Small sample unmanned aerial vehicle identification method based on communication perception integrated signal

Abstract

The invention relates to a small sample unmanned aerial vehicle identification method based on a communication perception integrated signal. The method comprises the following steps: firstly, carrying out orthogonal time-frequency air conditioning on an original transmitting signal to generate a transmitting signal sequence and transmitting the transmitting signal sequence; then, receiving a reflected signal of a target, and performing orthogonal time-frequency-space demodulation and filtering processing on the reflected signal to obtain a target spectrum feature image; then, carrying out image preprocessing on the target spectrum feature image, and extracting micro-Doppler spectrum features of the unmanned aerial vehicle and different background clutters; and finally, constructing a self-adaptive classifier, inputting the micro-Doppler spectrum features into the self-adaptive classifier for training, inputting the micro-Doppler spectrum features to be recognized into the trained self-adaptive classifier, and recognizing the unmanned aerial vehicle in the scene. Only small sample data is needed to summarize image features and divide image categories, so that the unmanned aerial vehicle is identified and the background of the unmanned aerial vehicle is detected, and the time cost required by the method is reduced compared with that of a current mainstream unmanned aerial vehicle identification method.

Description

Technical Field

[0001] The present application relates to the technical field of UAV communication perception integration and target recognition, and in particular to a small sample UAV recognition method based on communication perception integration signals. Background Art

[0002] Driven by the global wave of digitalization and intelligentization, the low-altitude economy is booming as an emerging industry. Drones, with their flexible, maneuverable, and easy-to-deploy features, have become deeply integrated into all areas of social and economic life. In aerial photography, drones offer unique perspectives to aid film and television production and geographic mapping. In security scenarios, drones provide dynamic protection of key areas through real-time monitoring. In the logistics and transportation industry, drones transcend geographical limitations, effectively solving the problem of material distribution in remote areas. As drone applications continue to expand, the need for accurate identification and effective control is becoming increasingly urgent. Drone identification technology has become a key core technology for ensuring safety and order in the low-altitude sector.

[0003] Current mainstream drone recognition technologies fall into two main categories: traditional image algorithms and deep learning algorithms. Recognition solutions based on traditional image algorithms achieve drone target detection through feature extraction, matching, and other techniques. However, in complex and ever-changing urban environments, interference factors such as pedestrian activity, swaying vegetation, and shifting light and shadows can easily lead to algorithm misjudgments or missed detections, making their robustness insufficient for practical applications. While deep learning algorithms demonstrate strong performance in image recognition, they rely heavily on large amounts of annotated data for model training. This makes the model's generalization capabilities particularly problematic when faced with new drone models or targets with special camouflage coatings, making it difficult to quickly adapt to the recognition needs of new scenarios and targets.

[0004] Therefore, the existing drone identification technology has obvious technical bottlenecks in terms of adaptability to complex environments and data dependence. Among the related technologies, there is an urgent need for a method that can accurately identify drone targets in complex urban environments under small sample data conditions. Summary of the Invention

[0005] Based on this, it is necessary to address the above technical problems and provide a small-sample UAV identification method based on communication-perception integrated signals that can accurately identify UAV targets in complex urban environments under small sample data conditions.

[0006] In a first aspect, the present application provides a small sample drone identification method based on integrated communication and perception signals. The method comprises:

[0007] Perform orthogonal time-frequency-space modulation on the original transmission signal to generate a transmission signal sequence and transmit it;

[0008] receiving a reflected signal from a target, performing orthogonal time-frequency-space demodulation and filtering on the reflected signal to obtain a target spectrum feature image;

[0009] Performing image preprocessing on the target spectrum feature image to extract micro-Doppler spectrum features of the UAV and different background clutter;

[0010] An adaptive classifier is constructed, the micro-Doppler spectrum features are input into the adaptive classifier for training, and the micro-Doppler spectrum features to be identified are input into the trained adaptive classifier to identify the drone in the scene.

[0011] Optionally, in one embodiment of the present application, performing orthogonal time-frequency-space modulation on the original transmission signal to generate a transmission signal sequence and transmitting the sequence includes:

[0012] Map the input bit stream modulation into a delay-Doppler domain matrix;

[0013] Performing an inverse symplectic finite Fourier transform on the delay-Doppler domain matrix to map it into a time-frequency domain matrix;

[0014] Performing Heisenberg transform and windowing processing on the time-frequency domain matrix to obtain a time-domain transmission signal matrix.

