CN-122027410-A - Unmanned aerial vehicle graph signaling detection and recognition method based on improvement YOLOv8

CN122027410ACN 122027410 ACN122027410 ACN 122027410ACN-122027410-A

Abstract

The invention belongs to the technical field of wireless communication signal processing, and relates to an unmanned aerial vehicle image transmission signal detection and recognition method based on improvement YOLOv, which is used for collecting unmanned aerial vehicle image transmission signals and interference data; generating a time-frequency diagram through short-time Fourier transformation, dimension reduction and gray conversion and marking, carrying out data enhancement by adopting frequency shift, noise addition and multi-signal fusion to construct an enhanced marking set, improving YOLOv8 a model, changing a convolution kernel of a CBS module into a non-square shape to adapt to signal morphology, adding a multi-branch variable receptive field unit in a Bottleneck module to extract multi-scale characteristics, training by utilizing enhanced data to obtain a detection model, preprocessing a signal to be detected, inputting the signal to be detected into the model, and synchronizing the central frequency, bandwidth and unmanned plane type of an output signal. The invention realizes the automatic extraction and identification of the end-to-end signal characteristics, and improves the detection precision and robustness under complex scenes.

Inventors

WANG FANFAN
WANG WENHAO
SONG YIXIN
ZENG YUJIE

Assignees

成都能通科技股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (10)

1. An unmanned aerial vehicle graph signaling signal detection and recognition method based on improvement YOLOv is characterized by comprising the following steps: S1, acquiring signal original data and interference data of a training unmanned aerial vehicle graph under multiple scenes; S2, processing the training unmanned aerial vehicle image signal original data according to preset preprocessing logic to obtain training gray scale image data, marking the unmanned aerial vehicle image signal area in the training gray scale image data, and generating initial marking data; s3, performing scene expansion processing of adapting the signal transmission characteristics of the unmanned aerial vehicle on the initial annotation data based on the interference data to obtain enhanced annotation data; S4, constructing an improved YOLOv model, adapting to morphological characteristics of unmanned aerial vehicle image transmission signals by adjusting parameters and structures of a characteristic extraction module of a backbone network of the YOLOv model, and training the improved YOLOv model by utilizing the enhanced marking data to obtain an unmanned aerial vehicle image transmission signal detection and identification model; S5, acquiring a radio frequency signal to be detected, processing the radio frequency signal to be detected according to the preset preprocessing logic of S2 to obtain gray scale image data to be detected, inputting the gray scale image data to be detected into the unmanned aerial vehicle image signal detection and identification model, and outputting the center frequency, bandwidth and corresponding unmanned aerial vehicle type of the unmanned aerial vehicle image signal.
2. The method for detecting and identifying the unmanned aerial vehicle map signaling based on the improvement YOLOv according to claim 1, wherein the training unmanned aerial vehicle map signaling raw data and the interference data in the step S1 include: Acquiring signal original data of the unmanned aerial vehicle graph for training from multiple dimensions of unmanned aerial vehicle attributes, signal parameters, transmission distances and working frequency bands; and acquiring interference data in the unmanned aerial vehicle signal environment, wherein the interference data is used for simulating an electromagnetic environment interference scene in practical application.
3. The unmanned aerial vehicle map signal detection and recognition method according to claim 1, wherein the pre-processing logic in step S2 comprises time-frequency conversion, dimension reduction processing and gray scale conversion, and the pre-processing logic is executed as follows: Performing time-frequency conversion processing on the training unmanned aerial vehicle graph signal original data to obtain time-frequency domain data; performing dimension reduction processing on the time-frequency domain data to reduce data redundancy; And converting the time-frequency domain data after the dimension reduction into gray scale image data.
4. The unmanned aerial vehicle map signaling detection and identification method based on improvement YOLOv according to claim 3, wherein the labeling process in step S2 is: And selecting an area corresponding to the unmanned aerial vehicle image signal in the training gray map data through a marking tool frame, marking the position range information of the area, and forming initial marking data.
5. The unmanned aerial vehicle map signal detection and recognition method based on the improvement YOLOv according to claim 1, wherein the step S3 comprises: processing training unmanned aerial vehicle graph signal original data corresponding to the initial annotation data in a frequency shift conversion mode, and adapting to a signal frequency fluctuation scene; Processing the training unmanned aerial vehicle graph signal original data corresponding to the initial annotation data by adopting a noise adding processing mode, and adapting to electromagnetic interference scenes; Carrying out fusion processing on the training unmanned aerial vehicle graph signal original data corresponding to the initial annotation data and the interference data by adopting a multi-signal fusion mode, and adapting to a multi-target signal superposition scene; And processing the signal subjected to any one of the processing according to the preset preprocessing logic of the S2, marking the region information corresponding to the processed signal, and generating the enhanced marking data.
6. The unmanned aerial vehicle map signal detection and recognition method based on the improvement YOLOv according to claim 1, wherein the step S4 adjusts the feature extraction module parameters and structures of the YOLOv model backbone network, including: the convolution kernel morphological parameters of a CBS module in the YOLOv model backbone network are adjusted, a non-square convolution kernel is adopted to replace an original square convolution kernel, and the rectangular morphology of the unmanned aerial vehicle image signal is adapted; the structure of the CBS module comprises a convolution layer, a batch normalization layer and an activation function layer.
7. The unmanned aerial vehicle map signal detection and recognition method based on improvement YOLOv according to claim 6, wherein the step S4 adjusts the feature extraction module structure of the YOLOv model backbone network, further comprising: optimizing a Bottleneck module structure in the YOLOv model backbone network, and adding a variable receptive field unit in the Bottleneck module; the variable receptive field unit comprises a plurality of paths of characteristic extraction branches with different receptive field sizes, and unmanned aerial vehicle image transmission signals with different parameters are adapted by carrying out fusion processing on the output characteristics of the plurality of paths of branches; And superposing the fused features with the input data of the Bottleneck module to finish feature reinforcement.
8. The unmanned aerial vehicle map signal detection and recognition method based on improvement YOLOv according to claim 1, wherein the training process in step S4 comprises: dividing the enhanced annotation data into a training set, a verification set and a test set; setting model training parameters, selecting a loss function adapting to a target detection task, and performing iterative training on the improved YOLOv model; After each training round, evaluating the model performance through the verification set, and stopping training when the model performance meets the preset stable condition; And performing performance verification on the model after training stopping by using the test set, and obtaining the unmanned aerial vehicle map signaling detection and identification model after verification.
9. The unmanned aerial vehicle graph signal detection and recognition method based on the improvement YOLOv according to claim 3, wherein the time-frequency conversion processing adopts a short-time Fourier transform mode, the signals are segmented through a window function, and the time-frequency joint characteristics of the signals are reserved.
10. The method for detecting and identifying the unmanned aerial vehicle map signal based on the improvement YOLOv according to claim 5, wherein in the multi-signal fusion mode, the fusion ratio of the training unmanned aerial vehicle map signal raw data and the interference data is in a preset interval, and the value range of the preset interval is 0.2 to 0.8.

