Radar radiation source individual identification method and device for airport low-altitude security

Abstract

The embodiment of the invention discloses a radar radiation source individual identification method and device for airport low-altitude security and protection. A specific embodiment of the method comprises the following steps: generating a radar signal fragment group set; performing self-supervised pre-training on the radiation source feature extraction network to update model parameters; constructing a radar radiation source individual identification model; performing joint optimization on the radar radiation source individual identification model through the known radiation source signal group set to generate a known radiation source center feature set; generating a known radiation source boundary parameter set according to the radar radiation source individual identification model and the known radiation source center feature set; and according to the known radiation source boundary parameter set and the radar radiation source individual identification model, performing radiation source individual identification on the real-time radar signal to generate a radar radiation source identification result. According to the embodiment, the generalization ability of the recognition model can be improved under the small sample condition, so that the individual recognition accuracy of the radiation source is improved, and the low-altitude security and protection safety of an airport is improved.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, and more specifically to a method and apparatus for individual radar radiation source identification for low-altitude security at airports. Background Technology

[0002] Bird deterrence radar or low-power surveillance radar is typically deployed around airport runways, taxiways, and air routes to detect flocks of birds and direct their movement. Meanwhile, with the widespread use of drones, airports are increasingly adopting drones equipped with passive radio frequency receivers and small radars for low-altitude security tasks. Currently, when identifying individual radiation sources in open environments such as airports, it is often assumed that a sufficient and diverse range of known category samples are used to train the identification model.

[0003] However, in real-world applications, airport bird deterrence radars and drone-mounted radars may differ in manufacturer, batch, or firmware version, and new devices of the same model may appear during maintenance and replacement. Furthermore, due to limitations such as security controls, data collection windows, and equipment operating times, airports cannot consistently collect large amounts of labeled data for each radar device, resulting in limited usable sample data for each device. This leads to poor generalization ability of the identification model and low accuracy in identifying individual radiation sources, ultimately reducing the security of low-altitude airport security. Summary of the Invention

[0004] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0005] Some embodiments of this disclosure propose a method and apparatus for individual radar radiation source identification for low-altitude airport security, in order to solve the technical problems mentioned in the background section above.

[0006] In a first aspect, some embodiments of this disclosure provide a method for identifying individual radar radiation sources for low-altitude security at airports. The method includes: generating a set of radar signal segments based on collected raw radar signals, wherein the raw radar signals are unmarked radar signals received by different receiving devices located in the airport area, including receiving stations and airborne receivers of patrol drones; performing self-supervised pre-training on a radiation source feature extraction network based on the radar signal segment set to update the network's model parameters; constructing an individual radar radiation source identification model based on the radiation source feature extraction network; and identifying individual radar radiation sources using the known set of radiation source signals. The radar source individual identification model is jointly optimized to generate a set of known radiation source center features, wherein the set of known radiation source signal groups is a labeled small sample dataset collected in the airport area. Based on the radar source individual identification model and the set of known radiation source center features, a set of known radiation source boundary parameters is generated, wherein each known radiation source boundary parameter corresponds to a known radar source individual in the airport area. Based on the set of known radiation source boundary parameters and the radar source individual identification model, the real-time radar signals received by each receiving device in the airport area are identified as radiation sources to generate radar source identification results.

[0007] Secondly, some embodiments of this disclosure provide a radar radiation source individual identification device for low-altitude security at airports. The device includes: a first generation unit configured to generate a set of radar signal segments based on various collected original radar signals, wherein the original radar signals are unmarked radar signals received by different receiving devices located in the airport area, including receiving stations and airborne receivers of patrol drones; a self-supervised pre-training unit configured to perform self-supervised pre-training on a radiation source feature extraction network based on the radar signal segment set to update the model parameters of the radiation source feature extraction network; a construction unit configured to construct a radar radiation source individual identification model based on the radiation source feature extraction network; and a joint optimization unit configured to optimize the model using known... A set of radiation source signal groups is used to jointly optimize the radar radiation source individual identification model to generate a set of known radiation source center features, wherein the set of known radiation source signal groups is a labeled small sample dataset collected in the airport area; a second generation unit is configured to generate a set of known radiation source boundary parameters based on the radar radiation source individual identification model and the set of known radiation source center features, wherein each known radiation source boundary parameter corresponds to a known radar radiation source individual in the airport area; an individual identification unit is configured to perform radiation source individual identification on the real-time radar signals received by each receiving device in the airport area based on the set of known radiation source boundary parameters and the radar radiation source individual identification model to generate radar radiation source identification results.

