Unmanned vehicle GPS generation type deception jamming method based on target azimuth detection

Abstract

The invention relates to the field of unmanned vehicle GPS interference, in particular to an unmanned vehicle GPS generation type deception jamming method based on target azimuth detection. According to the scheme, the method comprises the steps that a data set is labeled and preprocessed, an SSD-RepVGG network model is constructed, the SSD-RepVGG network model is trained through the labeled and preprocessed data set, target image pixel data are detected through the network model, unmanned vehicle speed information is solved, and unmanned vehicle angle information is calculated according to the camera imaging principle. And carrying out multi-modal fusion on the detection result of the SSD-RepVGG network model and the data information detected by the millimeter wave radar to finally obtain the angle and distance information of the unmanned vehicle. If it is judged that the unmanned vehicle enters the limited area, a forged GPS signal is generated to decoy the unmanned vehicle to drive away according to the planned path, and efficient and high-precision control over the unmanned vehicle is achieved. The method is suitable for unmanned vehicle management and control.

Description

Technical Field

[0001] The present invention relates to the field of GPS interference for unmanned vehicles, and particularly to a GPS generative spoofing interference method for unmanned vehicles based on target azimuth detection. Background Art

[0002] With the rapid development of autonomous driving technology, especially in military exercises and other security-blocked areas, ensuring the security of restricted areas has become an increasingly important issue. The entry of unmanned vehicles into restricted areas may lead to data leakage.

[0003] Traditional security measures usually rely on physical barriers and manual monitoring, but physical barriers and manual monitoring are difficult to meet real-time requirements, with low efficiency and low security.

[0004] Currently, GPS navigation spoofing interference technology is widely applied. For example, a GPS navigation spoofing system disclosed in CN218003735U includes an FPGA chip, a GPS receiver, a radio frequency transceiver, and a clock circuit. The GPS receiver is connected to the FPGA chip through UART, the FPGA chip is connected to the radio frequency transceiver, and the output end of the clock circuit is connected to the FPGA chip and the radio frequency transceiver. The GPS receiver is used to receive external real satellite navigation signals and transmit them to the FPGA chip. The FPGA chip is used to generate simulated satellite navigation signals and forward the simulated satellite navigation signals or real satellite navigation signals to the radio frequency transceiver. The radio frequency transceiver is used to send the simulated satellite navigation signals or real satellite navigation signals to the target receiver. The clock circuit is used to provide a reference clock for the FPGA chip and the radio frequency transceiver.

[0005] The above solution completes the precise control of the target aircraft by sending spoofing signals to the target UAV. However, this solution lacks precise detection and monitoring of the target and cannot perform real-time tracking of the target. Therefore, the accuracy of its control is not high. And the above solution is only applicable to the spoofing control of aircraft (UAVs) and is not applicable to the spoofing control of unmanned vehicles. Summary of the Invention

[0006] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a GPS generative spoofing interference method for unmanned vehicles based on target azimuth detection, achieving efficient and high-precision control of unmanned vehicles.

[0007] The present invention adopts the following technical solutions to achieve the above purpose. The present invention provides a GPS generative spoofing interference method for unmanned vehicles based on target azimuth detection, including:

[0008] S1, dataset annotation and data preprocessing;

[0009] S2, constructing an SSD-RepVGG network model;

[0010] S3. Train the SSD-RepVGG network model with the labeled and preprocessed dataset;

[0011] S4. Process the images captured by the camera;

[0012] S5. Calculate the angle and speed information of the unmanned vehicle;

[0013] S6. Obtain the distance information of the unmanned vehicle through multimodal fusion;

[0014] S7. Use the GPS-generated spoofing interference signal to deceive the GPS information of the unmanned vehicle.

[0015] Furthermore, step S1 specifically includes:

[0016] Use an open-source dataset, perform data augmentation on the dataset, including random rotation, random cropping, color jittering, brightness adjustment, and adding the Mosaic method, expand the dataset, and divide it into a training set, a validation set, and a test set according to a set ratio.

[0017] Furthermore, step S2 specifically includes:

[0018] Adopt the RepVGG module to replace the backbone VGG-16 network as the front network, convert the FC6 layer and FC7 layer in the VGG network from fully connected layers to convolutional layers, and at the same time remove the FC8 layer and the Dropout layer. Replace the 2×2 pooling layer with a stride of 2 in the VGG network with a 3×3 pooling layer with a stride of 1, and add a convolutional layer to the VGG network.

