Intelligent identification method and system based on unmanned aerial vehicle countering

Abstract

The invention discloses an intelligent identification method and system based on unmanned aerial vehicle countering, and relates to the technical field of unmanned aerial vehicles, and the method comprises the steps: calculating a modal initial weight based on an MLP supervised learning method, calculating an attention weight through a Transform attention mechanism, and fusing a feature vector through a collaborative attention mechanism based on the modal initial weight and the attention weight. By introducing a collaborative attention mechanism and combining MLP supervised learning and a Transform attention mechanism, adaptive modal weight adjustment based on environment vectors and threat intelligence is realized, the problem that the recognition precision of traditional fixed weight fusion in a complex environment is reduced is effectively solved, the recognition capability of a disguise unmanned aerial vehicle is enhanced, the false alarm rate and the missing report rate are remarkably reduced, and the recognition efficiency of the disguise unmanned aerial vehicle is improved. The robustness of feature extraction in diversified scenes is ensured, and the environmental adaptability, the recognition precision and the anti-deception capability of the unmanned aerial vehicle countering system are improved.

Description

Technical Field

[0001] The present invention relates to the field of drone technology, and in particular to an intelligent recognition method and system based on drone countermeasures. Background Art

[0002] In recent years, drone countermeasure technology has developed rapidly in the fields of aviation safety, military defense and public safety. With the popularization of drone technology and the widespread application of low-cost drones, the demand for intelligent identification and countermeasures of drones has become increasingly urgent. Existing technologies mainly rely on multimodal sensor fusion and deep learning models to achieve drone detection and classification. However, existing technologies are still limited in the dynamics of multimodal fusion, environmental adaptability and detection capabilities of camouflaged targets. Traditional multimodal fusion methods lack sufficient consideration of environmental changes and dynamic relationships between modalities, resulting in reduced recognition accuracy in complex environments (such as strong electromagnetic interference or low visibility). Existing methods are ineffective in dealing with camouflaged drones or abnormal behaviors. When using drones, there is a lack of effective anti-spoofing mechanisms, and they are easily affected by interference signals or forged features, with high false alarm and missed alarm rates. Existing drone countermeasure technologies have problems with insufficient environmental adaptability and limited camouflaged target recognition capabilities in terms of multimodal feature fusion and anti-spoofing detection. Our invention introduces a collaborative attention mechanism, combines MLP supervised learning with the Transformer attention mechanism to dynamically fuse radar, radio frequency, visible light, and infrared modal features, and adaptively adjusts weights based on environmental vectors and threat intelligence, thereby solving the dynamic problem of feature fusion in complex environments, and improving anti-spoofing capabilities through consistency detection and anomaly analysis, significantly improving recognition accuracy and countermeasure efficiency. Summary of the Invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the present invention provides an intelligent recognition method and system based on drone countermeasures to solve the problems of insufficient environmental adaptability and limited camouflaged target recognition capability in existing drone countermeasure technologies in terms of multimodal feature fusion and anti-spoofing detection.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides an intelligent identification method based on drone countermeasures, which includes collecting real-time environmental parameters, generating environmental vectors, combining the environmental vectors with threat intelligence to calculate an initial threat level, and generating a multimodal data set based on the initial threat level; Use short-time Fourier transform to extract the radar echo time-frequency spectrum. Input the time-frequency spectrum into the MobileNetV3 model to extract radar signal features. Use the 1D-CNN model to extract radio frequency signal features. Use the EfficientNet-B0 model to extract visual features of visible light images. Use IRNet to extract thermal imaging features of infrared images. Then generate a set of feature vectors. The initial modal weights are calculated based on the MLP supervised learning method, the attention weights are calculated through the Transformer attention mechanism, and the feature vectors are fused using the collaborative attention mechanism based on the initial modal weights and the attention weights. Based on the fused feature vector, the Transformer model is used for real-time classification and threat level generation. The countermeasure priority is calculated based on the threat level and real-time classification results. Anti-spoofing detection results and countermeasure instructions are generated based on the countermeasure priority. Based on anti-spoofing detection results and countermeasure instructions, dynamic countermeasure instruction execution and effect feedback optimization are performed.

[0006] As a preferred solution of the intelligent identification method based on drone countermeasures described in the present invention, wherein: the real-time environmental parameters are collected, the environmental vector is generated, the environmental vector is combined with the threat intelligence to calculate the initial threat level, and the multimodal data set is generated based on the initial threat level. The system is deployed at the drone countermeasure ground station, equipped with millimeter wave radar, radio frequency receiver, high-resolution visible light camera, long-wave infrared camera, light sensor, electromagnetic interference detector, weather station, collects real-time environmental parameters, generates environmental vector , and normalize it to get the environment vector , combining environmental vectors with threat intelligence to calculate the initial threat level ,according to , adjust the sampling frequency, based on the multimodal data after adjusting the sampling frequency, use the high-precision GPS clock to align the data and generate a data set with a unified timestamp .