[0008] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0009] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0010] The above-described embodiments of this disclosure have the following beneficial effects: the radar radiation source individual identification method for airport low-altitude security through some embodiments of this disclosure can improve the generalization ability of the identification model under small sample conditions, thereby improving the accuracy of radiation source individual identification and improving airport low-altitude security. Specifically, the reasons for the low accuracy of radiation source individual identification and the reduced airport low-altitude security are as follows: In actual application scenarios, airport bird deterrence radar and UAV-mounted radar may have differences in manufacturers, batches, or firmware versions, and new equipment of the same model will appear with maintenance and replacement. In addition, due to limitations such as security control, data collection window, and equipment operating time, airports cannot collect a large amount of labeled data for each radar device for a long period of time, resulting in less usable sample data for each radar device, which leads to poor generalization ability of the identification model and low accuracy of radiation source individual identification, making it difficult to effectively identify and determine unknown radiation sources, thus leading to a decline in airport low-altitude security. Based on this, the radar radiation source individual identification method for airport low-altitude security through some embodiments of this disclosure first generates a set of radar signal segments based on the collected original radar signals. The aforementioned raw radar signals are unmarked radar signals received by different receiving devices located in the airport area, including receiving stations and airborne receivers from patrol drones. Therefore, by dividing the raw radar signals continuously received by different receiving devices within the airport area into radar signal segments, and grouping radar signal segments belonging to the same radiation source based on radar signal correlation features, a set of radar signal segment groups is obtained. Then, based on the aforementioned set of radar signal segment groups, the radiation source feature extraction network undergoes self-supervised pre-training to update its model parameters. Thus, without manual annotation, the radiation source feature extraction network learns to maintain similar stable features for each type of signal from the same source under different noise and reception conditions, thereby improving its adaptability to noise, obstruction, and time-varying channels in the actual airport environment. Subsequently, based on the aforementioned radiation source feature extraction network, a radar radiation source individual recognition model is constructed. Therefore, the pre-trained radiation source feature extraction network with stable feature extraction capabilities can be concatenated with a classification output head to obtain a radar radiation source individual recognition model that can convert signal segments into signal segment features and output discriminable categories. Next, the radar source individual identification model is jointly optimized using the known set of radiation source signals to generate a set of known radiation source center features. This set of known radiation source signals is a labeled small-sample dataset collected from the aforementioned airport area.Therefore, under the hard constraint of a small sample dataset, joint loss is used to make the features of each known radar source more clustered under different acquisition periods and different receiving devices, forming stable and more representative central features of known sources. This also exacerbates the differences in central features between different individual radar sources, reducing the probability of misidentification of radar sources of the same model but different batches or firmware versions. Secondly, based on the aforementioned radar source individual identification model and the aforementioned set of known source central features, a set of known source boundary parameters is generated. Each known source boundary parameter corresponds to a known radar source individual within the aforementioned airport area. This allows the establishment of boundary parameters representing the normal range of variation for each known radar source individual. This enables the same device to experience certain feature drift due to fluctuations in the airport environment and operating conditions, while also determining the maximum boundary of feature drift. This significantly reduces the probability of misidentification when encountering new equipment or abnormal features caused by strong interference. Finally, based on the aforementioned set of known source boundary parameters and the aforementioned radar source individual identification model, radar source individual identification is performed on the real-time radar signals received by each receiving device within the aforementioned airport area to generate radar source identification results. Therefore, by averaging the features of multiple segments to suppress instantaneous disturbances, and then combining the boundary parameters to verify the credibility of the classification results, the accuracy of known device identification is improved, while the risk of new devices being misidentified as known devices under abnormal conditions is also reduced, thereby improving the reliability and security of airport low-altitude security. Attached Figure Description

[0011] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0012] Figure 1 This is a flowchart of some embodiments of the radar radiation source individual identification method for low-altitude security at airports according to the present disclosure; Figure 2 This is a schematic diagram of the process for generating signal comparison loss values ​​in the radar radiation source individual identification method for low-altitude security at airports disclosed herein. Figure 3 This is a schematic diagram of the radiation source feature extraction network structure in the radar radiation source individual identification method for low-altitude security at airports disclosed herein. Figure 4 This disclosure presents a feature visualization diagram of a radar radiation source individual identification method for low-altitude security at airports. Figure 5 This is a schematic diagram of the structure of some embodiments of a radar radiation source individual identification device for low-altitude security at airports according to the present disclosure. Figure 6 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0013] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0014] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0015] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0016] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0017] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0018] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] Figure 1 A flow 100 of some embodiments of a radar emitter identification method for low-altitude airport security according to this disclosure is shown. The radar emitter identification method for low-altitude airport security includes the following steps: Step 101: Construct a set of radar signal segment groups based on the collected raw radar signals.

[0020] In some embodiments, the implementing entity (e.g., a computing device) of the radar radiation source individual identification method for low-altitude security at airports can construct a set of radar signal segments based on the collected raw radar signals. These raw radar signals are unmarked radar signals received by receiving devices located in different areas of the airport. The receiving devices include receiving stations and airborne receivers from patrol drones. The airport area can refer to the low-altitude airspace within a 3km radius of the airport terminal or apron. The low-altitude airspace can refer to flight areas below 300 meters. Each collected raw radar signal can be a radar signal emitted by a radar radiation source within the airport area under actual operational conditions and received in real time by the receiving device, without any radiation source identification information. Specifically, the raw radar signal can be a continuous I / Q sampling signal. The receiving device can be a device with radio reception capabilities, capable of collecting electromagnetic signals from radar radiation sources and outputting digital signal data. The patrol drone can be a drone performing inspection, monitoring, or reconnaissance tasks within the airport area according to a preset route or real-time commands, typically equipped with airborne receiving devices. In practice, the aforementioned executing entity can divide each original radar signal into individual radar signal segments according to a fixed number of signal frames, and group the radar signal segments belonging to the same original radar signal into a set of radar signal segment groups.

[0021] It should be noted that the aforementioned computing devices can be either hardware or software. When the computing device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device. When the computing device is software, it can be installed on the hardware devices listed above. It can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here. It should be understood that the number of the aforementioned computing devices can be arbitrary, depending on the implementation requirements.

[0022] In some optional implementations of certain embodiments, the aforementioned execution entity may construct a set of radar signal segments based on the collected raw radar signals through the following steps: The first step is to perform the following time-domain processing steps for each raw radar signal acquired: The first sub-step involves performing time-frequency transformation processing on the original radar signal to generate signal time-frequency energy information. In practice, the executing entity can perform a short-time Fourier transform (STFT) or a continuous wavelet transform (CWT) on the original radar signal, thereby mapping the continuous-time signal (i.e., the transformed original radar signal) onto the time-frequency plane to obtain a time-frequency representation. Then, the executing entity can use the energy distribution of the time-frequency representation as the signal time-frequency energy information.

[0023] The second sub-step involves noise estimation of the aforementioned signal time-frequency energy information to generate a signal noise basis and a signal noise estimation scale. In practice, the executing entity can take the lower quantile (e.g., 10th percentile) of the aforementioned signal time-frequency energy information for each frequency channel as the signal noise basis, and determine the signal noise estimation scale by performing a moving average or exponential smoothing on the signal time-frequency energy information.

