A UAV detection method based on multi-sensor fusion

Abstract

The application discloses a kind of multiple sensor fusion's unmanned plane detection method, belong to unmanned plane detection technical field.The application uses radar, sound sensor and optical sensor to carry out multilevel detection to the target to be identified, obtains radar feature vector, sound feature vector, optical feature vector;Then three kinds of feature vectors are multimode fusion, and comprehensive feature vector is obtained;Finally, the comprehensive feature vector is classified using machine learning algorithm, and the type and attribute of the unmanned plane are identified;In addition, based on three kinds of feature vectors, more accurate unmanned plane position coordinates can be obtained to prevent missed detection and false detection.The application can improve the efficiency and accuracy of the target to be identified by multiple sensor fusion, while the type and attribute of the target to be identified can be identified, suitable for unmanned plane detection in various complex environments, and has wide application prospect.

Description

Technical Field

[0001] This invention belongs to the field of UAV detection technology, specifically relating to a UAV detection method that integrates multiple sensors. Background Technology

[0002] Existing drone detection methods are mainly based on single sensors, such as radar, acoustic sensors, and optical sensors. However, each of these sensors has its own advantages and disadvantages and cannot meet the comprehensive needs of drone detection. For example, radar can provide long-range, all-weather detection capabilities, but its detection effect is poor for low-altitude, low-speed drones, and it cannot provide image information of drones; acoustic sensors can provide short-range, low-altitude detection capabilities, but its detection effect is poor for high-altitude, high-speed drones, and it is greatly affected by noisy environments; optical sensors can provide high-resolution, visual detection capabilities, but their detection effect is poor for long-range, nighttime drones, and it is greatly affected by lighting conditions. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes a multi-sensor fusion method for UAV detection. This method leverages the complementary advantages of various sensors to improve the efficiency and accuracy of UAV detection. It can also identify the type and attributes of UAVs and is applicable to UAV detection in various complex environments, showing broad application prospects.

[0004] The technical solution adopted in this invention is as follows:

[0005] A method for UAV detection using multi-sensor fusion, characterized by comprising the following steps:

[0006] Step 1: Use radar to scan the airspace, extract radar feature vectors based on the received radar echo signals, and send the radar feature vectors to the data processing center; the radar feature vectors contain the distance and velocity information of the target to be identified;

[0007] Step 2: When the radar detects a target to be identified within a first preset distance, an optical sensor is used to further detect the target, obtain the optical feature vector of the target, and send the optical feature vector to the data processing center; the optical feature vector includes the shape information, size information, color information, texture information, and marking information of the target;

[0008] Step 3: When the radar detects the target to be identified within the second preset distance, the sound sensor is used to further detect the target to be identified, obtain the sound feature vector of the target to be identified, and send the sound feature vector to the data processing center; the sound feature vector contains the voiceprint features and location coordinate information of the target to be identified;

[0009] Step 4: The data processing center determines the UAV's position coordinates based on the received radar feature vector, optical feature vector, and acoustic feature vector; at the same time, it performs multimodal fusion of the three feature vectors to obtain a comprehensive feature vector, and uses machine learning algorithms to classify the comprehensive feature vector to identify the type and attributes of the target to be identified; finally, it outputs the detection results.

[0010] Furthermore, the first preset distance is 0.5km to 2km; the second preset distance is within 500m.

[0011] The advantages of this invention lie in its multi-layered detection of the target using radar, acoustic sensors, and optical sensors. Each sensor's signal undergoes preprocessing and feature extraction to obtain its own feature vector. These feature vectors are then fused using a multimodal approach to obtain a comprehensive feature vector. Finally, a machine learning algorithm is used to classify the comprehensive feature vector, identifying the type and attributes of the UAV. Furthermore, based on these three feature vectors, more accurate UAV position coordinates can be obtained, preventing missed detections and false detections. This invention, through multi-sensor fusion, improves the efficiency and accuracy of target detection, while also identifying the type and attributes of the target. It is suitable for UAV detection in various complex environments and has broad application prospects. Attached Figure Description

[0012] Figure 1 This is a flowchart of a UAV detection method that integrates multiple sensors according to the present invention. Detailed Implementation

[0013] To make the objectives, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0014] This embodiment of a multi-sensor fusion UAV detection method is implemented based on the following system, which includes: radar, optical sensors, sound sensors, and a data processing center.

[0015] The radar is used to scan the airspace and acquire radar feature vectors containing rough position and velocity information of the UAV. The radar used in this embodiment is a pulse-Doppler radar, which works by emitting a series of short pulses, receiving the reflected pulses, and calculating the target's range and velocity based on the time delay and Doppler frequency shift of the received pulses.

