CN-120543893-B - Non-cooperative unmanned aerial vehicle burst frequency point detection method and device based on Seg-Yolo model

CN120543893BCN 120543893 BCN120543893 BCN 120543893BCN-120543893-B

Abstract

The invention relates to the field of signal detection, in particular to a burst frequency point detection method and equipment of a non-cooperative unmanned aerial vehicle based on a Seg-Yolo model. The Seg-Yolo model provided by the invention does not need a priori sequence of the non-cooperative unmanned aerial vehicle signal at all, does not need to predict a specific sequence, and can directly realize continuous, discontinuous, overlapped and non-overlapped signal frequency point detection in the time-frequency diagram. According to the invention, through preprocessing, image noise reduction and a Seg-Yolo model, the image volume to be processed is reduced, the deep learning network scale is simplified, the center frequency point of each frequency point can be precisely locked on the basis of reducing the training time length and improving the training efficiency, and the precision and reliability of extracting signals from complex frequency hopping signals are improved.

Inventors

ZENG MIYI
CAO YONG
CHEN AIPING
WANG NENG
ZHENG DANYANG

Assignees

成都工业职业技术学院

Dates

Publication Date: 20260508
Application Date: 20250411

Claims (5)

1. A non-cooperative unmanned aerial vehicle burst frequency point detection method based on a Seg-Yolo model is characterized by comprising the following steps: s1, acquiring data to be detected, preprocessing the data, and outputting a time-frequency diagram; S2, denoising and binarizing the time-frequency diagram to output a binarized time-frequency diagram; S3, inputting the binarized time-frequency diagram into a pre-constructed Seg-Yolo model; s4, outputting a burst frequency point detection result by the Seg-Yolo model; Wherein, the Seg-Yolo model is composed of a plurality of yolo blocks, and the pre-construction comprises the following steps: a, establishing a Seg-Yolo dataset; b, constructing a Seg-Yolo initial model; c, preprocessing the Seg-Yolo dataset through S1, denoising and binarizing the dataset through S2, and outputting a training set; training the Seg-Yolo initial model through the training set, and outputting the Seg-Yolo model after the model converges; Wherein the number of yolo blocks is 3, including a transverse time axis yolo blocks, a longitudinal frequency domain axis yolo blocks and a comprehensive yolo blocks; The transverse time axis yolo is used for specially predicting time domains and intervals; The longitudinal frequency domain axis yolo is used for specially predicting frequency domains and intervals; the comprehensive yolo block is used for comprehensively predicting the time domain and the frequency domain; Each yolo block comprises a main network, a neck network, a head network and a prediction network which are connected in sequence; The backbone network comprises 9 residual network layers of 1 channel; The neck network comprises a convolution layer and 3 pooling layers; the header network comprises 6 convolution layers with 3×3 convolution kernels and 1 channel number; The prediction network comprises 3 category prediction networks and 3 position prediction networks; Loss function of the Seg-Yolo model The expression of (2) is: , , Wherein, the For the initial value of the loss function, For the confidence of the i yolo th block, As a conventional loss function for the i yolo th block, As a position information loss function, t, f, w, h is a comprehensive position parameter of yolo blocks, L t is a time axis information loss function, and L f is a frequency axis information loss function; The expression of the comprehensive position parameter in the position information loss function is as follows: , , , , Wherein, the The time domain center point, the frequency domain center point, the time domain length and the frequency domain bandwidth are respectively output by the integrated yolo blocks; the time domain center point and the time domain length are respectively output by the transverse time axis yolo blocks; The frequency domain center point and the frequency domain bandwidth are respectively output by the longitudinal frequency axis yolo blocks; the expression of the time axis information loss function and the frequency axis information loss function is as follows: Time axis information loss function: , , frequency axis information loss function: , , wherein X is input data of a composite yolo block, Input data for the transverse timeline yolo block output, As input data for the longitudinal frequency axis yolo block, And For the ith and jth input data.
2. The method for detecting burst frequency points of non-cooperative unmanned aerial vehicle based on Seg-Yolo model according to claim 1, wherein the preprocessing in S1 comprises the following steps: S11, setting conversion parameters, and converting data to be processed into a time-frequency diagram according to the conversion parameters; s12, dividing the time-frequency diagram according to a preset dividing coefficient; S13, inputting the divided time-frequency diagram into a pre-trained lightweight convolution model, and outputting screened data into the time-frequency diagram; The lightweight convolution model comprises three convolution networks with 3×3 convolution kernels and 1 channel number.
3. The method for detecting burst frequency points of non-cooperative unmanned aerial vehicle based on Seg-Yolo model according to claim 2, wherein the pre-training of the lightweight convolution model comprises: Acquiring a preprocessing data set, marking the divided time-frequency diagram after executing the processing of S11 and S12, generating a preprocessing training set, and carrying out model training on the lightweight convolution model through the preprocessing training set; The preprocessing data set comprises unmanned aerial vehicle communication signal samples and railway simulation noise samples in a set proportion, wherein the time-frequency diagram containing effective signals is marked as 1, the time-frequency diagram not containing effective signals is marked as 0, and the lightweight convolution model is used for screening out the time-frequency diagram marked as 1.
4. The method for detecting burst frequency points of non-cooperative unmanned aerial vehicle based on Seg-Yolo model according to claim 2, wherein S2 comprises the following steps: S21, carrying out normalization processing on the time-frequency diagram; s22, denoising the normalized time-frequency diagram according to a preset threshold; S23, performing intelligent clustering bi-classification on the denoised time-frequency diagram through a kmeans intelligent clustering algorithm, and outputting a binarized time-frequency diagram.
5. A Seg-Yolo model-based non-cooperative unmanned aerial vehicle burst frequency point detection device comprising at least one processor and a memory communicatively coupled to the at least one processor, the memory storing instructions executable by the at least one processor to enable the at least one processor to perform the method of any of claims 1 to 4.