[0024] The third sub-step involves extracting regional features from the aforementioned signal-noise basis and signal-noise estimation scale to obtain a time-frequency regional feature set. In practice, firstly, the executing entity subtracts the signal-noise basis from the corresponding time-frequency energy value of each time-frequency unit included in the signal-noise energy information. Then, the difference is divided by the signal-noise estimation scale to obtain a normalized energy value. Through this processing, a normalized energy matrix (i.e., a normalized saliency map) reflecting the significance of a time-frequency unit relative to background noise is obtained. This normalized energy matrix represents the deviation of signal energy from the noise level at each time-frequency location on the time-frequency plane. Then, the executing entity performs threshold segmentation on the normalized saliency map using a preset signal energy threshold. When the normalized energy value corresponding to a certain time-frequency unit is greater than the preset signal energy threshold, the time-frequency unit can be determined to be a unit that may contain valid radar signals; when it is less than or equal to the preset signal energy threshold, the time-frequency unit can be determined to be a noise unit. Through this threshold segmentation process, a binary time-frequency mask is obtained. Next, the execution entity can perform morphological processing on the aforementioned binary time-frequency mask to eliminate isolated noise points and enhance the integrity of continuous signal regions. This morphological processing may include, but is not limited to, opening operations, closing operations, or connected component analysis. Connected component analysis can refer to extracting multiple connected regions composed of adjacent time-frequency units on the time-frequency plane. Each connected region corresponds to a candidate signal region continuously distributed in the time and frequency dimensions. Each candidate signal region can characterize a possible radar transmission activity or effective signal segment, serving as a time-frequency candidate event. Finally, for each time-frequency candidate event (i.e., a candidate signal region), the execution entity can determine region-level feature parameters within the corresponding time-frequency region as time-frequency region features. These region-level feature parameters may include, but are not limited to, the start time and duration of the time-frequency candidate event in the time dimension, the bandwidth range in the frequency dimension, the sum of the energy of all time-frequency units within the candidate signal region, the energy-weighted centroid location, and the region area.

[0025] The fourth sub-step involves dividing the original radar signal into time-domain segments based on the aforementioned time-frequency region feature set, resulting in individual radar signal segments. In practice, firstly, the executing entity can divide the original radar signal into time-domain segments according to the start time included in each time-frequency region feature, obtaining individual radar signal segments. Then, the executing entity can use a clustering algorithm to cluster the aforementioned time-frequency region features according to the feature similarity or feature distance (e.g., cosine similarity or Euclidean distance) between the various time-frequency region features in the time-frequency region feature set, obtaining individual time-frequency region feature clusters. Then, for each time-frequency region feature cluster, the executing entity merges the radar signal segments corresponding to time-adjacent time-frequency region features (e.g., 1 minute 02 seconds to 1 minute 12 seconds and 1 minute 12 seconds to 1 minute 14 seconds are temporally connected) into one radar signal segment according to the start time included in the time-frequency region feature. As an example, the aforementioned clustering algorithm can be a density-based spatial clustering of applications with noise (DBSCAN) algorithm or a hierarchical clustering algorithm.

[0026] The fifth sub-step involves feature extraction for each radar signal segment to generate radar signal segment features. In practice, the aforementioned execution entity can determine the signal center frequency, signal bandwidth, number of spectral peaks, and signal energy distribution from each radar signal segment, and then concatenate these parameters to generate radar signal segment features.

[0027] The second step involves classifying the generated radar signal segments into groups based on their shared origin, thus obtaining a set of radar signal segment groups. In practice, the execution entity can use the aforementioned clustering algorithm to cluster the features of each radar signal segment according to the feature similarity or feature distance between them (e.g., cosine similarity or Euclidean distance), resulting in feature clusters for each radar signal segment. Then, the execution entity can determine the radar signal segments corresponding to each feature cluster as a radar signal segment group.

[0028] Step 102: Based on the radar signal segment set, perform self-supervised pre-training on the radiation source feature extraction network to update the model parameters of the radiation source feature extraction network.

[0029] In some embodiments, the execution entity can perform self-supervised pre-training on the radiation source feature extraction network based on the aforementioned set of radar signal segments to update the model parameters of the radiation source feature extraction network. The radiation source feature extraction network can be a neural network model for feature extraction from radar signals. As an example, the radiation source feature extraction network can be, but is not limited to, a convolutional neural network, a recurrent neural network model, or a temporal neural network model. In practice, firstly, the execution entity can directly use each radar signal segment from the aforementioned set of radar signal segments as input to the radiation source feature extraction network to generate radiation source features for each radar signal segment. Then, the execution entity can determine the loss function value of the radiation source feature extraction network using the generated radiation source features of each radar signal segment and a self-supervised loss function (e.g., a contrastive loss function, i.e., an information entropy loss function). Afterward, the parameters of the radiation source feature extraction network are updated using the determined loss function value and a backpropagation algorithm. Finally, the radiation source feature extraction network is trained for K rounds using the aforementioned set of radar signal segments until the difference between the loss function value determined in the previous training round and the loss function value determined in the current training round is less than or equal to a preset loss threshold. At this point, the radiation source feature extraction network is considered to have converged.

[0030] It should be noted that in airport low-altitude security scenarios, existing neural network models for radiation source feature extraction mostly employ fixed-scale convolutional structures or single-level feature representations, typically relying on supervised training with labeled data under limited reception conditions. When faced with highly time-varying channels, multipath effects, and changes in acquisition perspective introduced by mobile receiving platforms such as drone patrols, these models are prone to mislearning environmental disturbances and differences in reception conditions as radiation source features, leading to significant diffusion of feature distribution for the same radar equipment at different receiving locations or time periods. Furthermore, due to airport security controls and equipment operating conditions, the number of known labeled radiation source samples actually available for training is limited. Existing models struggle to stably distinguish individual differences in radar equipment under small sample conditions, resulting in insufficient generalization ability and poor feature extraction capabilities for radiation source feature extraction networks.

[0031] In some optional implementations of certain embodiments, the aforementioned execution entity may perform self-supervised pre-training on the radiation source feature extraction network based on the aforementioned set of radar signal fragments to update the model parameters of the radiation source feature extraction network through the following steps: The first step is to perform the following model training steps for each radar signal segment group in the above radar signal segment group set: The first sub-step involves performing sample signal enhancement processing on the aforementioned radar signal segment group to generate a set of signal segment sample pairs. In practice, the aforementioned execution entity can perform sample signal enhancement processing on the aforementioned radar signal segment group through time shifting and noise injection processing to generate a set of signal segment sample pairs.