[0007] As a preferred solution of the intelligent recognition method based on drone countermeasures described in the present invention, the short-time Fourier transform is used to extract the radar echo time-frequency spectrum, the time-frequency spectrum is input into the MobileNetV3 model to extract the radar signal features, the 1D-CNN model is used to extract the features of the radio frequency signal, and a feature vector set is generated to refer to the radar echo. Perform feature extraction and generate a time-frequency spectrum using short-time Fourier transform (STFT). The spectrum is input to the pre-trained MobileNetV3 model. The network extracts the spatial and frequency features of the radar signal through convolution operations and outputs a fixed-dimensional feature vector. ; For RF signals Feature extraction is performed to calculate the signal energy and the spectral entropy , when the electromagnetic noise level N( ) > , the RF signal is severely interfered, the signal energy has reduced reliability, increasing the spectral entropy weight, enhancing the representation of the signal frequency distribution, using a 1D-CNN model to output the feature vector of the RF signal ; Feature extraction is performed on the visible light image , if the light intensity <u, skip the processing to avoid invalid calculations, otherwise use the pre-trained EfficientNet-B0 model to extract the high-order vision of the visible light image , use IRNet to extract the thermal imaging features of the infrared image , generate a set of feature vectors through adaptive feature extraction .

[0008] As a preferred solution of the intelligent recognition method based on UAV countermeasure described in this invention, wherein: the feature vectors are fused by the collaborative attention mechanism based on the modal initial weight and the attention weight refers to calculating the modal initial weight based on the MLP supervised learning method , calculating the data-driven attention weight through the Transformer attention mechanism , adopting the collaborative attention mechanism through the initial weights of each modality , adaptively adjusting according to the environment vector and the threat level , using the data-driven attention weight , based on the dynamic importance of the modal features , multiplying and as the comprehensive weight, and performing weighted summation with the corresponding modal features to generate the fused feature vector .

[0009] ​As a preferred solution of the intelligent identification method based on drone countermeasures described in the present invention, the real-time classification and threat level generation based on the fused feature vector using the Transformer model refers to the training of a fully connected neural network through a drone data set, using cross-entropy loss, inputting the time series fusion features into a lightweight Transformer model, analyzing the drone behavior pattern to generate an intent vector, and calculating the dynamic threat level through a fully connected layer to reflect the real-time threat level. The contrast loss is combined with the cross-entropy loss to optimize the model to enhance the ability to distinguish camouflaged targets, and the results output category probability, dynamic threat level and intent vector.

[0010] As a preferred solution of the intelligent identification method based on drone countermeasures described in the present invention, the countermeasure priority is calculated based on the threat level and classification results, and the anti-spoofing detection results and countermeasure instructions are generated based on the countermeasure priority, which means calculating the cosine similarity between the features of each modality, evaluating the feature consistency, and if the average consistency score is lower than the preset threshold, marking it as a potential camouflaged target, using the One-Class SVM model to analyze the fused feature vector, detect abnormal behavior, and output a negative value to indicate an abnormality. The countermeasure priority is calculated based on the threat level and classification results. If the priority exceeds the threshold and there is no abnormality, a countermeasure instruction containing the target position, category and threat level is generated and sent to the radio frequency jammer. If camouflage is detected, the weight of the abnormal modality is reduced and the fusion feature vector step is returned to re-fusion, and the output result is the anti-spoofing detection result and countermeasure instruction.

[0011] As a preferred solution of the intelligent identification method based on drone countermeasures described in the present invention, the dynamic countermeasure instruction execution and effect feedback optimization based on anti-spoofing detection results and countermeasure instructions refers to using an adaptive countermeasure execution and feedback optimization method to generate an optimized countermeasure strategy, activating the radio frequency jammer to perform the countermeasure operation according to the target position and category in the countermeasure instruction, collecting the countermeasure effect data in real time, and comparing it with the category probability and threat level to evaluate the countermeasure success rate. If the countermeasure fails and camouflage is detected, the modal weight is adjusted, the fusion feature vector is regenerated and reclassified. If successful, the effect data is recorded to update the training data set of the Transformer model, and the optimized countermeasure strategy and effect evaluation report are output as a result to achieve dynamic closed-loop countermeasure optimization.

[0012] In a second aspect, the present invention provides an intelligent identification system based on drone countermeasures, comprising: The environmental perception and threat initialization module is used to collect real-time environmental parameters, generate environmental vectors, and calculate the initial threat level based on threat intelligence to generate a multimodal data set; Multimodal feature extraction module, used to extract features of each modality and generate a set of feature vectors; The feature fusion and weight calculation module is used to combine the normalized environment vector and the initial threat level of the environment perception module and generate a fused feature vector using the collaborative attention mechanism; Classification and threat assessment module, which uses the Transformer model and fully connected neural network for real-time classification and threat level assessment; The anti-spoofing detection and countermeasure decision module is used to fuse the feature vectors and features of each modality of the feature fusion module to perform anti-spoofing detection and generate countermeasure instructions.