[0019] Furthermore, step S4 specifically includes:

[0020] Use MATLAB software and the Opencv framework to calibrate the camera. After completing the image distortion correction, put the post-processed image into the SSD-RepVGG network model for detection to obtain the pixel position of the unmanned vehicle in the image captured by the camera.

[0021] Furthermore, step S5 specifically includes:

[0022] According to the camera imaging principle, convert the obtained pixel position of the unmanned vehicle into actual angle information, and the expression is:

[0023]

[0024] where α is the horizontal field of view angle of the camera, θ is the azimuth angle where the unmanned vehicle is located, the resolution of the captured image is m×n pixels, x ul ,x lrThey are the horizontal coordinate pixel values of the upper left and lower right corners of the driverless vehicle respectively;

[0025] Meanwhile, calculate the lateral movement speed v of the driverless vehicle according to the detection speed of the SSD-RepVGG network model x .

[0026] Furthermore, step S6 specifically includes:

[0027] Detect the angle θ of the object within the set range through the millimeter-wave radar i ′ , the lateral speed v ′ xi and the distance d i ′ , traverse the angle θ of all objects i ′ , compare θ i ′ with the azimuth angle θ where the driverless vehicle is located, screen out the objects with the angle difference less than the threshold angle, traverse the lateral speed v of all objects ′ xi , compare v ′ xi with the lateral movement speed v of the driverless vehicle x , screen out the objects with the lateral movement speed less than the threshold speed, correctly screen out the driverless vehicle according to the angle and speed, and extract the distance information of the driverless vehicle from the detection results of the millimeter-wave radar.

[0028] Furthermore, step S7 specifically includes:

[0029] If the driverless vehicle enters the restricted area, adopt the PID control method to adaptively adjust the steering and rotation speed of the turntable, so as to achieve the purpose of directional tracking of the driverless vehicle, and generate a pseudo-random code strongly correlated with the real signal and a spoofing navigation message with the same format as the real telegram according to the basic characteristics of the real GPS signal, and send the signal loaded with the spoofing navigation message through the transmitting antenna, so that the GPS receiver of the driverless vehicle receives the false signal.

[0030] The beneficial effects of the present invention are:

[0031] The present invention uses the SSD-RepVGG neural network model to achieve real-time detection of driverless vehicles. In the test scenario, it is necessary to preprocess the images captured by the camera. According to the camera imaging principle, after correcting the distortion of the collected images, the images are input into the trained SSD-RepVGG network model to detect the angle information of the driverless vehicle in the real scenario. The detection results of the SSD-RepVGG network model are fused with the data information detected by the millimeter-wave radar, and finally the angle and distance information of the driverless vehicle are obtained. It is judged that if the driverless vehicle enters the restricted area, a forged GPS signal is generated to lure it to drive away according to the planned path, comprehensively ensuring the safety of the restricted area and realizing the efficient and high-precision control of driverless vehicles. Description of the Drawings

[0032] Figure 1 is the flowchart of the GPS generation deception interference method for driverless vehicles based on target azimuth detection provided by the present invention;

[0033] Figure 2 is the schematic diagram of the SSD-RepVGG network model structure provided by the present invention;

[0034] Figure 3 is the schematic diagram of the RepVGG module architecture provided by the present invention;

[0035] Figure 4 is the schematic diagram of the structural reparameterization of the RepVGG module provided by the present invention;

[0036] Figure 5 is the schematic diagram of the camera imaging principle provided by the present invention;

[0037] Figure 6 is the schematic diagram of the interference result of the driverless vehicle when the dangerous threshold D = 6 meters provided by the present invention;

[0038] Figure 7 is the schematic diagram of the interference result of the driverless vehicle when the dangerous threshold D = 8 meters provided by the present invention;

[0039] Figure 8 is the schematic diagram of the interference result of the driverless vehicle when the dangerous threshold D = 10 meters provided by the present invention;

[0040] Figure 9 is the schematic diagram of the interference result of the driverless vehicle when the dangerous threshold D = 12 meters provided by the present invention. Detailed Embodiments

[0041] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0042] The present invention provides a GPS generation spoofing interference method for unmanned vehicles based on target azimuth detection, as Figure 1 shown, which specifically includes:

[0043] S1. Dataset annotation and data preprocessing;

[0044] The present invention uses the Stanford Cars open-source dataset, which contains 16,185 car images of different models, covering 196 different vehicle categories. To make the test remain robust under complex conditions, we perform data augmentation on the dataset, including random rotation, random cropping, color jittering, brightness adjustment, and adding the Mosaic method, expanding the dataset to 32,370 images, and dividing them in the ratio of training set: validation set: test set = 8:1:1.