[0032] The second sub-step involves performing the following self-supervised training steps based on the aforementioned set of signal segment sample pairs: Sub-step one involves extracting a predetermined batch number of signal segment sample pairs from the aforementioned signal segment sample pair set and inputting them into the radiation source feature extraction network to determine the batch signal loss value. In practice, firstly, the aforementioned execution entity can directly extract a predetermined batch number of signal segment sample pairs (x1, x2, ..., x...) from the signal segment sample pair set. B Each signal segment sample pair x i Includes x i 1 (i.e., the first enhanced signal segment), x i 2 (i.e., the second enhanced signal segment). Then, the aforementioned execution entity can use the first and second enhanced signal segments contained in each signal segment sample pair as positive samples, and the first and second enhanced signal segments contained in other signal segment sample pairs in the same batch as negative samples. Then, for each signal segment sample pair, the aforementioned execution entity can perform the following steps: First, determine a first intermediate value using the following first expression: exp(sim(z) i 1 , z i 2 ) / τ). Where, z i 1 z i 2 These are the signal segment features generated by the radiation source feature extraction network, corresponding to the first and second enhanced signal segments in the signal segment sample pair, respectively. `sim()` is the similarity function (e.g., cosine similarity function). `τ` is the temperature coefficient, controlling the distribution smoothness. `exp()` is the exponential function. The second step is to determine each second intermediate value using the following second expression: `exp(sim(z)` i 1 ， z k ) / τ). Where, z k This could be a signal segment feature corresponding to a negative sample in the same batch. Therefore, by substituting the signal segment feature corresponding to each negative sample into the expression, various second intermediate values ​​can be generated. The third step is to determine each third intermediate value using the following third expression: exp(sim(z) i 2, z k By substituting the signal segment features corresponding to each negative sample into the expression, each third intermediate value can be generated. The fourth step involves generating the signal contrast loss value using -log[first intermediate value / [(sum of all second intermediate values) + (sum of all third intermediate values)]]. Finally, the aforementioned execution entity can determine the sum of the generated signal contrast loss values ​​as the batch signal loss value.

[0033] like Figure 2 The signal contrast loss value generation process shown can be implemented in practice by performing time-shifting and noise-in / out processing on a radar signal segment 201 to generate a first enhanced signal segment 202 and a second enhanced signal segment 203 as a signal segment sample pair. Then, the execution entity can treat the first enhanced signal segment 202 and the second enhanced signal segment 203 as positive samples and input them into the radiation source feature extraction network 204 to obtain the corresponding first signal segment feature 205 and second signal segment feature 206. Afterwards, the execution entity can generate a first intermediate value 210 using the first signal segment feature 205, the second signal segment feature 206, and the first expression. Then, the aforementioned execution entity can use the first enhanced signal segment and the second enhanced signal segment included in other signal segment sample pairs in the same batch as negative samples. Using the signal segment feature 207 corresponding to each negative sample, the first signal segment feature 205, and the second expression, it generates various second intermediate values ​​209. Similarly, using the signal segment feature 208 corresponding to each negative sample, the second signal segment feature 206, and the third expression, it generates various third intermediate values ​​211. Finally, the aforementioned execution entity can generate a signal contrast loss value 212 using the various second intermediate values ​​209, the first intermediate value 210, and the various third intermediate values ​​211.

[0034] like Figure 3 The radiation source feature extraction network structure shown may include a feature mapping layer 301, a residual backbone network 302, a cross-layer feature fusion layer 303, and an embedding layer 304.

[0035] The aforementioned radiation source feature extraction network takes signal segments (including radar signal segments, first enhanced signal segments, and second enhanced signal segments) as input. The signal segments are complex baseband radar signals acquired by the receiving device. Before entering the network, the real and imaginary parts of the signal are arranged as different channels, thus constructing a dual-channel one-dimensional real-valued input tensor, denoted as X∈R. B*2*L Where R represents a real number, B represents the batch size, and L represents the number of sampling points of the signal segment in the time dimension, thus completely preserving the transmission structure and transient characteristics of the radar signal in the time dimension.

[0036] The aforementioned feature mapping layer 301 is used to map the original radar signal to a high-dimensional feature space. This layer consists of a one-dimensional convolutional layer, a batch normalization layer, and a nonlinear activation function (such as the SiLU activation function) connected sequentially. The one-dimensional convolutional layer uses a 7×7 kernel size (i.e., k=7) and a stride of 2 (i.e., s=2) to expand the receptive field in the time dimension, thereby capturing the local structural features of the radar pulse and obtaining the feature mapping result F1. In practice, the aforementioned feature mapping layer 301 can be expressed as F1=SiLU(BN(Conv1D) k=7，s=2 (X))). Here, Conv1D() can refer to a one-dimensional convolutional layer. BN() can refer to a normalization layer. SiLU() can refer to the SiLU activation function. Subsequently, the feature mapping result F1 can be further downsampled in the time dimension through a one-dimensional max-pooling layer to reduce the feature sequence length and suppress local noise interference, resulting in F1′∈R. B *C1*(L / 4) In the above, C1 represents the initial number of feature channels, which can be 64.