[0015] Optionally, in one embodiment of the present application, performing orthogonal time-frequency-space modulation on the original transmission signal to generate a transmission signal sequence and transmitting the sequence further includes:

[0016] A cyclic prefix is inserted into the time domain transmit signal matrix to obtain a transmit signal sequence.

[0017] Optionally, in one embodiment of the present application, receiving a reflection signal of a target, performing orthogonal time-frequency-space demodulation and filtering on the reflection signal to obtain a target spectrum feature image includes:

[0018] Perform serial-to-parallel conversion, windowing, and Wigner transform on the reflected signal to obtain a time-frequency domain symbol matrix;

[0019] Performing a symplectic finite Fourier transform on the time-frequency domain symbol matrix to obtain a delay-Doppler domain received signal matrix;

[0020] Matched filtering and range filtering are performed on the delay-Doppler domain received signal matrix, and short-time Fourier transform is performed to obtain a target spectrum feature image.

[0021] Optionally, in one embodiment of the present application, performing matched filtering and range filtering on the delay-Doppler domain received signal matrix, and performing short-time Fourier transform to obtain a target spectrum feature image includes:

[0022] Performing matched filtering on the delay-Doppler domain received signal matrix, counting the peak values of the matched filtering results, and obtaining the delay taps corresponding to the peak values;

[0023] Perform distance filtering on the elements in the time delay tap to eliminate communication symbol interference.

[0024] Optionally, in one embodiment of the present application, the adaptive classifier includes an input layer, an encoder layer, a feature mapping layer, a similarity calculation layer and a contrast loss optimization layer, wherein the encoder layer includes a visual encoder and a text encoder.

[0025] Optionally, in one embodiment of the present application, inputting the micro-Doppler spectrum features to be identified into a trained adaptive classifier to identify the drone in the scene includes:

[0026] Visual features are obtained based on the visual encoder, and a similarity matrix is obtained based on the similarity calculation layer;

[0027] Calculating a raw classification score based on the visual features and the similarity matrix;

[0028] Calculating a feature vector mean and a covariance matrix based on the visual features, and calculating a Gaussian discriminant classification score;

[0029] Determine key-value pairs based on one-hot encoding of visual features and labels, and calculate cache classification scores;

[0030] A target classification score is determined based on the original classification score, the Gaussian discriminant classification score, and the cache classification score, and a drone detection result is determined based on the target classification score.

[0031] In a second aspect, the present application also provides a small sample drone identification device based on integrated communication and perception signals. The device includes:

[0032] The transmission signal processing module is used to perform orthogonal time-frequency-space modulation on the original transmission signal, generate a transmission signal sequence and transmit it;

[0033] A reflection signal processing module is used to receive the reflection signal of the target, perform orthogonal time-frequency-space demodulation and filtering on the reflection signal, and obtain a target spectrum feature image;

[0034] An image preprocessing module is used to perform image preprocessing on the target spectrum feature image to extract micro-Doppler spectrum features of the UAV and different background clutter;

[0035] The drone identification module is used to construct an adaptive classifier, input the micro-Doppler spectrum features into the adaptive classifier for training, input the micro-Doppler spectrum features to be identified into the trained adaptive classifier, and identify drones in the scene.

[0036] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program and the processor executes the steps of the method described in each of the above embodiments.

[0037] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in each of the above embodiments.

[0038] The above-mentioned small-sample drone identification method based on integrated communication and perception signals first performs orthogonal time-frequency-space modulation on the original transmitted signal to generate a transmitted signal sequence and transmit it. Next, the target's reflected signal is received and orthogonal time-frequency-space demodulation and filtering are performed on the reflected signal to obtain a target spectrum feature image. Next, the target spectrum feature image is subjected to image preprocessing to extract the micro-Doppler spectrum features of the drone and different background clutter. Finally, an adaptive classifier is constructed, the micro-Doppler spectrum features are input into the adaptive classifier for training, and the micro-Doppler spectrum features to be identified are input into the trained adaptive classifier to identify drones in the scene. In other words, only a small sample of data is required to summarize image features and classify images, thereby achieving drone identification and detecting the background in which the drone is located. This method reduces the time cost required by current mainstream drone identification methods and can effectively identify drones in various backgrounds. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a diagram of the application environment of a small sample drone identification method based on communication-perception integrated signals in one embodiment;