[0016] The optical sensor is used to further detect the drone detected by the radar and obtain an optical feature vector containing image information of the drone. The optical sensor used in this embodiment is a visible light camera. Its working principle is to receive reflected light from the target through a lens, convert the light signal into an electrical signal, convert it from analog to digital to a digital image, and then extract features from the digital image to obtain the optical feature vector.

[0017] The sound sensor is used to further detect the drone detected by the radar and obtain a sound feature vector containing the drone's acoustic signature and position coordinate information. In this embodiment, the sound sensor is a multi-microphone array. The multi-microphone array is used to perform triangulation on the drone to obtain its position coordinate information. Simultaneously, the high-speed rotation of the drone's propellers during flight generates a certain amount of noise. Different drones have different acoustic signatures; therefore, acoustic signature features can also be extracted from the received drone noise.

[0018] The data processing center receives radar feature vectors, acoustic feature vectors, and optical feature vectors, determines the UAV's position coordinates based on these three feature vectors, performs multimodal fusion on the received feature vectors to obtain a comprehensive feature vector, then uses machine learning algorithms to classify the comprehensive feature vector to identify the UAV's type and attributes, and finally outputs the detection results.

[0019] The workflow of the UAV detection method based on multi-sensor fusion in this embodiment is as follows: Figure 1 As shown:

[0020] Step 1: Use radar to scan the airspace within 10km. When there is a target to be identified in the scanned area, the short pulse emitted by the radar is reflected by the target. The radar receives the reflected radar echo signal and calculates the radar feature vector containing the range and velocity information of the target based on the time delay and Doppler frequency shift of the radar echo signal.

[0021] Specifically, the steps for obtaining radar feature vectors are as follows:

[0022] Step 1.1: Use a matched filter to filter the radar echo signal to improve the signal-to-noise ratio, enhance the target echo, and suppress noise and interference;

[0023] Step 1.2: Use a constant false alarm rate detector to enhance the filtered radar echo signal in order to improve detection sensitivity and reduce false alarms and missed alarms.

[0024] Step 1.3: Use a phase calibrator to calibrate the enhanced radar echo signal to eliminate phase errors and improve the measurement accuracy of azimuth and elevation angles;

[0025] Step 1.4: Using a pulse compressor, feature extraction is performed on the time delay and Doppler shift of the calibrated radar echo signal to obtain a radar feature vector containing azimuth information, elevation information, radar cross-section information, and range and velocity information of the target to be identified; then the radar feature vector is sent to the data processing center.

[0026] Step 2: When the radar signal detects that the UAV is within 2km, the optical sensor further detects the UAV detected by the radar, receives the reflected light signal of the target to be identified, converts the reflected light signal into an electrical signal, and then converts it into a digital image through analog-to-digital conversion. The digital image is then used to extract features to obtain an optical feature vector containing the shape information, size information, color information, texture information, and marking information of the target to be identified.

[0027] Specifically, the steps to obtain the optical feature vector are as follows:

[0028] Step 2.1: Use a Gaussian filter to filter the digital image, remove noise and blur, and improve the image's clarity and contrast.

[0029] Step 2.2: Use histogram equalization to enhance the denoised image, adjust the brightness and contrast of the image, and improve the visual effect and information content of the image.

[0030] Step 2.3: Use a distortion calibrator to calibrate the enhanced image, eliminate image distortion and color difference, and improve the geometric and color realism of the image;

[0031] Step 2.4: Using a convolutional neural network, feature extraction is performed on the pixel and color values ​​of the calibrated image to obtain an optical feature vector containing shape, size, color, texture, and marking information of the target to be identified. This optical feature vector is then sent to the data processing center.

[0032] Step 3: When the radar signal detects that the UAV is within 500 meters, multiple microphone arrays are used to perform triangulation on the sound emitted by the UAV to obtain a sound feature vector containing the voiceprint features and location coordinates of the target to be identified; then the sound feature vector is sent to the data processing center.

[0033] Step 4: The data processing center determines the UAV's position coordinates based on the received radar feature vector, optical feature vector, and acoustic feature vector; at the same time, it performs multimodal fusion of the three feature vectors to obtain a comprehensive feature vector, and uses machine learning algorithms to classify the comprehensive feature vector to identify the type and attributes of the target to be identified; finally, it outputs the detection results.