[0037] The aforementioned residual backbone network 302 consists of multiple sequentially connected feature extraction stages (e.g., five). Each feature extraction stage is composed of several stacked selective kernel residual blocks. Between each feature extraction stage, one-dimensional convolutions with strides are used to progressively reduce the temporal resolution and progressively increase the number of feature channels. In each selective kernel residual block, the input feature is U∈R. B*Cin*T The input signals are fed in parallel into at least two one-dimensional convolutional branches, each using a different kernel size (e.g., k=3 and k=7) to perform multi-scale temporal modeling of the radar signal within the same layer. Here, Cin represents the number of input feature channels, and T represents the time dimension. Each convolutional branch consists of a one-dimensional convolutional layer, a normalization layer, and a SiLU activation function in sequence, outputting branch features: V1, V2 ∈ R. B*Cout*T ′. Here, V1 can be expressed as: V1=SiLU(BN(Conv1D(k=3, s=s1, p=1)(U))). V2 can be expressed as: V2=SiLU(BN(Conv1D(k=7, s=s1, p=3)(U)). s1 represents the time step of the convolution operation in the current selective kernel residual block. s1=1 indicates that the time length is not changed, only feature transformation is performed. s1=2 indicates that the time dimension is halved, i.e., downsampling is performed. T′ is the length of the output time dimension. Specifically, T′=T / s1. After the above multi-branch features enter the selective kernel fusion module, this module first performs element-wise summation on the branch features and performs global average pooling on the time dimension to obtain the channel-level statistical vector S∈R.B*Cout Where Cout is the number of output feature channels. The above statistical vector is further input into a weight generation network composed of fully connected layers, and channel weight coefficients corresponding to each branch are generated through nonlinear mapping and Softmax operation. Using the generated channel weight coefficients, the output features of each convolutional branch are weighted and summed to obtain the adaptively fused feature representation. The fused feature representation is then added element-wise with the residual branch. When the input and output features are inconsistent in the number of channels or temporal resolution, the residual branch is projected and aligned through a one-dimensional 1×1 convolution. Finally, the residual summation result is output through a nonlinear activation function to form the output feature of the selective kernel residual block. Through the cascading stacking of multiple stages, the above residual backbone network 302 maps the radar signal from low-level temporal texture features to high-level structured radiation source representation layer by layer, outputting multi-scale feature representations, denoted as C2, C3, C4, and C5.

[0038] It should be noted that, Figure 3 The number of selective kernel residual blocks included in each feature stage shown is only illustrative, and the actual number of selective kernel residual blocks is not limited here.

[0039] The aforementioned multi-scale feature representations C2, C3, C4, and C5 progressively advance in temporal resolution and semantic hierarchy. To fully utilize the complementary information of features at different levels, these multi-scale features are input into the aforementioned cross-layer feature fusion layer 303 for cross-layer feature fusion. During the cross-layer feature fusion process, the features of each layer are first aligned using a one-dimensional 1×1 convolution to obtain C... ^ i=Conv1D 1×1 (Ci). (i takes values ​​from 2 to 5) Then, a top-down feature fusion method similar to a feature pyramid network (FNP) is adopted. High-level features are upsampled to the same temporal resolution as low-level features, and then element-wise added to the corresponding level features to form the fused feature P5=C. ^ 5. P4=C ^ 4 + Up(P5), P3 = C ^ 3 + Up(P4), P2 = C ^ 2+Up(P3). Here, UP() is an upsampling operation using linear interpolation. Then, for each P... i (i takes values ​​from 2 to 5) Perform a smooth convolution operation, the specific expression is as follows: SiLU(BN(Conv1D(k=3,s=1,p=1)(P i())). Here, p represents the padding value. Therefore, after completing the stepwise fusion, the fusion result of each layer is smoothed by one-dimensional convolution, normalization, and nonlinear activation to eliminate the discontinuities introduced during the scale transformation. Through the cross-layer feature fusion layer 303, the network obtains a set of multi-scale feature representations that simultaneously contain fine-grained temporal information and high-level stable structural information at the output stage.

[0040] The fused multi-scale features are then input into the embedding layer 304. First, a global average pooling (GAP) operation is performed on each scale feature along the time dimension, compressing the variable-length temporal features into a fixed-length feature vector g. i =GAP(P i Then, the feature vectors from different scales are concatenated along the channel dimension to form a multi-scale joint feature representation, i.e., g=[g2, g3, g4, g5]. The above multi-scale joint feature g is further input into a projection network composed of multiple fully connected layers. Through linear transformation, normalization and SiLU activation function, the multi-scale joint feature is mapped to the embedding feature space of a preset dimension, and the embedding feature e (i.e. the radar signal segment feature corresponding to the radar signal segment) is output.

[0041] Sub-step two: In response to the determination that the radiation source feature extraction network has not converged, the parameters of the radiation source feature extraction network are updated based on the batch signal loss value, and the self-supervised training step described above is executed again. In practice, the executing entity can determine that the radiation source feature extraction network has not converged if the difference between the batch signal loss value determined in the previous execution of the self-supervised training step and the batch signal loss value determined in the current execution of the self-supervised training step is greater than the preset loss threshold. Then, the parameters of the radiation source feature extraction network are updated using the gradient descent algorithm and the backpropagation algorithm, and the self-supervised training step described above is executed again.

[0042] In some optional implementations of certain embodiments, the aforementioned execution entity may perform sample signal enhancement processing on the aforementioned radar signal segment group through the following steps to generate a set of signal segment sample pairs: The first step is to perform the following sample augmentation steps for each radar signal segment in the above radar signal segment group: The first sub-step involves injecting noise into the aforementioned radar signal segment to generate a first enhanced signal segment. In practice, the aforementioned execution entity can inject random noise into the radar signal segment x to simulate noise fluctuations and receiver thermal noise variations in the electromagnetic environment of an actual airport. As an example, the aforementioned execution entity can generate Gaussian white noise n∼N(0,σ) with the same dimension as x. 2The noise intensity σ is randomly determined according to a preset signal-to-noise ratio range, and Gaussian white noise n is superimposed onto the radar signal segment x. Here, σ can be determined based on the target signal-to-noise ratio (SNR) and the signal power P. x Estimate (e.g., σ) 2 =P x / 10 SNR / 10 ).

[0043] The second sub-step involves time-shifting the aforementioned radar signal segment to generate a second enhanced signal segment. In practice, the execution entity can perform time shifting on the radar signal segment x to simulate segment alignment errors caused by trigger point offset, inconsistent truncation windows, or propagation delay fluctuations. As an example, the execution entity can randomly sample the shift amount δ within the interval [−Δ, Δ] and perform cyclic shifting or zero-filling shifting to obtain the second enhanced signal segment.

[0044] The second sub-step involves determining the first enhanced signal segment and the second enhanced signal segment as a signal segment sample pair.

[0045] The second step is to determine the generated signal segment sample pairs as a signal segment sample pair set.

[0046] Step 103: Construct an individual radar radiation source identification model based on the radiation source feature extraction network.

[0047] In some embodiments, the execution entity can construct a radar radiation source individual identification model based on the radiation source feature extraction network. In practice, the execution entity can concatenate the radiation source feature extraction network with a classification output head to construct the radar radiation source individual identification model. The classification output head is used to output the probability values ​​of the features generated by the radiation source feature extraction network belonging to each known radiation source individual. The classification output head can be composed of a fully connected layer, a normalization layer, and a softmax function connected sequentially.

