanned aerial vehicle remote sensing image adversarial sample generation method according to claim 3, characterized in that, In the step S3, the loss function f is jointly optimized by using a constraint function F 可见光 , f 近红外 , f 中红外 , f 远红外 , so that the shapes of the adversarial samples of the single-band images are similar and do not interfere with each other, and after multiple optimizations, the attack success rate of the adversarial samples of the multi-band images on the multi-band detection system on the multi-band unmanned aerial vehicle reaches a target value, and the adversarial samples X of the multi-band images in this state are obtained; wherein the constraint function F is a minimum constraint function.

5. A multi-band unmanned aerial vehicle remote sensing image adversarial sample generation device, characterized in that, The device comprises: A first processing module configured to acquire multi-band images of a protected target acquired by a multi-band unmanned aerial vehicle, and align single-band images so that the position of the protected target in different band images is aligned; A second processing module configured to calculate a loss function by a machine learning method based on the multi-band images acquired by the multi-band unmanned aerial vehicle, and generate adversarial samples of the single-band images by continuous iteration; A third processing module configured to jointly optimize the loss function by a constraint function so that the attack success rate of the adversarial samples of the multi-band images on a multi-band detection system on the multi-band unmanned aerial vehicle reaches a target value, and the adversarial samples of the multi-band images in this state are obtained; A fourth processing module configured to apply the obtained adversarial samples of the multi-band images to the surface of the protected target by combining stealth material technology, without affecting the original function of the protected target, so as to counter the multi-band detection system on the multi-band unmanned aerial vehicle.

6. The multi-band adversarial sample generation device for unmanned aerial vehicle remote sensing images according to claim 5, characterized in that, The multi-band unmanned aerial vehicle is a multi-spectral version unmanned aerial vehicle, the protected target is a pedestrian and / or a vehicle, and the multi-band images are visible light images, near-infrared images, mid-infrared images, and far-infrared images.

7. The multi-band adversarial sample generation device for unmanned aerial vehicle remote sensing images of claim 6, wherein, The second processing module is specifically configured to: based on the multi-band image collected by the multi-band unmanned aerial vehicle, using a machine learning method based on gradient descent method or optimization algorithm, calculate a loss function f 可见光 、 近红外 、 中红外 、 远红外 , through continuous iteration to generate an adversarial sample x 可见光 、 近红外 、 中红外 、 远红外 of a single-band image.

8. The multi-band adversarial sample generation device for unmanned aerial vehicle remote sensing images of claim 7, wherein, The third processing module is specifically configured to perform joint optimization on the loss function f 可见光 , f 近红外 , f 中红外 , f 远红外 by using a constraint function F, so that the shapes of the adversarial samples of the single-band images are similar and do not interfere with each other, and the attack success rate of the adversarial samples of the multi-band images on the multi-band detection system on the multi-band unmanned aerial vehicle reaches a target value after multiple optimizations, and an adversarial sample X of the multi-band image in this state is obtained; wherein the constraint function F is a minimum constraint function.

9. An electronic device, comprising: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the computer program to realize the method for generating adversarial samples of multi-band unmanned aerial vehicle remote sensing images according to any one of claims 1-4.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to realize the method for generating adversarial samples of multi-band unmanned aerial vehicle remote sensing images according to any one of claims 1-4.

Images