CN-121982640-A - Unmanned aerial vehicle intrusion prediction method and system based on amplitude and phase decoupling, electronic equipment and storage medium

CN121982640ACN 121982640 ACN121982640 ACN 121982640ACN-121982640-A

Abstract

The unmanned aerial vehicle intrusion prediction method based on amplitude and phase decoupling comprises the steps of decomposing each frame of monitoring image into an amplitude component and a phase component, inputting the amplitude component and the phase component into a prediction model, outputting an unmanned aerial vehicle intrusion prediction result, inputting the phase component into a phase network, outputting a predicted value of the phase component, combining the predicted value with an amplitude component of a last frame of monitoring image, performing frequency domain time domain conversion to obtain a first predicted image sequence, inputting the unmanned aerial vehicle monitoring image sequence into the amplitude network, outputting a second predicted image sequence, extracting the amplitude to obtain a predicted value of the amplitude component, combining the predicted value of the phase component with the predicted value of the amplitude component through a fusion device, performing frequency domain time domain conversion to obtain a fusion result, and determining the unmanned aerial vehicle intrusion prediction result according to the fusion result, the first predicted image sequence and the second predicted image sequence. According to the method, the amplitude and the phase are decoupled, and the accuracy of unmanned aerial vehicle intrusion prediction is improved.

Inventors

LI XIAOLI
WANG WEI
DU ZHENLONG
CHEN DONG

Assignees

南京工业大学

Dates

Publication Date: 20260505
Application Date: 20260123

Claims (10)

1. An unmanned aerial vehicle intrusion prediction method based on amplitude and phase decoupling is characterized by comprising the following steps of: acquiring an unmanned aerial vehicle monitoring image sequence, and decomposing each frame of monitoring image into an amplitude component and a phase component; inputting the amplitude component and the phase component into a prediction model, and outputting an unmanned aerial vehicle intrusion prediction result, wherein the prediction model comprises a phase network, an amplitude network and a fusion device; The phase component is input into a phase network, a predicted value of the phase component is output, and after the predicted value of the phase component is combined with an amplitude component of a last frame of monitoring image, frequency domain time domain conversion is carried out to obtain a first predicted image sequence; Inputting the unmanned aerial vehicle monitoring image sequence into an amplitude network, outputting a second predicted image sequence, and extracting the amplitude of the second predicted image sequence to obtain a predicted value of an amplitude component; and combining the predicted value of the phase component with the predicted value of the amplitude component through a fusion device, performing frequency domain time domain conversion to obtain a fusion result, and determining an unmanned aerial vehicle intrusion prediction result according to the fusion result, the first predicted image sequence and the second predicted image sequence.
2. The unmanned aerial vehicle intrusion prediction method based on amplitude and phase decoupling according to claim 1, wherein: The phase component is input into the phase network, and the predicted value of the output phase component comprises: combining the time dimension of the phase component with the channel dimension; the frequency coordinates corresponding to the phase components are normalized and then spliced with the combined phase components in the channel dimension; Inputting the spliced phase components into a phase network to extract time characteristics; And re-separating the time dimension and the channel dimension from the extracted time features to obtain a predicted value of the phase component.
3. The unmanned aerial vehicle intrusion prediction method based on amplitude and phase decoupling according to claim 1, wherein: the amplitude network is including encoder, amplitude module and the decoder that connects gradually, with unmanned aerial vehicle monitoring image sequence input to the amplitude network, and output second predictive image sequence includes: extracting the spatial characteristics of the unmanned aerial vehicle monitoring image sequence by using an encoder; Carrying out Fourier transform on the input spatial features along the spatial dimension by using an amplitude module to obtain a Fourier spectrum and extracting time dynamic features in the Fourier spectrum; And performing image reconstruction based on the interaction relation by using a decoder to obtain a second predicted image sequence.
4. The unmanned aerial vehicle intrusion prediction method based on amplitude and phase decoupling according to claim 3, wherein: extracting temporal dynamic features in the fourier spectrum includes: extracting time dynamic characteristics in a Fourier spectrum by a two-layer complex multi-layer perceptron as follows: Wherein, the To extract the time-dynamic features in the fourier spectrum, The fourier spectrum is represented and, Is a complex number The activation function, MLP () represents a multi-layer perceptron, the first layer MLP performs linear transformation and activation in the time dimension, extracts time features from the historical time steps, and the second layer MLP maps the extracted time features to future time steps, resulting in time dynamic features in the Fourier spectrum.
5. The unmanned aerial vehicle intrusion prediction method based on amplitude and phase decoupling according to claim 1, wherein: The method further comprises the steps of: And calculating the phase loss, the amplitude loss and the weighted sum of the loss of the unmanned aerial vehicle intrusion prediction result and the true value to obtain a composite loss function, and training a prediction model by adopting the composite loss function.
6. The unmanned aerial vehicle intrusion prediction method based on amplitude and phase decoupling according to claim 5, wherein: the phase loss is calculated as follows: Wherein, the In order for the phase to be lost, In order to cut off the frequency of the signal, As a predicted value of the phase component, As the true value of the phase component, Is a frequency domain coordinate.
7. The unmanned aerial vehicle intrusion prediction method based on amplitude and phase decoupling according to claim 5, wherein: the amplitude loss is calculated as follows: Wherein, the In order for the amplitude to be lost, In order to cut off the frequency of the signal, For the time index of the time index, As the frequency domain coordinates of the antenna array, As a predicted value of the amplitude component, Is the true value of the amplitude component, And The mean and standard deviation of the amplitude components in the training set, respectively.
8. An amplitude and phase decoupling-based unmanned aerial vehicle intrusion prediction method system for implementing the amplitude and phase decoupling-based unmanned aerial vehicle intrusion prediction method according to any one of claims 1 to 7, wherein the system comprises: The acquisition and decomposition module is used for acquiring the unmanned aerial vehicle monitoring image sequence and decomposing each frame of monitoring image into an amplitude component and a phase component; The intrusion prediction module is used for inputting the amplitude component and the phase component into a prediction model and outputting an unmanned aerial vehicle intrusion prediction result, and the prediction model comprises a phase network, an amplitude network and a fusion device; The intrusion prediction module comprises a phase prediction module, an amplitude prediction module and a fusion module, wherein, The phase prediction module is used for inputting the phase component into a pre-constructed phase network, outputting a predicted phase sequence, combining the predicted phase sequence with the amplitude component of the last frame of monitoring image, and performing inverse Fourier transform processing to obtain a first predicted image sequence; The amplitude prediction module is used for inputting the unmanned aerial vehicle monitoring image sequence into an amplitude network, outputting a second predicted image sequence, and extracting the amplitude of the second predicted image sequence to obtain a predicted amplitude sequence; the fusion module is used for combining the predicted value of the phase component and the predicted value of the amplitude component through the fusion device, then performing frequency domain time domain conversion to obtain a fusion result, and determining an unmanned aerial vehicle intrusion prediction result according to the fusion result, the first predicted image sequence and the second predicted image sequence.
9. An electronic device comprises a processor and a storage medium, and is characterized in that: The storage medium is used for storing instructions; The processor is configured to operate in accordance with the instructions to perform the steps of the amplitude and phase decoupling based drone intrusion prediction method according to any one of claims 1 to 7.
10. A computer readable storage medium having stored thereon a computer program, characterized in that the program when executed by a processor implements the steps of the unmanned aerial vehicle intrusion prediction method based on amplitude and phase decoupling of any one of claims 1-7.