[0048] Step 104: Using the known set of radiation source signal groups, jointly optimize the radar radiation source individual identification model to generate a set of known radiation source center features.

[0049] In some embodiments, the aforementioned execution entity can jointly optimize the radiation source feature extraction network using a set of known radiation source signal groups to generate a set of known radiation source center features. This set of known radiation source signal groups is a labeled small sample dataset collected from the aforementioned airport area. Each known radiation source signal group in the set corresponds to a known radiation source individual label, representing the data source of each known radiation source signal in the set, and each known radiation source signal group contains less than or equal to 10 signal data points. In practice, firstly, the aforementioned execution entity can use the set of known radiation source signal groups to perform multiple rounds of model training on the radar radiation source individual identification model, using the cross-entropy loss function to determine the loss function value, and then using gradient descent and backpropagation algorithms to update the parameters of the radar radiation source individual identification model. Finally, the radiation source feature extraction network included in the converged radar radiation source individual identification model is used to extract the known radiation source signal features of each known radiation source signal. Then, the mean feature value of each known radiation source signal feature within each group is determined as the known radiation source center feature, resulting in the set of known radiation source center features.

[0050] In some optional implementations of certain embodiments, the aforementioned execution entity may perform joint optimization of the radiation source feature extraction network using a known set of radiation source signal groups to generate a known set of radiation source center features: The first step, for each known radiation source signal group in the above set of known radiation source signal groups, is to perform the following training steps: The first sub-step involves performing the following joint loss optimization steps based on the known radiation source signal set described above: In the first sub-step, the features of the known radiation source signal group are extracted using the radiation source feature extraction network included in the radar radiation source individual identification model, thereby obtaining the features of each known radiation source.

[0051] Sub-step two involves inputting the characteristics of each known radiation source mentioned above into the classification output head of the radar radiation source individual identification model to generate the individual identification probability of each radiation source.

[0052] Sub-step three involves generating the model loss value based on the individual identification probabilities of each radiation source. In practice, the aforementioned execution entity can generate a cross-entropy loss function as the model loss value using the known radiation source individual label represented by the individual identification probability of each radiation source, the difference between the actual known radiation source individual label of the corresponding known radiation source signal, and the cross-entropy loss function.

[0053] Sub-step four involves generating a category centering loss value based on the individual identification probabilities of each radiation source and the characteristics of each airport radiation source. In practice, the executing entity can calculate the average of the characteristics of each known radiation source corresponding to the known radiation source signal group to generate temporary centering features. Then, it can calculate the distance between each radiation source feature and the temporary centering feature, and use the average of these distances as the category centering loss value, thereby constraining the distribution of individuals of the same radiation source in the feature space to be more concentrated.

[0054] Sub-step five involves generating feature contrast loss values ​​based on the known radiation source signal groups and the radiation source feature extraction network described above. In practice, the specific implementation steps for generating the "feature contrast loss values" can refer to the specific implementation method for the "batch signal loss values" described above. For example, the known radiation source features corresponding to the known radiation source signal groups can be considered as positive samples, and the known radiation source features corresponding to different radiation source signal groups can be considered as negative samples. The feature contrast loss value is determined through a contrast loss function, thereby constraining the feature distance of the same radiation source individual to be less than the feature distance between different radiation source individuals.

[0055] Sub-step six involves generating a joint loss function value based on the generated model loss value, class center loss value, and feature contrast loss value. In practice, the joint loss function value = feature contrast loss value + model loss value + λ × class center loss value. λ is the adjustment weight.

[0056] Sub-step seven: In response to the conclusion that the radar radiation source individual identification model has converged, the known radiation source center features are generated based on the known radiation source features corresponding to the aforementioned known radiation source signal groups. In practice, the executing entity can determine the mean of the aforementioned known radiation source features as the known radiation source center features.

[0057] The second sub-step, in response to the determination that the radar source individual identification model has not converged, optimizes the parameters of the radar source individual identification model based on the joint loss function value, and repeats the aforementioned joint loss optimization step. In practice, the aforementioned executing entity can optimize the parameters of the radar source individual identification model using the aforementioned joint loss function value, the backpropagation algorithm, and the Adam optimizer, and repeat the aforementioned joint loss optimization step.

[0058] The second step is to determine the generated features of each known radiation source center as a set of known radiation source center features.

[0059] To further verify the performance of the jointly optimized radar source individual identification model, the feature extraction network included in the jointly optimized model was used to perform visual dimensionality reduction on the features extracted from the test set data. In the visualization, known class samples are marked with circles, unknown class samples are marked with pentagrams, and different radiation source individuals or categories are distinguished by color. The results are as follows: Figure 4 As shown in the visualization, features of known classes or individuals form distinct, compact, and well-defined clusters, while clear inter-class boundaries also exist between unknown and known classes. This directly benefits from the constraint of the class center loss value on intra-class aggregation and the enhancement of data augmentation invariance through contrastive learning. More importantly, this explicit and compact clustering structure for known classes naturally leaves ample empty regions in the feature space, making unknown class samples more easily identified as out-of-distribution data during testing, thus significantly improving the model's recognition performance and robustness in open environments.

[0060] Step 105: Generate a set of known radiation source boundary parameters based on the radar radiation source individual identification model and the known radiation source center feature set.