[0040] Figure 2 1 is a flow chart of a method for identifying small-sample drones based on integrated communication and perception signals in one embodiment;

[0041] Figure 3 1. A schematic diagram of the process of OTFS signal processing in one embodiment;

[0042] Figure 4 This is a structural block diagram of a small sample drone identification device based on communication and perception integrated signals in one embodiment;

[0043] Figure 5 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0045] The small sample drone identification method based on communication perception integrated signal provided by the embodiment of the present application can be applied to Figure 1 In the application environment shown, the terminal communicates with the server through the network. The data storage system can store data that the server needs to process. The data storage system can be integrated on the server or placed on the cloud or other network servers. The terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can be smart speakers, smart TVs, smart air conditioners, smart car devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server can be implemented as a standalone server or a server cluster consisting of multiple servers.

[0046] In one embodiment, Figure 2 As shown in the figure, a small sample drone identification method based on communication perception integrated signal is provided. Figure 1 The following steps are used as an example to illustrate the server in the example:

[0047] S201: Perform orthogonal time-frequency-space modulation on the original transmission signal to generate a transmission signal sequence and transmit it.

[0048] In the embodiment of the present application, the traditional Orthogonal Frequency Division Multiplexing (OFDM) system is difficult to adapt to the high-speed detection requirements of 6G drones due to Doppler shift sensitivity and multipath interference. At the detection radar signal transmitter, the original transmit signal is subjected to orthogonal time-frequency space modulation (OTFS). Through delay-Doppler domain signal mapping, the Doppler expansion caused by the drone is evenly dispersed to the signal block, and a transmit signal sequence is generated and transmitted, which can reduce the frequency offset bit error rate.

[0049] Specifically, in one embodiment of the present application, performing orthogonal time-frequency-space modulation on the original transmission signal to generate a transmission signal sequence and transmitting the sequence includes:

[0050] S301: Modulate the input bit stream into a delay-Doppler domain matrix.

[0051] S303: Perform an inverse symplectic finite Fourier transform on the delay-Doppler domain matrix to map it into a time-frequency domain matrix.

[0052] S305: Perform Heisenberg transform and windowing processing on the time-frequency domain matrix to obtain a time-domain transmission signal matrix.

[0053] In one embodiment of the present application, Figure 3 As shown, the input bit stream is first mapped into a symbol matrix X in the delay-Doppler domain (DD) by quadrature amplitude modulation (QAM) or quadrature phase shift keying (QPSK) modulation. DD , then it is converted to the time-frequency domain (TF) through the inverse symplectic finite Fourier transform (ISFFT), and then the time-domain transmission signal matrix is generated through Heisenberg transform and windowing.

[0054] Specifically, at the transmitter, the modulation alphabet of size Q is The NM information symbols are placed in the delay-Doppler domain matrix The elements in the matrix are represented as x(k, l). The above process is to digitally modulate the serial bit data stream in the channel encoder to obtain the corresponding data symbols x[k, l]. Currently, the commonly used digital signal modulation methods are orthogonal amplitude modulation and orthogonal phase shift keying.

[0055] Afterwards, X DD Perform inverse symplectic fast Fourier transform (ISFFT) mapping to the time-frequency domain and give the following formula:

[0056]

[0057] in, is the transformation matrix of Discrete Fourier Transform (DFT), and the specific expression is given by the following formula:

[0058]

[0059] Among them, ω M =e -j2π / M ,ω N =e -j2π / N Given a vector a of length N, its DFT is represented as Inverse Discrete Fourier Transform (IDFT) is expressed as

[0060] Afterwards, the time-frequency domain matrix After Heisenberg transform and windowing processing, the time domain transmission signal matrix can be obtained as follows:

[0061]

[0062] in, is an M×M unit matrix, G tx is the transmit signal window function matrix and It is worth noting that when G tx When it is a rectangular wave, G tx =I M .

[0063] Finally, the time domain transmit signal matrix S is converted from parallel to serial to obtain the transmit signal sequence s, as described in the following formula:

[0064]

[0065] Among them, vec(·) represents vectorization, Represents the Kronecker product (KP).