Description

Unmanned aerial vehicle intrusion prediction method and system based on amplitude and phase decoupling, electronic equipment and storage medium Technical Field The invention belongs to the technical field of combination of security monitoring and artificial intelligence, and particularly relates to an unmanned aerial vehicle intrusion prediction method and system based on amplitude and phase decoupling, electronic equipment and a storage medium. Background The unmanned aerial vehicle intrusion prediction technology which analyzes the current monitoring data through an algorithm, predicts the future unmanned aerial vehicle movement track and evaluates the risk distribution plays a key role in important scenes such as airport clearance area protection, military forbidden area security, and important activity site security management. Despite significant advances in the art, challenges remain in accurately predicting drone position and signal strength. The method mainly derives from the fact that the spatial movement of the unmanned aerial vehicle is closely related to the signal characteristic change, the coupling relationship between the two is tighter due to the complex electromagnetic environment and the changeable flight attitude, and the prediction difficulty is increased. Deep learning technology has been tried in the field of unmanned aerial vehicle monitoring because of its excellent feature extraction and pattern recognition capabilities. The target detection algorithm based on the convolutional neural network CNN can efficiently identify the unmanned aerial vehicle target, and the long-term memory network LSTM can process time sequence data to help analyze the movement trend of the unmanned aerial vehicle. However, existing deep learning models have difficulty in achieving independent modeling of both aspects due to the complex coupling between the spatial trajectory and the variation in signal strength. The position change of the unmanned aerial vehicle is mainly influenced by a flight control system and external air flow, for example, in a strong wind environment, the flight track may deviate from a preset route, the change of the signal strength is closely related to the communication distance, the antenna orientation and the interference environment, and when the unmanned aerial vehicle is far away from a signal source or the direction is adjusted, the signal strength tends to be significantly attenuated. The interaction of the two makes it difficult to accurately predict the intrusion path and threat level, and the traditional single factor or simple correlation analysis model has difficulty in meeting the actual requirements. Disclosure of Invention In order to solve the defects in the prior art, the invention provides an unmanned aerial vehicle intrusion prediction method and system based on amplitude and phase decoupling, electronic equipment and a storage medium, so that the accurate prediction of the position and signal strength of an unmanned aerial vehicle is realized, and the accuracy of the unmanned aerial vehicle intrusion prediction path and threat level is improved. Through intensive research on the characteristics of unmanned aerial vehicle signals in the frequency domain, the situation that the phase change and the spatial position movement have a high correspondence relationship is found, and the amplitude change is closely related to the signal intensity fluctuation. This important finding provides a new idea for solving the problem. The invention adopts the following technical scheme. The first aspect of the invention provides an unmanned aerial vehicle intrusion prediction method based on amplitude and phase decoupling, which comprises the following steps: acquiring an unmanned aerial vehicle monitoring image sequence, and decomposing each frame of monitoring image into an amplitude component and a phase component; inputting the amplitude component and the phase component into a prediction model, and outputting an unmanned aerial vehicle intrusion prediction result, wherein the prediction model comprises a phase network, an amplitude network and a fusion device; The phase component is input into a phase network, a predicted value of the phase component is output, and after the predicted value of the phase component is combined with an amplitude component of a last frame of monitoring image, frequency domain time domain conversion is carried out to obtain a first predicted image sequence; Inputting the unmanned aerial vehicle monitoring image sequence into an amplitude network, outputting a second predicted image sequence, and extracting the amplitude of the second predicted image sequence to obtain a predicted value of an amplitude component; and combining the predicted value of the phase component with the predicted value of the amplitude component through a fusion device, performing frequency domain time domain conversion to obtain a fusion result, and determining an unmanne