A method and device for unmanned aerial vehicle signal detection based on sparse mask and denoising reconstruction pre-training, a storage medium and an electronic device

Abstract

The application provides a kind of unmanned aerial vehicle signal detection method, device, storage medium and electronic equipment based on sparse mask and denoising reconstruction pre-training, related to signal detection field, the method comprises: collecting unmanned aerial vehicle signal in real environment;Based on the unmanned aerial vehicle signal, pre-training set, fine-tuning training set and verification set are obtained;Unmanned aerial vehicle signal denoising detection network based on the architecture of Transformer encoder and decoder is constructed, and the pre-training set is used to pre-train unmanned aerial vehicle signal denoising detection network;The fine-tuning training set is used to fine-tune the unmanned aerial vehicle signal denoising detection network that completes pre-training;The verification set is used to verify and evaluate the unmanned aerial vehicle signal denoising detection network after fine-tuning, and select the best performance model as the final model;Detection is carried out using the final model.The method solves the problems of low detection accuracy, local feature loss and other problems of unmanned aerial vehicle signal in complex background interference and low signal-to-noise ratio environment.

Description

Technical Field

[0001] This application relates to the field of signal detection technology, and more specifically, to a method, apparatus, storage medium, and electronic device for UAV signal detection based on sparse masking and denoising reconstruction pre-training. Background Technology

[0002] Research on UAV signal detection originated from the military's need to monitor low-altitude targets. With the rapid development of UAV technology, its applications in military reconnaissance, environmental monitoring, and disaster relief have become increasingly widespread, making UAV signal detection one of the core technologies for ensuring airspace safety and data acquisition efficiency. Among these technologies, UAV model identification has significant practical implications. With the surge in the number of UAVs, air traffic management and safety supervision face enormous challenges. Fast and accurate model identification technology can effectively distinguish between civilian, commercial, and military UAVs, providing technical support for low-altitude airspace management.

[0003] Currently, commonly used technologies are centered on electromagnetic spectrum analysis, integrating methods such as radio frequency fingerprinting, micro-Doppler effect analysis, and multimodal sensing. Although radio frequency fingerprinting can achieve a classification accuracy of over 98%, and micro-Doppler analysis can effectively distinguish drones from interfering targets such as birds, radio frequency fingerprinting mainly relies on signal distortion characteristics caused by hardware circuits, and the generation mechanism of the micro-Doppler effect originates from the periodic modulation of electromagnetic waves by the micro-motion of the target. Neither can distinguish different drone models and does not fully explore the structural relationship of drone frequency hopping signals in the time and frequency domains as a basis for identification. Therefore, they cannot be applied to anti-drone systems that require clear identification of drone models, and are easily affected by environmental noise and hardware differences in practical applications.

[0004] Modern unmanned aerial vehicles (UAVs) generally employ wireless communication protocols for control, with their communication links primarily consisting of an uplink control link and a downlink data link. The uplink control link sends control commands from the ground remote control station to the UAV, using a frequency-hopping communication system, and the signal frames transmitted during each hop are in narrowband format. The downlink data link, on the other hand, transmits large amounts of data, such as images and videos captured by the UAV, back to the ground remote control station; this link typically uses a wideband fixed-frequency communication method.

[0005] Drone signals typically exhibit characteristics such as high sparsity, low signal-to-noise ratio, non-stationarity, and complex background interference. Traditional signal processing methods, such as Fourier transform or wavelet analysis, often have limited performance when dealing with such signals because they cannot effectively capture local features.

[0006] Current mainstream methods for UAV signal detection can be broadly categorized into two types: those based on traditional signal processing and those based on deep learning. Traditional methods, such as compressed sensing and sparse filtering, utilize the sparsity of signals to achieve noise reduction and feature extraction. However, when processing multi-scale signals in dynamic and varied environments, their performance often degrades due to poor mask ratio control or insufficient integration of contextual information. Deep learning methods, such as convolutional neural networks (CNNs) and generative adversarial networks (GANs), while demonstrating advantages in feature learning, face bottlenecks such as scarce training data and poor model interpretability. For example, while denoising autoencoders (DAEs) can improve data reconstruction efficiency in UAV edge computing, modeling nonlinear spatial correlations still relies on a large amount of labeled data.