[0066] In this embodiment, the ISFFT and Heisenberg transform are equivalent only for the transmission symbol matrix X in the delay-Doppler domain. DD The modulation complexity is o(MNlogN) when N-point inverse fast Fourier transform (IFFT) is performed on the row vectors of the matrix. For orthogonal frequency division multiplexing (OFDM), which also uses multi-carrier modulation, OFDM performs an M-point IFFT on the column vectors of the transmit symbol matrix in the time-frequency domain. The modulation complexity is o(MNlogM). Because the number of subcarriers, M, is typically greater than the number of symbols, N, the modulation complexity of the simplified OTFS system is lower than that of the OFDM system.

[0067] In one embodiment of the present application, performing orthogonal time-frequency-space modulation on the original transmission signal to generate a transmission signal sequence and transmitting the sequence further includes:

[0068] A cyclic prefix is inserted into the time domain transmit signal matrix to obtain a transmit signal sequence.

[0069] In one embodiment of the present application, in order to cope with the inter-symbol interference caused by the multipath effect in the time-varying channel, a cyclic prefix needs to be added before the signal is transmitted in the OTFS communication system. The length of the transmitted signal sequence after the addition becomes NM+NCP , where N CP is the cyclic prefix length.

[0070] S203: Receive the reflected signal of the target, perform orthogonal time-frequency-space demodulation and filtering on the reflected signal, and obtain a target spectrum feature image.

[0071] In the embodiment of this application, Figure 3 As shown in the figure, the receiving radar receives the reflected signal from the target, performs orthogonal time-frequency-space demodulation on the reflected signal, including Wigner transform and Symplectic Finite Fourier Transform (SFFT), converts it back to the DD domain, and then performs filtering to extract the micro-Doppler spectrum characteristics of the target and draw a target spectrum characteristic image.

[0072] Specifically, in one embodiment of the present application, receiving a reflected signal of a target, performing orthogonal time-frequency-space demodulation and filtering on the reflected signal to obtain a target spectrum feature image includes:

[0073] S401: Perform serial-to-parallel conversion, windowing processing, and Wigner transform on the reflected signal to obtain a time-frequency domain symbol matrix.

[0074] S403: Performing a symplectic finite Fourier transform on the time-frequency domain symbol matrix to obtain a delay-Doppler domain received signal matrix.

[0075] S405: performing matched filtering and range filtering on the delay-Doppler domain received signal matrix, and performing short-time Fourier transform to obtain a target spectrum feature image.

[0076] In one embodiment of the present application, the receiving end obtains a received signal r after channel transmission. Specifically, the matrix representation of the input-output relationship after channel transmission is as follows:

[0077] r=Hs+w

[0078] in, and The (k+Nl) element of is equal to y[k,l],x[k,l],w[k,l], where w is the channel complex Gaussian noise vector, r is the received signal sequence, and H is the channel response matrix. The formula of H is as follows:

[0079]

[0080] Among them, h i is the channel coefficient under the i-th path, π is the permutation matrix (forward cyclic shift), is a diagonal matrix, π and The matrix expression of is as follows:

[0081]

[0082] Perform serial-to-parallel conversion on r to obtain the received signal matrix And give the following formula:

[0083] R=vec -1 (r)

[0084] Among them, vec -1 (·) represents matrixization. Apply window function to the matrix and perform Wigner transform to obtain the received time-frequency domain symbol matrix Y FT , Y FT The formula is as follows:

[0085] Y FT =F M G rx R

[0086] in, is the receiving matched filter window function matrix.

[0087] Afterwards, Y FT Perform SFFT transformation to obtain the received signal matrix in the delay-Doppler domain The elements in the matrix are represented as y(k,l), which is given by the following formula:

[0088]

[0089] The processed received signal is then subjected to matched filtering and range filtering, followed by a short-time Fourier transform (STFT) to obtain the target micro-motion characteristic spectrum. Matched filtering is used to extract the channel response peak to estimate the target delay, while range filtering is used to eliminate communication symbol interference.

[0090] Specifically, in one embodiment of the present application, performing matched filtering and range filtering on the delay-Doppler domain received signal matrix, and performing short-time Fourier transform to obtain a target spectrum feature image includes:

[0091] S501: performing matched filtering on the delay-Doppler domain received signal matrix, counting the peak values of the matched filtering results, and obtaining delay taps corresponding to the peak values.

[0092] S503: Perform distance filtering on the elements in the delay tap to eliminate communication symbol interference.