[0045] S2. Construct an SSD-RepVGG network model;

[0046] As Figure 3 shown, it is a schematic diagram of the RepVGG (Re-parameterized VGG, VGG improved by structural re-parameterization) module architecture. RepVGG has a simple and efficient network structure and strong feature learning ability, and it can learn rich image features. The SSD (Single Shot MultiBox Detector) algorithm needs to rely on the backbone network to extract effective features to detect and locate targets during the target detection process. Therefore, using the RepVGG module as the backbone network of the SSD algorithm can quickly extract more representative features to improve the detection accuracy of the neural network model for vehicle targets of different sizes, shapes, and categories.

[0047] The structural re-parameterization process of the RepVGG module is as shown in the appendix Figure 4 shown. During training, it is a multi-branch structure, including 3×3, 1×1 convolution branches and an identity mapping branch to enhance the expressiveness. After training, through structural re-parameterization, the weights of the 1×1 convolution and identity mapping branches are converted into the form of 3×3 convolution weights, and then the weights of each branch are added together. Finally, it is simplified to a single 3×3 convolution structure during inference, improving the inference efficiency without losing performance.

[0048] As Figure 2As shown in the figure, in the SSD-RepVGG network model, the present invention uses the RepVGG module to replace the backbone VGG-16 network as the pre-network, converts the FC6 layer and FC7 layer in the VGG network from fully connected layers to convolutional layers, and removes the FC8 layer and the Dropout layer at the same time. In addition, the last pooling layer in the network is replaced from the original 2×2 pooling layer with a stride of 2 to a 3×3 pooling layer with a stride of 1. Four additional convolutional layers are added to the network structure to extract more diverse and rich image features.

[0049] S3. Train the SSD-RepVGG network model with the labeled and preprocessed data set;

[0050] S4. Process the images captured by the camera;

[0051] In actual tests, it is necessary to process the images captured by the camera. Due to the limitations of lens design and manufacturing, the images captured by the camera will have radial distortion and tangential distortion, which will affect the accuracy of calculating the target angle based on the position of the target pixel point. Therefore, the present invention uses the Zhang-Zhengyou calibration method to calibrate the camera with MATLAB software and the Opencv framework. After completing the image distortion correction, the post-processed images are put into the SSD-RepVGG network model for detection to obtain the pixel positions of the unmanned vehicle in the images captured by the camera.

[0052] S5. Calculate the angle and speed information of the unmanned vehicle;

[0053] As shown in the appendix Figure 5 According to the camera imaging principle, the pixel position of the unmanned vehicle obtained above can be converted into actual angle information, and the expression is:

[0054]

[0055] Among them, α is the horizontal field of view angle of the camera, θ is the azimuth angle of the unmanned vehicle, which is what we want. The resolution of the captured image is m×n pixels. x ul , x lr are the horizontal coordinate pixel values of the upper left corner and the lower right corner of the unmanned vehicle respectively.

[0056] At the same time, according to the speed detected by the SSD-RepVGG network model, calculate the lateral movement speed v of the unmanned vehicle x .

[0057] S6. Obtain the distance information of the unmanned vehicle through multi-modal fusion;

[0058] Detect the angle θ of the object within the set range through the millimeter wave radar i ′ , lateral speed v ′xi and the distance d i ′ Traverse the angle θ of all objects i ′ Compare θ i ′ with the azimuth angle θ where the driverless vehicle is located, and filter out the objects with the angle difference less than the threshold angle Ψ. Traverse the lateral speed v of all objects ′ xi Compare v ′ xi with the lateral movement speed v of the driverless vehicle x to filter out the objects with the lateral movement speed less than the threshold speed V. After multi-modal data fusion, the driverless vehicle is correctly filtered out according to the angle and speed, and the distance information of the driverless vehicle is extracted from the millimeter-wave radar detection results.

[0059] S7. Generate a GPS spoofing interference signal.

[0060] GPS signals exist in binary form. GPS generative spoofing interference refers to using a high-precision satellite signal simulator to generate a pseudo-random code strongly correlated with the real signal and a fraudulent navigation message with the same format as the real message according to the basic characteristics of the real GPS signal. By transmitting an antenna to send a signal loaded with the fraudulent navigation message, the GPS receiver of the driverless vehicle receives a false signal. Since the false signal has a stronger power, the receiver misidentifies it as a real signal, and then calculates the wrong azimuth information, ultimately resulting in the azimuth of the driverless vehicle being spoofed and interfered.