[0013] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the intelligent identification method based on drone countermeasures as described in the first aspect of the present invention is implemented.

[0014] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent identification method based on drone countermeasures as described in the first aspect of the present invention.

[0015] The beneficial effects of the present invention are as follows: by introducing a collaborative attention mechanism, combining MLP supervised learning and Transformer attention mechanism, adaptive modal weight adjustment based on environmental vectors and threat intelligence is realized, effectively solving the problem of reduced recognition accuracy of traditional fixed weight fusion in complex environments (such as strong electromagnetic interference or low visibility). By calculating the cosine similarity between modalities and One-Class SVM anomaly detection, the recognition ability of camouflaged drones is enhanced, and the false alarm rate and missed alarm rate are significantly reduced. In view of the insufficient generalization ability of single modal feature extraction, this solution optimizes the radar, radio frequency, visible light and infrared feature extraction processes, and combines STFT, 1D-CNN, MobileNetV3, EfficientNet-B0 and IRNet models to ensure the robustness of feature extraction in diverse scenarios, significantly improving the environmental adaptability, recognition accuracy and anti-deception ability of the drone countermeasure system, and providing a more efficient and reliable solution for aviation safety and public safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 This is a flow chart of the intelligent identification method based on drone countermeasures in Example 1.

[0018] Figure 2 This is a schematic diagram of the structure of the intelligent recognition system based on drone countermeasures in Example 1.

[0019] Figure 3 This is a diagram of the multimodal feature fusion data structure of the intelligent recognition method based on drone countermeasures in Example 1. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0023] Example 1, with reference to Figures 1 to 3 , which is the first embodiment of the present invention, provides an intelligent identification method based on drone countermeasures, comprising the following steps: S1. Collect real-time environmental parameters and generate environmental vectors. The environmental vectors are combined with threat intelligence to calculate the initial threat level, and a multimodal data set is generated based on the initial threat level. Use short-time Fourier transform to extract the radar echo time-frequency spectrum. Input the time-frequency spectrum into the MobileNetV3 model to extract radar signal features. Use the 1D-CNN model to extract radio frequency signal features. Use the EfficientNet-B0 model to extract visual features of visible light images. Use IRNet to extract thermal imaging features of infrared images. Then generate a set of feature vectors. Specifically, real-time environmental parameters are collected to generate environmental vectors. The environmental vectors are combined with threat intelligence to calculate the initial threat level, and a multimodal data set is generated based on the initial threat level. The system (referring to a comprehensive hardware and software platform for drone detection, identification and countermeasures) is deployed at the drone countermeasure ground station, equipped with millimeter-wave radar (detection distance, speed, angle), radio frequency receiver (capture communication and control signals), high-resolution visible light camera (RGB image), long-wave infrared camera (thermal imaging), light sensor (TSL2561, accuracy ±5 lux), electromagnetic interference detector (spectrum analyzer, accuracy ±0.1 dBm), weather station (visibility meter, accuracy ±0.1 km), collect real-time environmental parameters, generate environmental vectors , and normalize it to get the environment vector : , in, is the discrete time point of the current moment, t is the time variable, i is the time series index, Is the light intensity, indicating the current moment The ambient light level, reflecting day and night or weather conditions (such as sunny or foggy), is directly measured by the light sensor. is the electromagnetic noise level, indicating the current moment The electromagnetic environment interference intensity reflects the reliability of the radio frequency signal and is measured by an electromagnetic interference detector. is visibility, indicating the current moment Atmospheric visibility affects the performance of optical and infrared sensors and is measured by the visibility meter at the weather station. is the wind speed, indicating the current moment The wind intensity, which affects the drone trajectory and radar signal stability, is measured by the wind speed sensor at the weather station; Calculate the initial threat level by combining environmental vectors with threat intelligence (external information related to drone activity, including regional drone activity frequency (e.g., number of drone appearances per hour) and a database of known enemy drone characteristics (e.g., model, signal characteristics, behavior patterns)). : , Among them, σ is the Sigmoid function, which ensures that the output is in [0,1], and r is the transpose operation of the vector. is the normalized environment vector The weight vector is obtained through the MLP supervised learning method. Threat intelligence features (activity frequency and feature library matching) are generated by querying intelligence databases (such as military intelligence databases) to reflect the potential threat level in the region. is the threat intelligence feature vector, which includes the regional drone activity frequency and the matching degree of the feature library. It is generated by querying the intelligence database. B is the bias term with an initial value of 0. It is obtained by MLP supervised learning through training with historical data (including known drone activities and environmental conditions); according to , adjust the sampling frequency (such as 25Hz when high threat > threshold; otherwise 15Hz), based on the multimodal data after adjusting the sampling frequency, use a high-precision GPS clock (accuracy ±10ns) to align the data and generate a data set with a unified timestamp : , in, is the radar echo (intensity, distance, speed), is the RF signal (spectral characteristics), is an infrared image, For visible light images.