[0007] In summary, current UAV signal detection technologies rarely use the time-frequency relationship of frequency-hopping signals as a detection basis, and face the problem of insufficient adaptability to complex electromagnetic environments. On the other hand, methods based on deep learning to extract UAV signal features often rely heavily on labeled data due to the sparsity of UAV frequency-hopping signals, and lack interpretability and generalization ability. Summary of the Invention

[0008] The embodiments of this application provide a method, apparatus, storage medium, and electronic device for UAV signal detection based on sparse masking and denoising reconstruction pre-training, in order to solve problems such as low detection accuracy and loss of local features of UAV signals in complex background interference and low signal-to-noise ratio environments.

[0009] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0010] According to a first aspect of the embodiments of this application, a method for detecting unmanned aerial vehicle (UAV) signals based on sparse masking and denoising reconstruction pre-training is provided, comprising: Collect drone signals in real-world environments, including high signal-to-noise ratio drone signals and drone signals with different signal-to-noise ratios; The UAV signals are converted into two-dimensional time-frequency spectra with bounding box annotations, and a pre-training set, a fine-tuning training set, and a validation set are constructed. A UAV signal denoising and detection network based on a Transformer encoder and decoder architecture is constructed, and the UAV signal denoising and detection network is pre-trained using the pre-training set. The pre-trained UAV signal denoising and detection network was fine-tuned using the fine-tuning training set. The fine-tuned UAV signal denoising and detection network was validated and evaluated using the validation set, and the model with the best performance was selected as the final model. The final model is used to detect drone signals.

[0011] In some embodiments of this application, based on the foregoing scheme, the step of converting the UAV signal into a two-dimensional time-frequency spectrum with bounding box annotations to construct a pre-training set, a fine-tuning training set, and a validation set includes: The UAV signal is processed using the Welch power spectrum estimation method to generate a two-dimensional time spectrum of the UAV signal; The target signal in the two-dimensional time spectrum is labeled with bounding boxes using a frequency hopping signal contour tracking algorithm and manual drawing. The labeled two-dimensional time spectrum is divided into a pre-training set, a fine-tuning training set, and a validation set. The pre-training set contains only two-dimensional time-spectrum maps of UAV signals with high signal-to-noise ratio, while the fine-tuning training set and the validation set contain two-dimensional time-spectrum maps of UAV signals with different signal-to-noise ratios, and the two-dimensional time-spectrum map data in the fine-tuning training set and the validation set do not overlap.

[0012] In some embodiments of this application, based on the aforementioned scheme, before inputting the two-dimensional time-frequency spectrum in the pre-training set into the UAV signal denoising and detection network for training, the target signals marked in the two-dimensional time-frequency spectrum are first masked.

[0013] In some embodiments of this application, based on the aforementioned scheme, the Transformer encoder of the UAV signal denoising and detection network extracts the performance of the target signal under different noise environments and the time-frequency features of the frequency hopping signal through a masked multi-head self-attention mechanism. It stabilizes the training process by introducing residual connections and layer normalization, and uses a feedforward neural network to perform nonlinear transformation on the features.

[0014] In some embodiments of this application, based on the foregoing scheme, the decoder of the human-machine signal denoising detection network is used to perform image reconstruction and classification tasks during pre-training, and the classification task distinguishes only noise and target signals.

[0015] In some embodiments of this application, based on the aforementioned scheme, in the reconstruction task, only the target signal selected by the frequency hopping signal contour tracking algorithm is randomly masked by a certain proportion. In the classification task, after the target signal is randomly masked by a certain proportion, the noise signal is randomly masked by the same number of patches.

[0016] In some embodiments of this application, based on the aforementioned scheme, in the process of verifying and evaluating the fine-tuned UAV signal denoising detection network, a standard target detection evaluation protocol is adopted, and the comprehensive performance index is calculated based on the intersection-union ratio matching and precision-recall curve.