[0034] The data processing center performs the following multimodal fusion and classification steps:

[0035] Step 4.1: Use a multimodal fusion algorithm to fuse radar feature vectors, acoustic feature vectors, and optical feature vectors to obtain a comprehensive feature vector. Multimodal fusion of three different feature vectors can leverage their complementary advantages to improve the integrity and reliability of the features.

[0036] Step 4.2: Input the comprehensive feature vector into the classification and recognition model to determine the type and attributes of the target to be identified.

[0037] The classification and recognition model is trained in the following way:

[0038] Step 4.2.1: Perform model-level fusion of the radar feature vector database, optical feature vector database, and acoustic feature vector database using a multimodal feature fusion method to obtain a comprehensive feature vector database;

[0039] Step 4.2.2: Use a support vector machine to classify and identify the elements in the comprehensive feature vector database to obtain a trained classification and identification model.

Claims

1. A multi-sensor fusion unmanned aerial vehicle detection method, characterized in that, The method comprises the following steps: Step 1, scanning the airspace using a radar, extracting a radar feature vector from the received radar echo signal, and sending the radar feature vector to a data processing center; the radar feature vector contains distance information and velocity information of the target to be identified; Step 2, when the radar detects that the target to be identified is within a first preset distance, further detecting the target to be identified using an optical sensor, obtaining an optical feature vector of the target to be identified, and sending the optical feature vector to the data processing center; the optical feature vector contains shape information, size information, color information, texture information, and logo information of the target to be identified; Step 3, when the radar detects that the target to be identified is within a second preset distance, further detecting the target to be identified using a sound sensor, obtaining a sound feature vector of the target to be identified, and sending the sound feature vector to the data processing center; the sound feature vector contains voiceprint features and position coordinate information of the target to be identified; Step 4, the data processing center determines the position coordinates of the unmanned aerial vehicle according to the received radar feature vector, optical feature vector, and sound feature vector; meanwhile, the three feature vectors are fused in multiple modes to obtain a comprehensive feature vector, the comprehensive feature vector is classified by using a machine learning algorithm, the type and attribute of the target to be identified are identified, and finally a detection result is output.

2. The unmanned aerial vehicle detection method of claim 1, wherein, The data processing center undergoes the following steps of multi-modal fusion and classification: Step 4.1, using a multi-modal fusion algorithm to fuse the radar feature vector, the sound feature vector, and the optical feature vector in multiple modes to obtain a comprehensive feature vector; Step 4.2, inputting the comprehensive feature vector into a classification and identification model to determine the type and attribute of the target to be identified; The classification and identification model is obtained by the following training method: Step 4.2.1, performing model-level fusion in a multi-modal feature fusion method on a radar feature vector database, an optical feature vector database, and a sound feature vector database to obtain a comprehensive feature vector database; Step 4.2.2, using a support vector machine to classify and identify elements in the comprehensive feature vector database to obtain a trained classification and identification model. 3.The unmanned aerial vehicle detection method of claim 2, wherein, In step 1, the steps of obtaining the radar feature vector are as follows: Step 1.1, using a matched filter to filter the radar echo signal to improve the signal-to-noise ratio, enhance the target echo, and suppress noise and interference; Step 1.2, using a constant false alarm rate detector to enhance the filtered radar echo signal to improve the detection sensitivity and reduce false alarms and missed alarms; Step 1.3, using a phase calibrator to calibrate the enhanced radar echo signal to eliminate phase errors and improve the measurement accuracy of the azimuth angle and the elevation angle; Step 1.4, using a pulse compressor to extract the time delay and Doppler shift of the calibrated radar echo signal to obtain a radar feature vector containing distance information and velocity information of the target to be identified. 4.The unmanned aerial vehicle detection method of claim 3, wherein, In step 2, the steps of obtaining the optical feature vector are as follows: Step 2.1, using a Gaussian filter to filter a digital image to remove noise and blur in the image and improve the clarity and contrast of the image; Step 2.2, using histogram equalization, the image after denoising is enhanced, the brightness and contrast of the image are adjusted, and the visual effect and information amount of the image are improved; Step 2.3, using distortion calibrator, the enhanced image is calibrated, the distortion and chromatic aberration of the image are eliminated, and the geometric and color authenticity of the image is improved; Step 2.4, using convolutional neural network, the pixel value and color value of the calibrated image are feature extracted, and an optical feature vector containing shape information, size information, color information, texture information and mark information of the to-be-identified target is obtained.

5. The unmanned aerial vehicle detection method of claim 1, wherein, The first preset distance is 0.5km-2km; the second preset distance is 200m-500m.

Images