[0024] By collecting real-time environmental parameters to generate normalized environmental vectors, and combining threat intelligence with MLP supervised learning to calculate the initial threat level, adaptive sampling frequency adjustment and multimodal data alignment are achieved, enhancing the environmental adaptability of data collection. Compared with the traditional fixed-weight fusion method, by dynamically adjusting the sampling frequency (25Hz for high threat and 15Hz for low threat) and high-precision GPS clock alignment data, the reliability of multimodal data in complex environments is effectively improved, the false alarm and missed alarm rates caused by environmental interference are reduced, and the robustness and recognition accuracy of the drone countermeasure system are significantly improved.

[0025] Furthermore, the short-time Fourier transform is used to extract the time-frequency spectrum of the radar echo, and the time-frequency spectrum is input into the MobileNetV3 model to extract the radar signal features. The 1D-CNN model is used to extract the features of the radio frequency signal. The EfficientNet-B0 model is used to extract the visual features of the visible light image. The IRNet is used to extract the thermal imaging features of the infrared image. Then, a feature vector set is generated to refer to the radar echo. Perform feature extraction and generate a time-frequency spectrum using short-time Fourier transform (STFT): , in, is the Hanning window, the length is based on Adjusted (High Threat 128ms, Low Threat 64ms), is the frequency variable, obtained by the Cooley-Tukey FFT method, is a complex exponential function used to convert the time domain signal Converted to the frequency domain, where ω is the angular frequency, τ is the time variable, and j is the imaginary unit, is obtained by short-time Fourier transform (STFT) calculation, is the small time increment in the integral, which is approximated in actual calculations by numerical integration methods; Input the spectrogram to the pre-trained MobileNetV3 model. The network extracts the spatial and frequency features of the radar signal through convolution operations and outputs a fixed-dimensional feature vector: , in, ∈ is the radar signal at time The feature vector represents the 96-dimensional features extracted after the STFT spectrogram is processed by MobileNetV3; For RF signals Perform feature extraction and calculate signal energy and spectral entropy: , in, is the signal energy, For RF signals The amplitude of the nth frequency component in the frequency domain is obtained from the RF signal by Fast Fourier Transform (FFT). Calculated, is the total number of frequency components, is the spectrum entropy, which is calculated by the Shannon information entropy method. is the normalized probability distribution of the spectrum, which represents the probability of frequency component q, and is obtained by normalizing the spectrum generated by fast Fourier transform (FFT); When the electromagnetic noise level N( )> (preset maximum value of electromagnetic noise) (determined by experimental analysis on RF data sets), RF signals are severely interfered with, and signal energy Reduced reliability, increased spectral entropy Weights are enhanced to represent the signal frequency distribution. A 1D-CNN model (a custom model trained from scratch on the RF dataset. The RF dataset is collected, including UAV communication and control signal samples (about 100,000, including signal energy and spectral entropy features), and the signal types are labeled (such as enemy, commercial). The model architecture is designed with three layers of convolution (64, 128, 96 neurons), using ReLU activation and max pooling, and outputs a 96-dimensional feature vector. With the cross-entropy loss as the objective, the Adam optimizer is used, and it is trained on the dataset for 50 epochs with 32 data per batch. Hyperparameters are adjusted through 5-fold cross-validation to ensure the generalization ability of the model. Finally, the classification accuracy on the validation set reaches more than 95%, generating a custom model suitable for RF feature extraction), and outputs the feature vector of the RF signal : , For visible light images feature extraction is performed. If the light intensity <u, indicating that the environment is too dark (such as at night or in extremely low light), the quality of the visible light camera image is poor, and feature extraction is unreliable. The threshold u is determined through experimental verification, and the processing is skipped to avoid invalid calculations. Otherwise, the pre-trained EfficientNet-B0 model is used to extract the high-order visual features of the visible light image ∈<00​​​​​​​​​​​​​​​​​​​By using short-time Fourier transform (STFT) combined with an adaptive Hanning window on radar echoes to generate a time-frequency spectrum and inputting it into MobileNetV3 for feature extraction, the present invention enhances feature robustness by calculating signal energy and spectral entropy of radio frequency signals and increasing the spectral entropy weight in high electromagnetic noise environments. A customized 1D-CNN model (three-layer convolution, 95% classification accuracy) is used to enhance feature robustness. Visible light images are filtered by illumination thresholds to avoid invalid calculations and high-order features are extracted using EfficientNet-B0. An optimized IRNet is used to extract thermal imaging features for infrared images. This invention achieves adaptive extraction and optimization of multimodal features, generating a unified 96-dimensional feature vector set. Compared with traditional fixed feature extraction methods, the present invention significantly improves the accuracy and reliability of feature extraction in complex environments, reduces false alarm and omission rates, and provides a more efficient and robust feature representation for drone countermeasures.