[0017] According to a second aspect of the embodiments of this application, a UAV signal detection device based on sparse masking and denoising reconstruction pre-training is provided, comprising: The acquisition unit is used to acquire drone signals in real-world environments, including high signal-to-noise ratio drone signals and drone signals with different signal-to-noise ratios. The conversion unit is used to convert the UAV signal into a two-dimensional time-frequency spectrum with bounding box annotations, and to construct a pre-training set, a fine-tuning training set, and a validation set. The building unit is used to construct a drone signal denoising and detection network based on a Transformer encoder-decoder architecture; The training unit is used to pre-train the UAV signal denoising detection network using the pre-training set; The fine-tuning unit is used to fine-tune the pre-trained UAV signal denoising and detection network using the fine-tuning training set; The verification unit is used to verify and evaluate the fine-tuned UAV signal denoising detection network using the verification set, and select the model with the best performance as the final model. The detection unit is used to detect UAV signals using the final model.

[0018] According to a third aspect of the embodiments of this application, a computer-readable storage medium is provided, the storage medium storing computer instructions that, when executed on a computer, cause the computer to perform the method as described in the first aspect.

[0019] According to a fourth aspect of the embodiments of this application, an electronic device is provided, including: a memory and a processor; The memory is used to store computer instructions; The processor is configured to invoke computer instructions stored in the memory, causing the electronic device to execute the method described in the first aspect.

[0020] This application innovatively constructs a UAV signal detection system based on the sparse features of UAV frequency-hopping signals, overcoming the technical bottlenecks of traditional methods' inability to adapt to complex channels and their weak generalization ability. By employing data annotation technology from the Frequency-Hopping Signal Contour Tracking (FHCT) algorithm, it solves the problem of reliance on manual annotation in traditional methods, achieving self-supervised pre-training while enabling the model to specifically mask the target signal, significantly improving the model's generalization ability.

[0021] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0022] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings: Figure 1 The diagram illustrates a step-by-step flowchart of a UAV signal detection method based on sparse masking and denoising reconstruction pre-training according to an embodiment of this application.

[0023] Figure 2 The diagram illustrates a detailed flowchart of a UAV signal detection method based on sparse masking and denoising reconstruction pre-training according to an embodiment of this application. Figure 3 A flowchart of a preprocessing procedure according to an embodiment of this application is shown; Figure 4 A schematic diagram of a targeted mask according to an embodiment of this application is shown; Figure 5 A block diagram of a UAV signal detection device based on sparse masking and denoising reconstruction pre-training is shown according to an embodiment of this application; Figure 6 A block diagram of an electronic device according to one embodiment of this application is shown; Figure 7 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown. Detailed Implementation

[0024] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.

[0025] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0026] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0027] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0028] It should be noted that "multiple" in this article refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0029] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such uses of these terms can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described.

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0031] The following detailed description of some embodiments of this application will be provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0032] See Figure 1 The diagram illustrates a step flow of a UAV signal detection method based on sparse masking and denoising reconstruction pre-training according to an embodiment of this application.

[0033] See Figure 2 The diagram illustrates a detailed flowchart of a UAV signal detection method based on sparse masking and denoising reconstruction pre-training according to an embodiment of this application.

[0034] like Figure 1 , Figure 2 As shown, a UAV signal detection method based on sparse masking and denoising reconstruction pre-training is demonstrated, specifically including steps S100 to S500.

[0035] refer to Figure 1 Step S100: Collect drone signals in a real environment, including high signal-to-noise ratio drone signals and drone signals with different signal-to-noise ratios.

[0036] It should be noted that the high signal-to-noise ratio UAV signal is obtained by the acquisition device at close range with the UAV and is regarded as a clean signal in this invention, while the signals with different signal-to-noise ratios are obtained by the acquisition device at different distances from the UAV.

[0037] For example, the specific process of this step is as follows: A high-performance radio frequency transceiver (ADRV9009) is used to sample the UAV control link signal to obtain the real UAV signal. By changing the distance between the device and the UAV, various signal-to-noise ratio signals, including high signal-to-noise ratio signals, are collected to form a sample library containing high signal-to-noise ratio UAV signals and UAV signals with different signal-to-noise ratios.

[0038] Continue to refer to Figure 1 In step S200, the UAV signal is converted into a two-dimensional time-frequency spectrum with bounding box annotations, and a pre-training set, a fine-tuning training set, and a validation set are constructed.