[0093] In one embodiment of the present application, the received signal Y DD Perform matched filtering and give the following formula:

[0094]

[0095] where the (k+Nl) element of the h matrix is h[k,l], The (i,j) element of the matrix is given by:

[0096]

[0097] Specifically, The first column of , ie, k″=l″=0, has the value x, and the remaining columns are cyclic shifts of x with a specific phase shift.

[0098] DD domain channel response estimated by matched filtering It is expressed as the following formula:

[0099]

[0100] in, is the conjugate transposed matrix of matrix X, and the matched filtering result is obtained by adding a threshold If there are j peaks after matched filtering, the delay taps corresponding to each peak form a single vector, denoted as l = [l0,l1,...,l j-1 ], if the drone exists, the delay tap corresponding to the drone is located in l.

[0101] Since the integrated base station transmits an OTFS communication and perception integrated signal, the communication symbols will interfere with the perception performance. Therefore, before further signal processing, it is necessary to first eliminate the interference of the transmitted communication data symbol x[k,l] in order to analyze other signal characteristics. The element-wise division operation of the received DD domain symbol y[k,l] and the transmitted symbol x[k,l] is performed, and the following formula is given:

[0102]

[0103] along Perform N-point IFFT operation on the Doppler axis to obtain the time-delay domain signal And give the following formula:

[0104]

[0105] Each different l q Represents the delay corresponding to the delay tap The distance corresponding to this delay is Different l q Corresponding to different distances R q Discrete channel response in the time delay domain.

[0106] After that, the distance filtering operation is performed on the elements in the delay tap vector l obtained by matched filtering. Each frame received by the receiver contains N f subframes, and the l corresponding to each subframe j ∈l distance filter vector is reconstructed into a new vector And give the following formula:

[0107]

[0108] in, Indicates that the i-th subframe is at l=l j The distance filtering result under the value, (·) T Represents the transpose operation, n'=0,1,...,NN f -1.

[0109] Afterwards, Perform short-time Fourier transform (STFT) to obtain its micro-motion characteristic spectrum and give the following formula:

[0110]

[0111] in, The matrix is represented as S STFT , where g is the discrete frequency coefficient, g = 0,...,G-1, w(d) is the window function, and the length of the window is D.

[0112] S205: performing image preprocessing on the target spectrum feature image to extract micro-Doppler spectrum features of the UAV and different background clutters.

[0113] In the embodiment of the present application, the experimental images under different backgrounds are read N times. And convert to RGB format image Specifically, the image Resize to n*n. If the image is not square after resizing, crop the central square part to ensure that the image size is n*n and the pixel values of the image are normalized to the range [0.0,1.0]. Represents N different categories of <image, text> pairs in a training batch group, where represents the i-th image, and Ti represents the i-th text (category label).

[0114] S207: Construct an adaptive classifier, input the micro-Doppler spectrum features into the adaptive classifier for training, input the micro-Doppler spectrum features to be identified into the trained adaptive classifier, and identify the drone in the scene.

[0115] In an embodiment of the present application, an adaptive classifier model is constructed and the micro-Doppler spectrum features are used to train the adaptive classifier so that it can accurately calculate the image classification score and select the category with the highest score as the predicted label. After training, the micro-Doppler spectrum features to be identified are input, and the model outputs whether a drone is present in the scene based on the predicted label, and the results are displayed in a visual manner.

[0116] Specifically, in one embodiment of the present application, the adaptive classifier includes an input layer, an encoder layer, a feature mapping layer, a similarity calculation layer and a contrast loss optimization layer, wherein the encoder layer includes a visual encoder and a text encoder.

[0117] In one embodiment of the present application, the adaptive classifier includes an input layer, an encoder layer, a feature mapping layer, a similarity calculation layer, and a contrast loss optimization layer, wherein the input layer receives input in two modalities, image and text, the encoder layer includes a visual encoder and a text encoder, the visual Transformer (ViT) encodes the image, and the Transformer is used to encode the text to extract their respective feature representations, the feature mapping layer maps these features into a common feature space, and calculates the similarity between the image and text features. The contrast loss optimization layer optimizes by contrastive learning loss function to maximize the similarity between matching image-text pairs while minimizing the similarity between unmatched pairs.

[0118] Specifically, all images and texts are passed through the image and text encoders respectively, and then normalized to obtain the image matrix I and text matrix T, and the following formula is given:

[0119]

[0120] Among them, φ I (·) and φ T (·) represents the image encoder (resnet) and text encoder (transformer), respectively. The image encoder and text encoder are used to process the <image, text> pair, map the image and text to a common subspace, and calculate the image-text similarity matrix. And give the following formula:

[0121] S=IT T ·e t

[0122] Among them, e t is the learnable temperature coefficient.