[0061] When the driverless vehicle enters the restricted area, the PID control method is used to adaptively adjust the steering and rotation speed of the turntable, so as to achieve the purpose of directional tracking of the driverless vehicle. At the same time, the SDR device generates a forged GPS signal with a matching power, which is amplified by a power amplifier and then radiates a GPS spoofing interference signal through a horn antenna to lure the driverless vehicle to drive away from the restricted area according to the planned path, ensuring the safety of the restricted area.

[0062] The following details the solution of the present invention with specific implementation cases.

[0063] (1) The SSD-RepVGG network model identifies the angle information of the driverless vehicle

[0064] The hardware configuration used in the present invention is an Intel(R) Core(TM) i5-11260H CPU @ 2.40 GHz, 16 GB RAM, an NVIDIA GeForce GTX 3050Ti (4GB) GPU, and a Windows 10 (64-bit) operating system, which is used to train the deep neural network.

[0065] To evaluate the feasibility of the proposed method, the present invention uses precision, recall, average precision, F1 score, and FPS as the main evaluation indicators for the detection performance of the algorithm. The accuracy of the SSD-RepVGG network in identifying autonomous vehicles is verified through ablation experiments, as shown in Table 1. It can be seen from Table 1 that although the VGG16 model is small and has a fast running speed, its detection accuracy is the lowest; the accuracy of the SSD model is improved compared to the VGG16 model, but the detection speed drops significantly. For the SSD-RepVGG network model, since the SSD backbone network is replaced with the RepVGG module, compared with the VGG16 and SSD network models, the proposed SSD-RepVGG model in the present invention achieves a balance between the detection speed and accuracy of autonomous vehicles. The test results show that the SSD-RepVGG network model has a 1.59% higher average detection precision and a 3.91fps higher detection speed than the SSD model.

[0066] Table 1 Comparison of Different Deep Neural Network Models

[0067] Model Accuracy (%) Recall (%) Average Precision (%) F1(％) FPS (fps) VGG16 88.23 92.21 91.54 90.17 52.95 SSD 92.13 96.01 95.17 94.03 42.95 SSD-RepVGG 93.97 97.43 96.76 95.67 46.86

[0068] Use the SSD-RepVGG model to detect autonomous vehicles and obtain their pixel positions in the image. The average error of the autonomous vehicle angle is calculated according to the following expression, and the average error of detecting the autonomous vehicle angle is obtained as 1.68%.

[0069]

[0070] where θ measure represents the angle of the autonomous vehicle detected by the SSD-RepVGG model, θ true represents the true azimuth angle of the autonomous vehicle, and N represents the total number of images of the detected target scene, with N taken as 3237.

[0071] (2) Multimodal Fusion to Obtain the Distance Information of Autonomous Vehicles

[0072] Through the configuration of multiple radar filters and multimodal fusion screening, the autonomous vehicle is uniquely selected with an accuracy of 99.80%. The set thresholds are Ψ = 3.28° and V = 0.76m / s. The following expression is used to evaluate the radar ranging performance, and according to the test data, the average error of the radar detecting the radial distance of the autonomous vehicle is 0.24m.

[0073]

[0074] where d measure is the distance of the millimeter-wave radar detecting the autonomous vehicle, d true is the true radial distance of the autonomous vehicle, and N represents the number of radar detections, with N taken as 500.

[0075] (3) GPS-generated spoofing interference on the orientation of the unmanned vehicle

[0076] According to the PID control principle, the parameters are repeatedly adjusted to achieve the directional tracking of the turntable to the unmanned vehicle. To measure the sensitive power threshold of the SDR device to transmit a matched GPS spoofing interference signal at different distances.

[0077] The sensitive power thresholds of the GPS-generated spoofing interference signal transmitted at different distances are shown in Table 2. The latitude and longitude, the number of L1 satellites, and the orientation information are used as the disturbed observation parameters of the unmanned vehicle. When the driverless vehicle is within the range of 2 - 16 meters from the interference source and not disturbed, (latitude, longitude) ≈ (40.154°, 116.266°), the number of L1 satellites is 11, and the orientation is 292.220°.

[0078] Table 2 Sensitive power thresholds at different distances

[0079]

[0080] Repeated experiments show that the orientation information is the most stable disturbed parameter. Therefore, the criterion of orientation = 0 is adopted as the standard for the complete disturbance of the unmanned vehicle. The least squares method is used to fit the interference distance of the unmanned vehicle and the power of the GPS-generated spoofing signal. As the unmanned vehicle gradually approaches the dangerous area, the interference conditions of the unmanned vehicle under different danger thresholds D are obtained, as shown in the attached Figure 6 - attached Figure 9 As shown, they are the interference results of the unmanned vehicle when the danger threshold D is 6m, 8m, 10m, and 12m respectively. It is verified that the present invention realizes spoofing interference on the GPS signal of the unmanned vehicle and makes it drive away from the restricted area according to the specified route.