[0039] In some feasible embodiments, based on the foregoing scheme, the step of converting the UAV signal into a two-dimensional time-frequency spectrum with bounding box annotations to construct a pre-training set, a fine-tuning training set, and a validation set includes: The UAV signal is processed using the Welch power spectrum estimation method to generate a two-dimensional time spectrum of the UAV signal; The target signal in the two-dimensional time spectrum is labeled with bounding boxes using a frequency hopping signal contour tracking algorithm and manual drawing. The labeled two-dimensional time spectrum is divided into a pre-training set, a fine-tuning training set, and a validation set. The pre-training set contains only two-dimensional time-spectrum maps of UAV signals with high signal-to-noise ratio, while the fine-tuning training set and the validation set contain two-dimensional time-spectrum maps of UAV signals with different signal-to-noise ratios, and the two-dimensional time-spectrum map data in the fine-tuning training set and the validation set do not overlap.

[0040] For example, the specific process of this step is as follows: The acquired UAV signals are preprocessed to convert them into a two-dimensional time-spectrum graph and to obtain bounding boxes for effective signal segments in the training set, facilitating targeted signal feature learning by the network. The time-spectrum graph is divided into a pre-training dataset, a fine-tuning training set, and a validation set based on signal-to-noise ratio (both the fine-tuning and validation sets use manually labeled bounding boxes). (Refer to...) Figure 3 The preprocessing process is as follows: Step S210: Calculate the time spectrum of the I and Q signals respectively using the Welch power spectrum estimation method. The Welch estimator is an improvement on the periodogram estimator. The core idea of ​​the Welch method is to cleverly utilize segmentation, windowing, overlap, and averaging operations to significantly improve the stability of the estimation.

[0041] A finite-length sequence of length N The sequence is divided into segments, each of length K, and the segments overlap. Then the number of segments L is: ; The Welch estimator expression is then: ; in, Power of the windowed function:

[0042] In this example, the FFT has 512 points, and the time dimension of the time spectrum after segmented averaging is 512.

[0043] Step S220: Use the FHCT algorithm to label the bounding box of the target signal in the time spectrum generated in step S210. .

[0044] S230, Convert the time spectrum to The RGB image is used as input for steps S300 and S400.

[0045] Continue to refer to Figure 1 Step S300: Construct a drone signal denoising and detection network based on the Transformer encoder and decoder architecture, and pre-train the drone signal denoising and detection network using the pre-training set.

[0046] In some feasible embodiments, based on the aforementioned scheme, before inputting the two-dimensional time-frequency spectrum in the pre-training set into the UAV signal denoising and detection network for training, the target signals marked in the two-dimensional time-frequency spectrum are first masked.

[0047] In some feasible embodiments, based on the aforementioned scheme, the Transformer encoder of the UAV signal denoising and detection network extracts the performance of the target signal under different noise environments and the time-frequency features of the frequency hopping signal through a masked multi-head self-attention mechanism. It stabilizes the training process by introducing residual connections and layer normalization, and uses a feedforward neural network to perform nonlinear transformation on the features.

[0048] In some feasible embodiments, based on the aforementioned scheme, the decoder of the human-machine signal denoising detection network is used to perform image reconstruction and classification tasks during pre-training, and the classification task distinguishes only noise and target signals.

[0049] In some feasible embodiments, based on the aforementioned scheme, in the reconstruction task, only the target signal selected by the frequency hopping signal contour tracking algorithm is randomly masked by a certain proportion. In the classification task, after the target signal is randomly masked by a certain proportion, the noise signal is randomly masked by the same number of patches.

[0050] For example, the training process is as follows: The UAV signal denoising and detection network adopts a Transformer encoder-decoder architecture. The encoder extracts the features of the target signal under different noise environments and the global time-frequency features. The two decoders only complete the target signal judgment and time-spectrum reconstruction tasks during the pre-training phase. Before the image is input to the encoder, the model's input preprocessing module first performs targeted masking on the target signal annotated in the time-spectrum, referencing... Figure 4 In the reconstruction task, 75% of the target signal contained in the FHCT bounding box is first randomly selected for masking, and then 75% of the remaining patches are randomly partially masked. ； In the classification task, the target signal is randomly masked by 75%, and the noise signal is randomly masked by the same number of patches.