[0123] In one embodiment of the present application, inputting the micro-Doppler spectrum features to be identified into the trained adaptive classifier to identify the drone in the scene includes:

[0124] S601: Obtain visual features based on the visual encoder, and obtain a similarity matrix based on the similarity calculation layer.

[0125] S603: Calculate an original classification score based on the visual features and the similarity matrix.

[0126] S605: Calculate the feature vector mean and covariance matrix based on the visual features, and calculate the Gaussian discriminant classification score.

[0127] S607: Determine a key-value pair based on the one-hot encoding of the visual features and the label, and calculate a cache classification score.

[0128] S609: Determine a target classification score based on the original classification score, the Gaussian discriminant classification score, and the cached classification score, and determine a drone detection result based on the target classification score.

[0129] In one embodiment of the present application, take out the test picture The visual feature X is obtained by processing the visual encoder test , and gives the following formula:

[0130]

[0131] The score of the test image is calculated by matching the obtained similarity matrix S, which represents the original classification score CLIP obtained by the original classifier. score , and gives the following formula:

[0132] CLIP score =X test S T

[0133] Among them, CLIP score Represents the classification probability of the predicted test image on all N categories, that is, the classification performance of the pre-trained original classifier.

[0134] Perform statistical analysis based on the visual features I of the training set and calculate the mean μ of the feature vector of each category i And covariance matrix ∑, calculate the Gaussian discriminant classification score GDA of the test image based on the eigenvector mean and covariance matrix score , and gives the following formula:

[0135] GDA score =X test W T +b

[0136] Where W is the weight of the test image, W = Σ -1 μ i ,Σ -1 represents the inverse of the covariance matrix, b is the bias of the test image,

[0137] The visual features of the small sample training set and their labels are stored as key-value pairs in the cache model, and the X of the test image is calculated. test Cache classification score CACHE score , and gives the following formula:

[0138] CACHE score =G·T

[0139] Where G is the relationship score between the query and the key, and is given by the following formula:

[0140] G=exp(-β(1-X test I T ))

[0141] Among them, β represents the hyperparameter whose value is the optimal moment of classification accuracy selected by the grid search method, and X test I T For the test feature X test With all small sample training set features I T The cosine similarity between .

[0142] Finally, the final classification score logits is calculated, the category with the highest score is used as the predicted label, and the result of whether the drone is detected is displayed in a visual way. The following formula is given:

[0143] logits=CLIP score +αGDA score +(1-α)CACHE score

[0144] Among them, α is the hyperparameter for balancing the Gaussian discriminant classifier and the cache classifier. The hyperparameter is set adaptively by grid search. The more the target dataset conforms to the premise of Gaussian distribution with the same covariance, the higher the accuracy of the Gaussian discriminant classifier. Increasing the hyperparameter α can increase GDA. score At the same time, since the cache classifier and the Gaussian discriminant classifier are classifiers calculated or cached from the visual features of the small sample training set, the α balance factor achieves a balance between the output scores of the two classifiers, making the joint cache Gaussian classifier more stable.

[0145] In the above-mentioned small-sample drone identification method based on integrated communication and perception signals, first, the original transmitted signal is subjected to orthogonal time-frequency-space modulation to generate a transmitted signal sequence and transmit it. Next, the target's reflected signal is received, and the reflected signal is subjected to orthogonal time-frequency-space demodulation and filtering to obtain a target spectrum feature image. Next, the target spectrum feature image is subjected to image preprocessing to extract the micro-Doppler spectrum features of the drone and different background clutter. Finally, an adaptive classifier is constructed, the micro-Doppler spectrum features are input into the adaptive classifier for training, and the micro-Doppler spectrum features to be identified are input into the trained adaptive classifier to identify drones in the scene. In other words, only a small sample of data is required to summarize image features and classify images, thereby achieving drone identification and detecting the background in which the drone is located. This method reduces the time cost required by current mainstream drone identification methods and can effectively identify drones in various backgrounds.

[0146] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0147] Based on the same inventive concept, the embodiments of the present application also provide a small sample drone identification device based on communication-sensing integrated signals for implementing the aforementioned small sample drone identification method based on communication-sensing integrated signals. The implementation solution provided by this device is similar to the implementation solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the small sample drone identification device based on communication-sensing integrated signals provided below can be found in the above-mentioned limitations of the small sample drone identification method based on communication-sensing integrated signals, and will not be repeated here.