[0027] S2. Calculate the initial modal weights based on the MLP supervised learning method, calculate the attention weights through the Transformer attention mechanism, and use the collaborative attention mechanism to fuse the feature vectors based on the initial modal weights and the attention weights; Use the Transformer model to perform real-time classification and generate threat levels based on the fused feature vector; Specifically, the modal initial weight is calculated based on the MLP supervised learning method, the attention weight is calculated through the Transformer attention mechanism, and the feature vector is fused based on the modal initial weight and the attention weight using the collaborative attention mechanism. : , in, is the adaptive function of mode j, implemented by MLP (2 layers, 128 neurons), is the concatenation vector, is an exponential function, that is , (such as e≈2.718), Used to output MLP Convert to positive values ​​to ensure weight normalization, It is an adaptive function of modality k∈{R,F,V,I}, outputting a scalar value representing the initial weight contribution of modality k. MLP is trained through supervised learning (based on environment and threat datasets, cross entropy loss, Adam optimizer) to optimize the weights to reflect the importance of the modality under different environments and threats. Computing data-driven attention weights via the Transformer attention mechanism (single layer, 4-head encoder) : , Among them, W is the weight matrix in the Transformer attention mechanism, b is the bias vector, W and b are trained by the Transformer encoder (single layer, 4 heads), tanh is the hyperbolic tangent activation function, , mapping the input to [−1,1] to enhance feature expression capabilities, For the moment A single modal eigenvector of , from the eigenvector set F( ), is the weight vector in the Transformer attention mechanism, which is obtained by training the Transformer encoder using supervised learning on a multimodal dataset; Adopt collaborative attention mechanism through the initial weights of each modality , according to the environment vector and threat level Adaptive adjustment, using data-driven attention weights , based on modal features The dynamic importance of and Multiplied as a comprehensive weight, and the corresponding modal features Weighted summation to generate fused feature vector .

[0028] The initial modal weights are calculated based on MLP supervised learning combined with environmental vectors and threat levels to reflect the importance of the modality in different environments (such as lighting and noise). The Transformer attention mechanism is used to calculate data-driven attention weights to capture the dynamic importance of modal features. The initial weights are multiplied by the attention weights through the collaborative attention mechanism to generate comprehensive weights. The radar, radio frequency, visible light, and infrared features are weighted and summed to generate a 96-dimensional fused feature vector. This overcomes the limitation of traditional fixed-weight fusion methods that are insufficiently responsive to environmental changes, and significantly improves the feature fusion robustness and recognition accuracy in complex environments.

[0029] Furthermore, the Transformer model is used for real-time classification and threat level generation based on the fused feature vector. A fully connected neural network (three layers, 256-64 categories) is trained on a drone dataset, and cross entropy loss is used to input the time series fusion features into a lightweight Transformer model (two layers, four heads). The drone behavior pattern is analyzed, an intent vector is generated, and the dynamic threat level is calculated through a fully connected layer (the time series fusion feature vectors of the first five frames are obtained and input into a lightweight Transformer model (two layers, four attention heads). The model is trained on the drone behavior dataset. The Transformer processes the sequence through a multi-head self-attention mechanism, captures time dependencies and behavior patterns, such as trajectory changes or communication frequency, and generates an intent vector that represents the context of the drone behavior. The intent vector is input into the fully connected neural network layer, which applies the trained weight matrix and bias to transform the vector into a high-dimensional representation, and then the probability distribution of different threat levels (such as low, medium, and high) is calculated through the Softmax activation function. The network ensures accurate threat assessment through supervised learning optimization on the labeled threat scene, and outputs a dynamic threat level, which is a probability representation of the current threat severity of the drone. Rate distribution), reflecting the real-time threat level, using contrast loss (InfoNCE) combined with cross entropy loss to optimize the model (designing the model structure, including a feature extractor (such as a convolutional neural network) to generate high-dimensional feature representations, and a fully connected layer to output classification results. The feature extractor converts input data (such as images or text) into feature vectors, and the fully connected layer maps the feature vectors to category probabilities, ensuring that the model can simultaneously output feature representations and classification results. Preprocessing the dataset, including data enhancement (such as flipping, cropping) and normalization processing to improve model robustness. During the training process, for each batch of data, calculate The cross-entropy loss is calculated to optimize classification performance, and contrast loss is used at the same time (to enhance feature discrimination by making the features of positive sample pairs more similar and the features of negative sample pairs more different). These two losses are weighted and combined. The weights can be adjusted according to the experiment to balance classification accuracy and feature learning effect. The optimizer (such as Adam) is used to update the model parameters, and the weights and hyperparameters are adjusted through the validation set to optimize performance. The classification accuracy and feature representation quality of the model are regularly evaluated on the test set to ensure that the model converges and has good generalization ability. The ability to distinguish camouflaged targets is enhanced, and the output is category probability, dynamic threat level, and intent vector.