[0051] This example encoder operates on only a small fraction (25%) of the complete image. Masked patches are removed; no masking is used. This allows this method to train very large encoders using only a small fraction of the computation and memory. The encoder in this example contains the following key components: 1. Image Patch Embedding Layer: The image is segmented into multiple non-overlapping patches, and each patch is converted into a vector representation.

[0052] 2. Position encoding: Add position information to image patches, because Transformer itself does not have the ability to process spatial position.

[0053] 3. Feature Extraction Module: The encoder performs deep feature extraction through 12 Transformer blocks. Each block contains a multi-head self-attention mechanism to capture global dependencies between image patches, an MLP for non-linear feature transformation, and residual connections and layer normalization for stable training.

[0054] 4. Output layer normalization: The encoder output is normalized to stabilize the training process and accelerate convergence.

[0055] The decoder takes as input a complete set of tokens, including the encoded visible patch and the learnable mask token. This example decoder contains the following key components: 1. Feature projection layer: Transforms the feature dimensions output by the encoder to the decoder dimensions.

[0056] 2. Feature Processing Module: The decoder performs deep feature processing through 8 Transformer blocks. Each block also includes a multi-head self-attention mechanism, a feedforward neural network (MLP), residual connections, and layer normalization.

[0057] 1) The multi-head self-attention mechanism in the decoder is used to capture long-distance dependencies between image patches and integrate information from different locations using unmasked image patches to reconstruct masked areas; 2) Feedforward neural networks (MLPs) are used to perform non-linear transformations on the features extracted by the attention mechanism, helping to reconstruct the detailed information of occluded areas and enhancing the model's ability to understand the local structure of the image. In this example, the MLP uses GELU as its activation function, which provides a smooth non-linear transformation and outperforms ReLU in the Transformer class.

[0058] 3) Layer normalization in the decoder is used to standardize the input features so that the mean is 0 and the variance is 1, which prevents gradient explosion or vanishing and accelerates model convergence.

[0059] 4) Residual connections are used to preserve the original feature information and prevent features from being lost during deep propagation.

[0060] 3. Output generation module: Uses a linear layer to output the reconstructed image (reconstruction task) or the determination result of the target signal (classification task).

[0061] The classification task uses the binary cross-entropy loss function between the predicted label and the true label to quantify the prediction error. Its mathematical expression is: ; in These are real labels, with values ​​of 0 or 1. This is the model's raw output (logits), and its value can be any real number.

[0062] The reconstruction loss is quantized using the mean square error function between the reconstructed time-frequency spectrum and the time-frequency spectrum of the clean signal (loss is calculated only for the masked patch): ; During model training, the total loss function is a weighted sum of the classification loss and the reconstruction loss. This combination helps to simultaneously optimize the model's classification ability and signal denoising ability. .

[0063] Continue to refer to Figure 1 Step S400: Fine-tune the pre-trained UAV signal denoising detection network using the fine-tuning training set.

[0064] For example, the fine-tuning process is as follows: The fine-tuned training set generated in step S200 is input into the pre-trained UAV signal denoising and detection model to fine-tune the model. The encoder from the pre-training stage is removed, and ViT is used as the feature extractor, replacing the traditional CNN (such as ResNet) backbone. Simultaneously, feature pyramid extraction is combined to fuse multi-scale features, and then the features extracted from the backbone network are fed into Faster R-CNN. The loss value between the detection boxes and labels output by Faster R-CNN and the ground truth boxes and labels is calculated. By backpropagating the loss value, the model parameters are updated and optimized.

[0065] Continue to refer to Figure 1 In step S500, the fine-tuned UAV signal denoising detection network is validated and evaluated using the validation set, and the model with the best performance is selected as the final model.

[0066] In some feasible embodiments, based on the aforementioned scheme, during the verification and evaluation of the fine-tuned UAV signal denoising detection network, a standard target detection evaluation protocol is adopted, and the comprehensive performance index is calculated based on the intersection-union ratio matching and precision-recall curve.

[0067] For example, comprehensive performance metrics could be map (average precision across all categories with multiple thresholds), mar_small (average recall for small-sized targets in the dataset), mar_medium, mar_large, etc.

[0068] Continue to refer to Figure 1 Step S600: Use the final model to detect the UAV signal.