[0061] In some embodiments, the executing entity can generate a set of known radiation source boundary parameters based on the radar radiation source individual identification model and the known radiation source center feature set. Each known radiation source boundary parameter corresponds to a known radar radiation source individual within the airport area. The known radar radiation source individual refers to a specific radar device in the airport low-altitude security application scenario whose device identity, deployment location, or device number has been confirmed, and a certain number of radar signal samples have been collected and labeled. Furthermore, a known radar radiation source individual is different from a radar device model or category. Even if multiple radar devices have the same manufacturer, model, or operating parameters, as long as they belong to different physical device instances, they are considered different radar radiation source individuals. For example, the known radar radiation source individual could be a bird deterrent radar, a monitoring radar, or an airborne radar deployed on a drone. In practice, for each known radiation source center feature in the aforementioned set of known radiation source center features, the executing entity can use the known radiation source center feature as the spatial center, and the corresponding known radiation source features as the feature data sample set to construct a one-class SVM model (i.e., a single-class support vector machine model). The one-class SVM model is then trained using the aforementioned feature data sample set, and the model parameters of the trained one-class SVM model are used as the boundary parameters of the known radiation sources. These model parameters may include, but are not limited to, discriminant function parameters or model weight parameters used to determine whether a feature vector lies within the support domain. The one-class SVM model can be trained by the executing entity using a preset kernel function (e.g., Gaussian kernel function, polynomial kernel function) and regularization parameters on the aforementioned feature data sample set, thereby obtaining a discriminant function used to distinguish between "belonging to the individual radiation source" and "not belonging to the individual radiation source". After training, the one-class SVM model can determine a support domain in the feature space, such that most feature vectors in the feature sample set are located inside the support domain, while feature vectors outside the support domain are considered untrusted samples (i.e., do not belong to the individual of the radiation source).

[0062] In some optional implementations of certain embodiments, the aforementioned execution entity can generate a set of known radiation source boundary parameters based on the aforementioned radar radiation source individual identification model and the aforementioned set of known radiation source center features through the following steps: The first step, for each known radiation source center feature in the above set of known radiation source center features, is to perform the following steps: The first sub-step involves determining the spatial distance between each known radiation source feature and the aforementioned known radiation source center feature as the feature distance. This spatial distance can be an Euclidean distance.

[0063] The second sub-step involves selecting at least one feature distance from the aforementioned feature distances that satisfies the spacing condition as the tail feature distance set. The executing entity can sort the feature distances according to their distance size and select feature distances whose distance values ​​are located in a preset quantile interval (e.g., the top 5% or top 10%) as the tail feature distance set to characterize the maximum feature offset range of the radiation source individual under normal operating conditions.

[0064] The third sub-step involves fitting the distribution of the aforementioned tail feature distance set to generate feature distribution location parameters, feature scale parameters, and feature shape parameters. In practice, the executing entity can employ the generalized Pareto distribution model (EVT model) from extreme value theory to estimate the parameters of the tail feature distance set, thereby obtaining the feature distribution location parameters, feature scale parameters, and feature shape parameters describing the tail distribution characteristics. Specifically, the executing entity can determine the aforementioned feature distribution location parameters, feature scale parameters, and feature shape parameters using the following cumulative distribution function expression: F(d) = 1 - exp[-[(d - μ) / λ] k The aforementioned feature distribution location parameter, feature scale parameter, and feature shape parameter are used to characterize the normal fluctuation boundary of the known radiation source individual in the feature space. The aforementioned d can be a tail feature distance from the aforementioned set of tail feature distances. The aforementioned feature distribution location parameter μ can characterize the degree of center shift of the feature distribution. The aforementioned feature scale parameter λ can characterize the degree of dispersion of the feature distribution. The aforementioned feature shape parameter k can characterize the degree of deviation of the tail features.

[0065] The fourth sub-step involves determining the aforementioned characteristic distribution location parameters, characteristic scale parameters, and characteristic shape parameters as known radiation source boundary parameters.

[0066] The second step is to determine the generated boundary parameters of each known radiation source as a set of known radiation source boundary parameters.

[0067] Step 106: Based on the known set of radiation source boundary parameters and the radar radiation source individual identification model, perform individual radar source identification on the real-time radar signals received by each receiving device in the airport area to generate radar radiation source identification results.

[0068] In some embodiments, the execution entity can perform individual radar source identification on the real-time radar signals received by each receiving device within the airport area based on the known set of radiation source boundary parameters and the radar radiation source individual identification model, thereby generating a radar radiation source identification result. In practice, firstly, the execution entity can input the real-time radar signal into the radar radiation source individual identification model to obtain the individual identification probability and real-time radar signal features. Then, the execution entity can select radiation source boundary parameters from the known set of radiation source boundary parameters that match the target known radiation source individual represented by the individual identification probability, and verify the credibility of the real-time radar signal features using the selected radiation source boundary parameters (e.g., inputting the real-time radar signal features into a one-Class SVM model loaded with the corresponding radiation source boundary parameters), obtain the credibility verification result, and use the credibility verification result and the individual identification probability as the radar radiation source identification result.

[0069] In some optional implementations of certain embodiments, the aforementioned execution entity may perform individual radar source identification on the real-time radar signals received by each receiving device within the aforementioned airport area based on the aforementioned known set of radiation source boundary parameters and the aforementioned individual radar source identification model, thereby generating radar source identification results: The first step is to segment the aforementioned real-time radar signals to obtain a set of real-time radar signal segments. In practice, the executing entity can divide each real-time radar signal into individual real-time radar signal segments according to a fixed number of signal frames, thus obtaining a set of real-time radar signal segments.

[0070] The second step involves extracting features from the set of real-time radar signal segments using the radiation source feature extraction network included in the radar radiation source individual identification model, in order to generate features for each real-time radar signal segment.

[0071] The third step is to generate the mean feature of the real-time radar signal based on the characteristics of each of the aforementioned real-time radar signal segments. In practice, the executing entity can average the characteristics of the aforementioned real-time radar signal segments to generate the mean feature of the real-time radar signal.

[0072] The fourth step is to input the aforementioned real-time radar signal mean features into the classification output head of the radar radiation source individual identification model to obtain the individual identification probability.

[0073] Fifth, based on the aforementioned individual identification probabilities, determine the target known radiation source center features and target known radiation source boundary parameters from the aforementioned set of known radiation source center features and the aforementioned set of known radiation source boundary parameters, respectively. In practice, the executing entity can select known radiation source center features and known radiation source boundary parameters that match the target known radiation source individual as represented by the aforementioned individual identification probabilities, from the aforementioned set of known radiation source center features and the aforementioned set of known radiation source boundary parameters, respectively, as the target known radiation source center features and target known radiation source boundary parameters.

[0074] The sixth step is to generate the feature distance between the known radiation source center feature of the target and the mean feature of the real-time radar signal as the feature distance to be verified. In practice, the feature distance can be the Euclidean distance between the two features.