[0030] The fused feature vector is processed through a fully connected neural network to output the drone category probability. The lightweight Transformer model is used to analyze the fused features of the first five frames of the time series, capture the drone's trajectory changes and communication frequency and other behavioral patterns, generate the intent vector, and then calculate the dynamic threat level through the fully connected layer and the Softmax function. The cross-entropy loss and contrast loss optimization model are combined to enhance the ability to distinguish camouflaged targets and overcome the limitations of traditional methods in identifying dynamic behaviors and camouflaged targets. The robustness of the model is improved through data enhancement and normalization preprocessing, which significantly improves the accuracy and real-time performance of drone classification and threat assessment in complex environments, providing a more efficient and reliable decision-making basis for drone countermeasures.

[0031] S3. Calculate the countermeasure priority based on the threat level and real-time classification results, and generate anti-spoofing detection results and countermeasure instructions based on the countermeasure priority; Based on anti-spoofing detection results and countermeasure instructions, dynamic countermeasure instruction execution and effect feedback optimization are carried out; Specifically, the countermeasure priority is calculated based on the threat level and classification results, and the anti-spoofing detection results and countermeasure instructions are generated based on the countermeasure priority. The cosine similarity between the features of each modality is calculated, and the feature consistency is evaluated (for each pair of modal features (such as radar and radio frequency, visible light and infrared), the cosine similarity of the feature vector is calculated, the dot product operation is performed on the two feature vectors, and the product is divided by the norm to obtain a similarity score in the range of [-1,1], the similarity scores of all modal pairs are summarized, and the average consistency score is calculated). If the average consistency score is lower than the preset threshold (such as 0.75), it is marked as a potential camouflage target and One-Class is used. The SVM model (based on the radial basis kernel) analyzes the fusion feature vector, detects abnormal behavior, outputs a negative value to indicate anomaly, combines the threat level and classification results (category probability, dynamic threat level and intent vector), calculates the countermeasure priority (referring to the indicator calculated based on the classification result), the system receives input data (such as text or file), performs in-depth processing through context analysis capabilities, uses the classification algorithm to generate category probability, reflects the possibility that the input data belongs to each predefined category, combines the real-time network information provided by the agent system, evaluates the dynamic threat level to quantify the potential risk, generates the intent vector through the intent recognition mechanism, captures the potential intention of the input content, integrates the category probability, threat level and intent vector. The system calculates the countermeasure priority based on the threat level and intent vector, and determines the urgency and priority of the response measures. For example, inputs with a high threat level and malicious intent vector will generate a high countermeasure priority, prompting the system to take immediate action, such as isolating the threat source or strengthening defense measures. If the priority exceeds the threshold and there are no abnormalities, a countermeasure instruction containing the target location (provided by the radar), category, and threat level is generated and sent to the RF jammer. If camouflage is detected, the weight of the abnormal mode is reduced and the fusion feature vector step is returned to re-fusion. The output result is the anti-spoofing detection result (whether it is camouflaged) and the countermeasure instruction (target location, category, threat level, priority) to achieve precise countermeasures.

[0032] Feature consistency is evaluated by calculating the cosine similarity between the features of each modality. If the average consistency score is lower than the preset threshold, it is marked as a potential camouflaged target. This overcomes the limitations of traditional methods in insufficient recognition of camouflaged drones and static countermeasure decisions, significantly improves the accuracy of anti-spoofing detection in complex environments and the dynamic adaptability of countermeasure commands, and provides a more accurate and reliable solution for drone countermeasures.

[0033] Furthermore, based on the anti-spoofing detection results and countermeasure instructions, dynamic countermeasure instruction execution and effect feedback optimization is performed, which means using an adaptive countermeasure execution and feedback optimization method to generate an optimized countermeasure strategy. According to the target location and category in the countermeasure instruction (referring to the physical or virtual coordinates of the detected target (such as geographic location or network address), the category refers to the type of target (such as drone, malware)), the radio frequency jammer is activated to perform countermeasure operations (such as signal suppression or physical interception), and countermeasure effect data is collected in real time (monitoring target state changes through radar and infrared sensors), and compared with the category probability and threat level to evaluate the countermeasure success rate. If the countermeasure fails and camouflage is detected, the modal weight is adjusted, the fusion feature vector is regenerated and reclassified. If successful, the effect data is recorded to update the training data set of the Transformer model. The results output the optimized countermeasure strategy (updated instruction) and effect evaluation report (success rate and target state) to achieve dynamic closed-loop countermeasure optimization.