Description

Unmanned aerial vehicle graph signaling detection and recognition method based on improvement YOLOv8 Technical Field The invention belongs to the technical field of wireless communication signal processing, and particularly relates to an unmanned aerial vehicle image signal detection and recognition method based on improvement YOLOv. Background Currently, unmanned aerial vehicle type identification is one of the key technologies in low-altitude security, countering systems and spectrum supervision. Currently, the main current unmanned aerial vehicle type recognition can be divided into recognition based on radio frequency signals, recognition based on radar, recognition based on photoelectricity/vision and recognition based on acoustics, and the recognition method based on radio frequency signals belongs to passive detection and has long acting distance, and becomes the most main current method. With the development of unmanned aerial vehicle technology, signals of the unmanned aerial vehicle are more and more complex and diversified, so that accurate identification of the type of the unmanned aerial vehicle signals has become a challenge. The current method for identifying the signal type of the unmanned aerial vehicle is that: based on the traditional signal processing and feature engineering method, the core flow is 1) signal preprocessing (down-conversion, filtering and the like), 2) feature extraction (time domain features, frequency domain features, modulation domain features, protocol features and the like), and 3) classifier identification (support vector machine, decision tree, random forest, K nearest neighbor and the like). Based on protocol analysis and behavior fingerprinting, this approach focuses more on the "behavior pattern" of the communication. The core is to extract a unique identifier or a behavior model by intercepting signals and analyzing communication protocols (frame structures, handshake processes, retransmission mechanisms and the like) so as to identify the signal type of the unmanned aerial vehicle. The end-to-end recognition method based on deep learning is to directly input a signal into a deep learning network for feature learning and classification after the signal is converted into a time-frequency diagram (two-dimensional time-frequency image, intensity is represented by color), and a typical deep learning method is to use a network model such as ResNet, VGG, mobileNet. However, the method based on traditional signal processing and feature engineering is highly dependent on expert experience, complex features are required to be designed manually, and when facing a novel unmanned aerial vehicle, the previous features are not used and need to be redesigned, the method based on protocol analysis and behavior fingerprint is only effective on the type of unmanned aerial vehicle with a known or reversible protocol, once the protocol is encrypted or changed, the method fails, the current main stream method only can identify the type of unmanned aerial vehicle, parameters such as center frequency, bandwidth and the like can not be output, and meanwhile, the method can not adapt to the scene of multiple unmanned aerial vehicles. Disclosure of Invention In order to solve the problems in the background art, the invention provides an unmanned aerial vehicle graph signal detection and recognition method based on an improvement YOLOv, which aims to solve the problem of low unmanned aerial vehicle graph signal detection and recognition accuracy. In order to achieve the purpose, the invention provides the following technical scheme that the unmanned aerial vehicle graph signal detection and recognition method based on the improvement YOLOv comprises the following steps: S1, acquiring signal original data and interference data of a training unmanned aerial vehicle graph under multiple scenes; S2, processing the training unmanned aerial vehicle image signal original data according to preset preprocessing logic to obtain training gray scale image data, marking the unmanned aerial vehicle image signal area in the training gray scale image data, and generating initial marking data; s3, performing scene expansion processing of adapting the signal transmission characteristics of the unmanned aerial vehicle on the initial annotation data based on the interference data to obtain enhanced annotation data; S4, constructing an improved YOLOv model, adapting to morphological characteristics of unmanned aerial vehicle image transmission signals by adjusting parameters and structures of a characteristic extraction module of a backbone network of the YOLOv model, and training the improved YOLOv model by utilizing the enhanced marking data to obtain an unmanned aerial vehicle image transmission signal detection and identification model; S5, acquiring a radio frequency signal to be detected, processing the radio frequency signal to be detected according to the preset preprocessing logic of S2 to obtain gr