[0148] In one embodiment, Figure 4 As shown, a small sample drone identification device 400 based on communication and perception integrated signals is provided, including: a transmission signal processing module 401, a reflection signal processing module 403, an image preprocessing module 405 and a drone identification module 407, wherein:

[0149] The transmission signal processing module 401 is used to perform orthogonal time-frequency-space modulation on the original transmission signal to generate a transmission signal sequence and transmit it.

[0150] The reflected signal processing module 403 is used to receive the reflected signal of the target, perform orthogonal time-frequency-space demodulation and filtering on the reflected signal, and obtain a target spectrum feature image;

[0151] The image preprocessing module 405 is used to perform image preprocessing on the target spectrum feature image to extract the micro-Doppler spectrum features of the UAV and different background clutter.

[0152] The drone identification module 407 is used to build an adaptive classifier, input the micro-Doppler spectrum features into the adaptive classifier for training, input the micro-Doppler spectrum features to be identified into the trained adaptive classifier, and identify drones in the scene.

[0153] In one embodiment of the present application, the transmission signal processing module is further configured to:

[0154] Map the input bit stream modulation into a delay-Doppler domain matrix;

[0155] Performing an inverse symplectic finite Fourier transform on the delay-Doppler domain matrix to map it into a time-frequency domain matrix;

[0156] Performing Heisenberg transform and windowing processing on the time-frequency domain matrix to obtain a time-domain transmission signal matrix.

[0157] In one embodiment of the present application, the transmission signal processing module is further configured to:

[0158] A cyclic prefix is inserted into the time domain transmit signal matrix to obtain a transmit signal sequence.

[0159] In one embodiment of the present application, the reflected signal processing module is further configured to:

[0160] Perform serial-to-parallel conversion, windowing, and Wigner transform on the reflected signal to obtain a time-frequency domain symbol matrix;

[0161] Performing a symplectic finite Fourier transform on the time-frequency domain symbol matrix to obtain a delay-Doppler domain received signal matrix;

[0162] Matched filtering and range filtering are performed on the delay-Doppler domain received signal matrix, and short-time Fourier transform is performed to obtain a target spectrum feature image.

[0163] In one embodiment of the present application, the reflected signal processing module is further configured to:

[0164] Performing matched filtering on the delay-Doppler domain received signal matrix, counting the peak values of the matched filtering results, and obtaining the delay taps corresponding to the peak values;

[0165] Perform distance filtering on the elements in the time delay tap to eliminate communication symbol interference.

[0166] In one embodiment of the present application, the adaptive classifier includes an input layer, an encoder layer, a feature mapping layer, a similarity calculation layer and a contrast loss optimization layer, wherein the encoder layer includes a visual encoder and a text encoder.

[0167] In one embodiment of the present application, the drone identification module is further configured to:

[0168] Visual features are obtained based on the visual encoder, and a similarity matrix is obtained based on the similarity calculation layer;

[0169] Calculating a raw classification score based on the visual features and the similarity matrix;

[0170] Calculating a feature vector mean and a covariance matrix based on the visual features, and calculating a Gaussian discriminant classification score;

[0171] Determine key-value pairs based on one-hot encoding of visual features and labels, and calculate cache classification scores;

[0172] A target classification score is determined based on the original classification score, the Gaussian discriminant classification score, and the cache classification score, and a drone detection result is determined based on the target classification score.

[0173] Each module in the aforementioned small-sample drone identification device based on integrated communication and perception signals can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor within a computer device in hardware form, or stored in a computer device's memory as software, allowing the processor to call and execute the corresponding operations of each module.

[0174] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 5As shown. The computer device includes a processor, memory, communication interface, display screen and input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be achieved through WIFI, mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a small sample drone identification method based on integrated communication and perception signals is implemented. The display screen of the computer device can be a liquid crystal display or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a key, trackball or touchpad provided on the computer device housing, or an external keyboard, touchpad or mouse.

[0175] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0176] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0177] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0178] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0179] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0180] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processor involved in the various embodiments provided herein may be, but are not limited to, a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, and the like.