[0069] In summary, this invention provides a UAV signal detection method based on sparse masking and denoising reconstruction pre-training. By using the FHCT algorithm to label the boundary of the target signal, targeted masking of the target signal is achieved, improving the accuracy and generalization of the target detection method based on the Transformer framework. Furthermore, this method does not rely on manual annotation, and with its lightweight Transformer model design, it can realize a pure Transformer end-to-end target detection framework. Self-supervised pre-training also significantly improves the model's generalization ability.

[0070] The following describes an embodiment of the apparatus described in this application, which can be used to execute a UAV signal detection method based on sparse masking and denoising reconstruction pre-training as described in the above embodiments of this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in the above applications.

[0071] Reference Figure 5 As shown, a UAV signal detection device 500 based on sparse masking and denoising reconstruction pre-training according to an embodiment of this application includes: The acquisition unit 501 is used to acquire drone signals in real-world environments, including high signal-to-noise ratio drone signals and drone signals with different signal-to-noise ratios. The conversion unit 502 is used to convert the UAV signal into a two-dimensional time-frequency spectrum with bounding box annotations, and to construct a pre-training set, a fine-tuning training set, and a validation set. Building unit 503 is used to build a drone signal denoising and detection network based on a Transformer encoder and decoder architecture; Training unit 504 is used to pre-train the UAV signal denoising detection network using the pre-training set; The fine-tuning unit 505 is used to fine-tune the pre-trained UAV signal denoising and detection network using the fine-tuning training set. The verification unit 506 is used to verify and evaluate the fine-tuned UAV signal denoising detection network using the verification set, and select the model with the best performance as the final model. The detection unit 507 is used to detect UAV signals using the final model.

[0072] like Figure 6 As shown, this application embodiment also provides an electronic device 600, including a memory 610, a processor 620, and a computer program 611 stored in the memory 610 and executable on the processor. When the processor 620 executes the computer program 611, it implements the steps of the above-mentioned UAV signal detection method based on sparse mask and denoising reconstruction pre-training.

[0073] Since the electronic device described in this embodiment is the device used to implement the UAV signal detection device based on sparse masking and denoising reconstruction pre-training in this application embodiment, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this embodiment based on the method described in this application embodiment. Therefore, how the electronic device implements the method in this application embodiment will not be described in detail here. Any device used by those skilled in the art to implement the method in this application embodiment is within the scope of protection of this application.

[0074] In practice, when the computer program 611 is executed by the processor, it can implement any of the embodiments corresponding to the first aspect.

[0075] Figure 7 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown.

[0076] It should be noted that, Figure 7 The computer system 700 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0077] like Figure 7 As shown, the computer system 700 includes a Central Processing Unit (CPU) 701, which can perform various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) 702 or programs loaded from storage portion 708 into Random Access Memory (RAM) 703, such as performing the methods described in the above embodiments. The RAM 703 also stores various programs and data required for system operation. The CPU 701, ROM 702, and RAM 703 are interconnected via a bus 704. An Input / Output (I / O) interface 705 is also connected to the bus 704.

[0078] The following components are connected to I / O interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to I / O interface 705 as needed. A removable medium 711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 710 as needed so that computer programs read from it can be installed into storage section 708 as needed.

[0079] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 709, and / or installed from removable medium 711. When the computer program is executed by central processing unit (CPU) 701, it performs various functions defined in the system of this application.

[0080] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such transmitted data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0081] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0082] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0083] In another aspect, this application also provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the UAV signal detection method based on sparse masking and denoising reconstruction pre-training described in the above embodiments.

[0084] In another aspect, this application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to implement the UAV signal detection method based on sparse masking and denoising reconstruction pre-training as described in the above embodiments.

[0085] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0086] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the methods according to the embodiments of this application.