[0081] The above is only the preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, should not be regarded as an exclusion of other embodiments, but can be used in various other combinations, modifications, and environments, and can be changed within the scope of the concept described herein through the above teachings or the technology or knowledge in the relevant field. And the changes and modifications made by those skilled in the art without departing from the spirit and scope of the present invention should all be within the protection scope of the appended claims of the present invention.

Claims

1. The GPS generation spoofing interference method for unmanned vehicles based on target azimuth detection, characterized in that, Including: S1. Dataset annotation and data preprocessing; S2. Constructing the SSD-RepVGG network model; S3. Training the SSD-RepVGG network model with the annotated and preprocessed dataset; S4. Processing the images captured by the camera; S5. Calculating the angle and speed information of the unmanned vehicle; S6. Obtaining the distance information of the unmanned vehicle through multimodal fusion; S7. Using the GPS generated spoofing interference signal to spoof the GPS information of the unmanned vehicle.

2. The method for generating spoofing interference of the GPS of an unmanned vehicle based on target azimuth detection according to claim 1, wherein Step S1 specifically includes: Using an open-source dataset, performing data augmentation on the dataset, including random rotation, random cropping, color jitter, brightness adjustment, and adding the Mosaic method to expand the dataset, and dividing it into a training set, a validation set, and a test set according to a set ratio.

3. The method for generating spoofing interference for the GPS of an unmanned vehicle based on target azimuth detection according to claim 1, wherein, Step S2 specifically includes: Adopting the RepVGG module to replace the backbone VGG-16 network as the front network, converting the FC6 layer and FC7 layer in the VGG network from fully connected layers to convolutional layers, removing the FC8 layer and the Dropout layer at the same time, replacing the 2×2 pooling layer with a stride of 2 in the VGG network with a 3×3 pooling layer with a stride of 1, and adding convolutional layers in the VGG network.

4. The method for generating spoofing interference for the GPS of an unmanned vehicle based on target azimuth detection according to claim 1, wherein Step S4 specifically includes: Using MATLAB software, calibrating the camera using the Opencv framework. After completing the image distortion correction, the post-processed image is then put into the SSD-RepVGG network model for detection to obtain the pixel position of the unmanned vehicle in the image captured by the camera.

5. The method for generating spoofing interference for the GPS of an unmanned vehicle based on target azimuth detection according to claim 4, characterized in that, Step S5 specifically includes: According to the camera imaging principle, converting the obtained pixel position of the unmanned vehicle into actual angle information. The expression is: where α is the horizontal field of view angle of the camera, θ is the azimuth angle of the driverless vehicle, the resolution of the captured image is m×n pixels, x ul , x lr are the horizontal coordinate pixel values of the upper left corner and the lower right corner of the driverless vehicle respectively; Meanwhile, calculate the lateral movement speed v of the driverless vehicle according to the detection speed of the SSD-RepVGG network model x .

6. The method for generating spoofing interference of the unmanned vehicle GPS based on target azimuth detection according to claim 5, wherein Step S6 specifically includes: Detect the angle θ of an object within a set range using a millimeter-wave radar i ′ , the lateral velocity v ′ xi and the distance d i ′ , traverse the angle θ of all objects i ′ , compare θ i ′ with the azimuth angle θ of the driverless vehicle, and filter out the objects with an angle difference less than the threshold angle. Traverse the lateral velocity v of all objects ′ xi , compare v ′ xi with the lateral movement velocity v of the driverless vehicle x , filter out the objects with a lateral movement velocity less than the threshold velocity, correctly filter out the driverless vehicle based on the angle and velocity, and extract the distance information of the driverless vehicle from the millimeter-wave radar detection results.

7. The method for generating spoofing interference of the unmanned vehicle GPS based on target azimuth detection according to claim 1, characterized in that Step S7 specifically includes: If the unmanned vehicle enters a restricted area, the PID control method is used to adaptively adjust the steering and rotation speed of the turntable, so as to achieve the purpose of directional tracking of the driverless vehicle. According to the basic characteristics of the real GPS signal, a pseudo-random code strongly correlated with the real signal and a spoofing navigation message with the same format as the real telegram are generated. The signal loaded with the spoofing navigation message is sent through the transmitting antenna, so that the GPS receiver of the unmanned vehicle receives the false signal.

Images