[0034] By dynamically adjusting the modal weights in multimodal feature fusion, regenerating the fused feature vector and triggering reclassification, subsequent identification and countermeasure decisions are optimized. If the countermeasure is successful, the effect data is recorded and the training data set of the Transformer model is updated to further improve the model's generalization ability and threat assessment accuracy. Compared with the traditional static countermeasure process, the closed-loop feedback mechanism significantly enhances the system's environmental adaptability and anti-deception capabilities, ensuring the efficiency and reliability of the countermeasure strategy in complex and changing scenarios.

[0035] This embodiment also provides an intelligent identification system based on drone countermeasures, including: an environmental perception and threat initialization module for collecting real-time environmental parameters, generating environmental vectors, and calculating the initial threat level in combination with threat intelligence to generate a multimodal data set; Multimodal feature extraction module, used to extract features of each modality and generate a set of feature vectors; The feature fusion and weight calculation module is used to combine the normalized environment vector and the initial threat level of the environment perception module and generate a fused feature vector using the collaborative attention mechanism; Classification and threat assessment module, which uses the Transformer model and fully connected neural network for real-time classification and threat level assessment; The anti-spoofing detection and countermeasure decision module is used to fuse the feature vectors and features of each modality of the feature fusion module to perform anti-spoofing detection and generate countermeasure instructions.

[0036] This embodiment also provides a computer device, which is suitable for the case of an intelligent identification method based on drone countermeasures, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the intelligent identification method based on drone countermeasures proposed in the above embodiment.

[0037] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0038] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent identification method and system for drone countermeasures as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0039] In summary, the present invention introduces a collaborative attention mechanism, combines MLP supervised learning with the Transformer attention mechanism, and realizes adaptive modal weight adjustment based on environmental vectors and threat intelligence. This effectively solves the problem of reduced recognition accuracy of traditional fixed weight fusion in complex environments (such as strong electromagnetic interference or low visibility). By calculating the cosine similarity between modalities and One-Class SVM anomaly detection, the recognition ability of camouflaged drones is enhanced, and the false alarm rate and missed alarm rate are significantly reduced. In view of the insufficient generalization ability of single modal feature extraction, this solution optimizes the radar, radio frequency, visible light and infrared feature extraction processes, and combines STFT, 1D-CNN, MobileNetV3, EfficientNet-B0 and IRNet models to ensure the robustness of feature extraction in diverse scenarios, significantly improving the environmental adaptability, recognition accuracy and anti-spoofing ability of the drone countermeasure system, and providing a more efficient and reliable solution for aviation safety and public safety.

[0040] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An intelligent identification method based on drone countermeasures, characterized by: include, Collect real-time environmental parameters, generate environmental vectors, combine environmental vectors with threat intelligence to calculate the initial threat level, and generate a multimodal data set based on the initial threat level; Use short-time Fourier transform to extract the radar echo time-frequency spectrum. Input the time-frequency spectrum into the MobileNetV3 model to extract radar signal features. Use the 1D-CNN model to extract radio frequency signal features. Use the EfficientNet-B0 model to extract visual features of visible light images. Use IRNet to extract thermal imaging features of infrared images. Then generate a set of feature vectors. The initial modal weights are calculated based on the MLP supervised learning method, the attention weights are calculated through the Transformer attention mechanism, and the feature vectors are fused using the collaborative attention mechanism based on the initial modal weights and the attention weights. Based on the fused feature vector, the Transformer model is used for real-time classification and threat level generation. The countermeasure priority is calculated based on the threat level and real-time classification results. Anti-spoofing detection results and countermeasure instructions are generated based on the countermeasure priority. Based on anti-spoofing detection results and countermeasure instructions, dynamic countermeasure instruction execution and effect feedback optimization are performed.

2. The intelligent identification method based on drone countermeasures according to claim 1, characterized in that: The system is deployed at a UAV countermeasure ground station, equipped with millimeter-wave radar, radio frequency receiver, high-resolution visible light camera, long-wave infrared camera, light sensor, electromagnetic interference detector, weather station, collects real-time environmental parameters, generates environmental vector, combines environmental vector with threat intelligence to calculate initial threat level, and generates multimodal data set based on the initial threat level. , and normalize it to get the environment vector , combining environmental vectors with threat intelligence to calculate the initial threat level ,according to , adjust the sampling frequency, based on the multimodal data after adjusting the sampling frequency, use the high-precision GPS clock to align the data and generate a data set with a unified timestamp .