[0087] Other embodiments of this application will readily conceive of by those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. It should be understood that this application is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A UAV signal detection method based on sparse masking and denoising reconstruction pre-training, characterized in that, include: Collect drone signals in real-world environments, including high signal-to-noise ratio drone signals and drone signals with different signal-to-noise ratios; The UAV signals are converted into two-dimensional time-frequency spectra with bounding box annotations, and a pre-training set, a fine-tuning training set, and a validation set are constructed. A UAV signal denoising and detection network based on a Transformer encoder and decoder architecture is constructed, and the UAV signal denoising and detection network is pre-trained using the pre-training set. The pre-trained UAV signal denoising and detection network was fine-tuned using the fine-tuning training set. The fine-tuned UAV signal denoising and detection network was validated and evaluated using the validation set, and the model with the best performance was selected as the final model. The final model is used to detect drone signals; The process of converting the UAV signal into a two-dimensional time-frequency spectrum with bounding box annotations, and constructing a pre-training set, a fine-tuned training set, and a validation set, includes: The UAV signal is processed using the Welch power spectrum estimation method to generate a two-dimensional time-frequency spectrum of the UAV signal; The target signal in the two-dimensional time-frequency spectrum is labeled with bounding boxes using a frequency-hopping signal contour tracking algorithm and manual drawing. The labeled two-dimensional time-frequency spectrum is divided into a pre-training set, a fine-tuning training set, and a validation set. The pre-training set contains only two-dimensional time-frequency spectra of UAV signals with high signal-to-noise ratio, while the fine-tuning training set and the validation set contain two-dimensional time-frequency spectra of UAV signals with different signal-to-noise ratios, and the two-dimensional time-frequency spectra data in the fine-tuning training set and the validation set do not overlap. Before inputting the two-dimensional time-frequency spectrum from the pre-training set into the UAV signal denoising and detection network for training, the target signals marked in the two-dimensional time-frequency spectrum are first masked. The decoder of the UAV signal denoising detection network is used to perform image reconstruction and classification tasks during pre-training. The classification task distinguishes only noise and target signals. In the reconstruction task, only the target signal selected by the frequency hopping signal contour tracking algorithm is randomly masked by a certain proportion. In the classification task, after the target signal is randomly masked by a certain proportion, the noise signal is randomly masked by the same number of patches. The frequency hopping signal contour tracking algorithm includes: Initial detection and parameter extraction: Traverse the spectrum in chronological order and output the peak power of the signal during this time period and the approximate starting point of the frequency of the target signal segment; Time-frequency correlation and segment fusion: splicing and fusing similar signals from adjacent time periods; Multi-parameter matching and trajectory management: For the signal trajectory set fused in the previous step, a matching threshold based on bandwidth, frequency boundary and power consistency is set, and signal segment fusion and update operations are performed according to the matching results.

2. The method according to claim 1, characterized in that, The Transformer encoder of the UAV signal denoising and detection network extracts the performance of the target signal under different noise environments and the time-frequency features of the frequency hopping signal through a masked multi-head self-attention mechanism. It stabilizes the training process by introducing residual connections and layer normalization, and uses a feedforward neural network to perform nonlinear transformation on the features.

3. The method according to claim 1, characterized in that, In the process of verifying and evaluating the fine-tuned UAV signal denoising detection network, a standard target detection evaluation protocol was adopted, and the comprehensive performance index was calculated based on cross-union ratio matching and precision-recall curves.

4. A UAV signal detection device based on sparse masking and denoising reconstruction pre-training, applied to the method as described in any one of claims 1-3, characterized in that, include: The acquisition unit is used to acquire drone signals in real-world environments, including high signal-to-noise ratio drone signals and drone signals with different signal-to-noise ratios. The conversion unit is used to convert the UAV signal into a two-dimensional time-frequency spectrum with bounding box annotations, and to construct a pre-training set, a fine-tuning training set, and a validation set. The building unit is used to construct a drone signal denoising and detection network based on a Transformer encoder-decoder architecture; The training unit is used to pre-train the UAV signal denoising detection network using the pre-training set; The fine-tuning unit is used to fine-tune the pre-trained UAV signal denoising and detection network using the fine-tuning training set; The verification unit is used to verify and evaluate the fine-tuned UAV signal denoising detection network using the verification set, and select the model with the best performance as the final model. The detection unit is used to detect UAV signals using the final model.

5. A computer-readable storage medium, characterized in that, The storage medium stores computer instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1-3.

6. An electronic device, characterized in that, include: Memory and processor; The memory is used to store computer instructions; The processor is configured to invoke computer instructions stored in the memory, causing the electronic device to perform the method as described in any one of claims 1-3.

Images