[0181] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0182] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A small sample drone identification method based on communication and perception integrated signals, characterized in that: The method comprises: Perform orthogonal time-frequency-space modulation on the original transmission signal to generate a transmission signal sequence and transmit it; receiving a reflected signal from a target, performing orthogonal time-frequency-space demodulation and filtering on the reflected signal to obtain a target spectrum feature image; Performing image preprocessing on the target spectrum feature image to extract micro-Doppler spectrum features of the UAV and different background clutter; An adaptive classifier is constructed, the micro-Doppler spectrum features are input into the adaptive classifier for training, and the micro-Doppler spectrum features to be identified are input into the trained adaptive classifier to identify the drone in the scene.

2. The small sample drone identification method based on communication-perception integrated signals according to claim 1 is characterized in that: The performing orthogonal time-frequency-space modulation on the original transmission signal to generate a transmission signal sequence and transmitting the sequence includes: Map the input bit stream modulation into a delay-Doppler domain matrix; Performing an inverse symplectic finite Fourier transform on the delay-Doppler domain matrix to map it into a time-frequency domain matrix; Performing Heisenberg transform and windowing processing on the time-frequency domain matrix to obtain a time-domain transmission signal matrix.

3. The small sample drone identification method based on communication and perception integrated signals according to claim 2 is characterized in that: The performing orthogonal time-frequency-space modulation on the original transmission signal to generate a transmission signal sequence and transmitting the sequence further comprises: A cyclic prefix is inserted into the time domain transmit signal matrix to obtain a transmit signal sequence.

4. The small sample drone identification method based on communication-perception integrated signals according to claim 1 is characterized in that: The receiving target's reflected signal, performing orthogonal time-frequency-space demodulation and filtering on the reflected signal to obtain the target spectrum feature image includes: Perform serial-to-parallel conversion, windowing, and Wigner transform on the reflected signal to obtain a time-frequency domain symbol matrix; Performing a symplectic finite Fourier transform on the time-frequency domain symbol matrix to obtain a delay-Doppler domain received signal matrix; Matched filtering and range filtering are performed on the delay-Doppler domain received signal matrix, and short-time Fourier transform is performed to obtain a target spectrum feature image.

5. The small sample drone identification method based on communication-perception integrated signals according to claim 4 is characterized in that: The performing matched filtering and range filtering on the delay-Doppler domain received signal matrix and performing short-time Fourier transform to obtain a target spectrum feature image includes: Performing matched filtering on the delay-Doppler domain received signal matrix, counting the peak values of the matched filtering results, and obtaining the delay taps corresponding to the peak values; Perform distance filtering on the elements in the time delay tap to eliminate communication symbol interference.

6. The small sample drone identification method based on communication-perception integrated signals according to claim 1 is characterized in that: The adaptive classifier includes an input layer, an encoder layer, a feature mapping layer, a similarity calculation layer and a contrast loss optimization layer, wherein the encoder layer includes a visual encoder and a text encoder.

7. The small sample drone identification method based on communication-perception integrated signals according to claim 6 is characterized in that: Inputting the micro-Doppler spectrum features to be identified into the trained adaptive classifier to identify the drone in the scene includes: Visual features are obtained based on the visual encoder, and a similarity matrix is obtained based on the similarity calculation layer; Calculating a raw classification score based on the visual features and the similarity matrix; Calculating a feature vector mean and a covariance matrix based on the visual features, and calculating a Gaussian discriminant classification score; Determine key-value pairs based on one-hot encoding of visual features and labels, and calculate cache classification scores; A target classification score is determined based on the original classification score, the Gaussian discriminant classification score, and the cache classification score, and a drone detection result is determined based on the target classification score.

8. A small sample drone identification device based on integrated communication and perception signals, characterized in that: The device comprises: The transmission signal processing module is used to perform orthogonal time-frequency-space modulation on the original transmission signal, generate a transmission signal sequence and transmit it; A reflection signal processing module is used to receive the reflection signal of the target, perform orthogonal time-frequency-space demodulation and filtering on the reflection signal, and obtain a target spectrum feature image; An image preprocessing module is used to perform image preprocessing on the target spectrum feature image to extract micro-Doppler spectrum features of the UAV and different background clutter; The drone identification module is used to construct an adaptive classifier, input the micro-Doppler spectrum features into the adaptive classifier for training, input the micro-Doppler spectrum features to be identified into the trained adaptive classifier, and identify drones in the scene.

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 method according to 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 method according to any one of claims 1 to 7 are implemented.

Images