3. The intelligent identification method based on drone countermeasures according to claim 2, characterized in that: The short-time Fourier transform is used to extract the time-frequency spectrum of the radar echo, the time-frequency spectrum is input into the MobileNetV3 model to extract the radar signal features, the 1D-CNN model is used to extract the features of the radio frequency signal, and a feature vector set is generated to refer to the radar echo. Perform feature extraction and generate a time-frequency spectrum using short-time Fourier transform (STFT). The spectrum is input to the pre-trained MobileNetV3 model. The network extracts the spatial and frequency features of the radar signal through convolution operations and outputs a fixed-dimensional feature vector. ; For RF signals Perform feature extraction and calculate signal energy and spectral entropy , when the electromagnetic noise level N( )> , the RF signal is seriously interfered with, and the signal energy Reduced reliability, increased spectral entropy Weight, enhance the representation of signal frequency distribution, use 1D-CNN model to output the feature vector of RF signal ; For visible light images feature extraction is performed. If the light intensity <u, skip the processing to avoid invalid calculations. Otherwise, use the pre-trained EfficientNet-B0 model to extract the high-order vision of visible light images , use IRNet to extract the thermal imaging features of infrared images , and generate a set of feature vectors through adaptive feature extraction .​ 4. The intelligent identification method based on drone countermeasures according to claim 3, characterized in that: The method of fusing feature vectors based on modal initial weights and attention weights using a collaborative attention mechanism refers to calculating modal initial weights based on an MLP supervised learning method. , calculate the data-driven attention weights through the Transformer attention mechanism , using the collaborative attention mechanism through the initial weights of each modality , according to the environment vector and threat level Adaptive adjustment, using data-driven attention weights , based on modal features The dynamic importance of and Multiplied as a comprehensive weight, and the corresponding modal features Weighted summation to generate fused feature vector .

5. The intelligent identification method based on drone countermeasures according to claim 4, characterized in that: The real-time classification and threat level generation based on the fused feature vector using the Transformer model refers to training a fully connected neural network with a drone dataset, using cross-entropy loss, inputting time series fusion features into a lightweight Transformer model, analyzing drone behavior patterns to generate intent vectors, and calculating dynamic threat levels through fully connected layers to reflect the real-time threat level. The model is optimized using contrast loss combined with cross-entropy loss to enhance the ability to distinguish camouflaged targets. The results output category probabilities, dynamic threat levels, and intent vectors.

6. The intelligent identification method based on drone countermeasures according to claim 5, characterized in that: The countermeasure priority is calculated by the threat level and classification result, and the anti-spoofing detection result and countermeasure instruction are generated based on the countermeasure priority. The cosine similarity between the features of each modality is calculated to evaluate the feature consistency. If the average consistency score is lower than the preset threshold, it is marked as a potential camouflage target. The One-Class SVM model is used to analyze the fused feature vector, detect abnormal behavior, and output a negative value to indicate an abnormality. The countermeasure priority is calculated based on the threat level and classification result. If the priority exceeds the threshold and there is no abnormality, a countermeasure instruction containing the target location, category and threat level is generated and sent to the radio frequency jammer. If camouflage is detected, the weight of the abnormal modality is reduced and the fusion feature vector step is returned to be re-fused. The output result is the anti-spoofing detection result and countermeasure instruction.

7. The intelligent identification method based on drone countermeasures according to claim 6, characterized in that: The anti-spoofing detection results and countermeasure instructions are used to perform dynamic countermeasure instruction execution and effect feedback optimization. An optimized countermeasure strategy is generated using an adaptive countermeasure execution and feedback optimization method. According to the target location and category in the countermeasure instruction, the radio frequency jammer is activated to perform the countermeasure operation. The countermeasure effect data is collected in real time and compared with the category probability and threat level to evaluate the countermeasure success rate. If the countermeasure fails and camouflage is detected, the modal weight is adjusted, the fused feature vector is regenerated and reclassified. If successful, the effect data is recorded to update the training data set of the Transformer model. The optimized countermeasure strategy and effect evaluation report are output as the result, realizing dynamic closed-loop countermeasure optimization.

8. An intelligent identification system based on drone countermeasures, based on the intelligent identification method based on drone countermeasures according to any one of claims 1 to 7, characterized in that: include, The environmental perception and threat initialization module is used to collect real-time environmental parameters, generate environmental vectors, and calculate the initial threat level based on threat intelligence to generate a multimodal data set; Multimodal feature extraction module, used to extract features of each modality and generate a set of feature vectors; The feature fusion and weight calculation module is used to combine the normalized environment vector and the initial threat level of the environment perception module and generate a fused feature vector using the collaborative attention mechanism; Classification and threat assessment module, which uses the Transformer model and fully connected neural network for real-time classification and threat level assessment; The anti-spoofing detection and countermeasure decision module is used to fuse the feature vectors and features of each modality of the feature fusion module to perform anti-spoofing detection and generate countermeasure instructions.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the intelligent identification method based on drone countermeasures described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent identification method based on drone countermeasures according to any one of claims 1 to 7 are implemented.

Images