Description

Non-cooperative unmanned aerial vehicle burst frequency point detection method and device based on Seg-Yolo model Technical Field The invention relates to the field of signal detection, in particular to a burst frequency point detection method and equipment of a non-cooperative unmanned aerial vehicle based on a Seg-Yolo model. Background The unmanned aerial vehicle is an important role for accelerating intelligent railway construction, and is also a key ring of railway communication networking. In general, unmanned aerial vehicle communication mostly adopts the frequency hopping communication mode, and this mode has better interference resistance's performance, and the secrecy of unmanned aerial vehicle communication has also been guaranteed to the random change of frequency point. However, this also increases the difficulty of supervision of the unmanned aerial vehicle communication, in particular a non-cooperative unmanned aerial vehicle. Non-cooperative unmanned aerial vehicles refer to unmanned aerial vehicles that are not known to the own side, are foreign, invasive, unmanned aerial vehicle individuals or groups that may have malicious attacks, and may pose the following hazards: The method comprises the steps of 1) enabling a non-cooperative unmanned aerial vehicle to impersonate a legal unmanned aerial vehicle identity to invade a railway monitoring section, providing error railway operation information for a railway dispatcher through a technical means, enabling the dispatching to be in fault with railway operation or even accident, 2) enabling the non-cooperative unmanned aerial vehicle to invade the railway operation section, destroying running trains, railway tracks, mountain bodies along the lines and the like, and causing serious railway accidents, 3) enabling the non-cooperative unmanned aerial vehicle to invade a railway key section, monitoring railway environment along the lines and railway operation information in real time, and causing key information leakage, and 4) enabling the non-cooperative unmanned aerial vehicle to impersonate the legal unmanned aerial vehicle identity to achieve legal communication with a railway ground station, invade a railway information network, and cause key information leakage. Therefore, the unmanned aerial vehicle application cutting in tends to cause the safety problem of intelligent railway construction, and the use supervision of the unmanned aerial vehicle joining in the railway communication network, particularly the management of the unmanned aerial vehicle communication frequency spectrum, needs to be enhanced. The frequency hopping communication mode brings obstruction to unmanned aerial vehicle information detection, particularly to a non-cooperative unmanned aerial vehicle, and on the basis of not having the prior frequency hopping frequency spectrum of the non-cooperative unmanned aerial vehicle, the burst frequency point communication signal invaded by the non-cooperative unmanned aerial vehicle is difficult to detect rapidly and accurately. In addition, the complex environment along the railway and the huge noise of the track operation also improve the difficulty of acquiring the communication signals of the unmanned aerial vehicle. The existing unmanned aerial vehicle signal detection technology has the following problems that 1) small-bandwidth frequency hopping signal processing does not have the capability of coping with conditions such as burst frequency points and burst bandwidths, 2) reliability of complex environments is not achieved, 3) overlapping frequency band detection of multiple unmanned aerial vehicles in the same time period cannot be completed, practicability is not achieved, and 4) accuracy of frequency points extracted by an intersection field algorithm based on image and deep learning cannot be guaranteed. Therefore, a method and a device for detecting burst frequency points of a non-cooperative unmanned aerial vehicle, which can adapt to the burst frequency points, have stronger anti-interference capability and higher practicability and precision, are needed. Disclosure of Invention The invention aims to overcome the defects in the prior art and provides a burst frequency point detection method of a non-cooperative unmanned aerial vehicle based on a Seg-Yolo model. In order to achieve the above object, the present invention provides the following technical solutions: A non-cooperative unmanned aerial vehicle burst frequency point detection method based on a Seg-Yolo model comprises the following steps: s1, acquiring data to be detected, preprocessing the data, and outputting a time-frequency diagram; S2, denoising and binarizing the time-frequency diagram to output a binarized time-frequency diagram; S3, inputting the binarized time-frequency diagram into a pre-constructed Seg-Yolo model; s4, outputting a burst frequency point detection result by the Seg-Yolo model; Wherein, the Seg-Yolo model is composed of a plurality o