[0075] Step 7: Based on the known boundary parameters of the target radiation source, verify the credibility of the aforementioned feature distance to be verified, and obtain the credibility verification result. In practice, the executing entity can determine the probability value or confidence value of the feature distance to be verified under the tail distribution model (Weibull distribution model) characterized by the known boundary parameters of the target radiation source as the credibility verification result, thereby determining whether the feature distance falls within the preset credibility range.

[0076] The eighth step is to determine the above individual identification probability and credibility verification results as the radar radiation source identification results.

[0077] The above-described embodiments of this disclosure have the following beneficial effects: the radar radiation source individual identification method for airport low-altitude security through some embodiments of this disclosure can improve the generalization ability of the identification model under small sample conditions, thereby improving the accuracy of radiation source individual identification and improving airport low-altitude security. Specifically, the reasons for the low accuracy of radiation source individual identification and the reduced airport low-altitude security are as follows: In actual application scenarios, airport bird deterrence radar and UAV-mounted radar may have differences in manufacturers, batches, or firmware versions, and new equipment of the same model will appear with maintenance and replacement. In addition, due to limitations such as security control, data collection window, and equipment operating time, airports cannot collect a large amount of labeled data for each radar device for a long period of time, resulting in less usable sample data for each radar device, which leads to poor generalization ability of the identification model and low accuracy of radiation source individual identification, making it difficult to effectively identify and determine unknown radiation sources, thus leading to a decline in airport low-altitude security. Based on this, the radar radiation source individual identification method for airport low-altitude security through some embodiments of this disclosure first generates a set of radar signal segments based on the collected original radar signals. The aforementioned raw radar signals are unmarked radar signals received by different receiving devices located in the airport area, including receiving stations and airborne receivers from patrol drones. Therefore, by dividing the raw radar signals continuously received by different receiving devices within the airport area into radar signal segments, and grouping radar signal segments belonging to the same radiation source based on radar signal correlation features, a set of radar signal segment groups is obtained. Then, based on the aforementioned set of radar signal segment groups, the radiation source feature extraction network undergoes self-supervised pre-training to update its model parameters. Thus, without manual annotation, the radiation source feature extraction network learns to maintain similar stable features for each type of signal from the same source under different noise and reception conditions, thereby improving its adaptability to noise, obstruction, and time-varying channels in the actual airport environment. Subsequently, based on the aforementioned radiation source feature extraction network, a radar radiation source individual recognition model is constructed. Therefore, the pre-trained radiation source feature extraction network with stable feature extraction capabilities can be concatenated with a classification output head to obtain a radar radiation source individual recognition model that can convert signal segments into signal segment features and output discriminable categories. Next, the radar source individual identification model is jointly optimized using the known set of radiation source signals to generate a set of known radiation source center features. This set of known radiation source signals is a labeled small-sample dataset collected from the aforementioned airport area.Therefore, under the hard constraint of a small sample dataset, joint loss is used to make the features of each known radar source more clustered under different acquisition periods and different receiving devices, forming stable and more representative central features of known sources. This also exacerbates the differences in central features between different individual radar sources, reducing the probability of misidentification of radar sources of the same model but different batches or firmware versions. Secondly, based on the aforementioned radar source individual identification model and the aforementioned set of known source central features, a set of known source boundary parameters is generated. Each known source boundary parameter corresponds to a known radar source individual within the aforementioned airport area. This allows the establishment of boundary parameters representing the normal range of variation for each known radar source individual. This enables the same device to experience certain feature drift due to fluctuations in the airport environment and operating conditions, while also determining the maximum boundary of feature drift. This significantly reduces the probability of misidentification when encountering new equipment or abnormal features caused by strong interference. Finally, based on the aforementioned set of known source boundary parameters and the aforementioned radar source individual identification model, radar source individual identification is performed on the real-time radar signals received by each receiving device within the aforementioned airport area to generate radar source identification results. Therefore, by averaging the features of multiple segments to suppress instantaneous disturbances, and then combining the boundary parameters to verify the credibility of the classification results, the accuracy of known device identification is improved, while the risk of misidentifying new devices or abnormal conditions as known devices is reduced, thereby improving the reliability and security of airport low-altitude security.

[0078] Further reference Figure 5 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a radar radiation source individual identification device for low-altitude security at airports. These device embodiments are similar to... Figure 1 Corresponding to the method embodiments shown, this radar radiation source individual identification device for low-altitude airport security can be specifically applied to various electronic devices.

[0079] like Figure 5As shown, a radar radiation source individual identification device 500 for low-altitude security at airports, according to some embodiments, includes: a first generation unit 501, a self-supervised pre-training unit 502, a construction unit 503, a second generation unit 505, and an individual identification unit 506. The first generation unit 501 is configured to generate a set of radar signal segments based on the collected original radar signals. These original radar signals are unmarked radar signals received by different receiving devices located in the airport area, including receiving stations and airborne receivers from patrol drones. The self-supervised pre-training unit 502 is configured to perform self-supervised pre-training on a radiation source feature extraction network based on the radar signal segment set to update the network's model parameters. The construction unit 503 is configured to construct a radar radiation source individual identification model based on the radiation source feature extraction network. The joint optimization unit 504 is configured to perform radar radiation source individual identification using a known set of radiation source signals. The model is jointly optimized to generate a set of known radiation source center features, wherein the set of known radiation source signal groups is a labeled small sample dataset collected in the airport area; the second generation unit 505 is configured to generate a set of known radiation source boundary parameters based on the radar radiation source individual identification model and the set of known radiation source center features, wherein each known radiation source boundary parameter corresponds to a known radar radiation source individual in the airport area; the individual identification unit 506 is configured to perform radiation source individual identification on the real-time radar signals received by each receiving device in the airport area based on the set of known radiation source boundary parameters and the radar radiation source individual identification model, so as to generate radar radiation source identification results.

[0080] It is understandable that the units described in the radar radiation source individual identification device 500 for low-altitude security at airports are related to the reference... Figure 1 The steps in the described method correspond accordingly. Therefore, the operations, features, and beneficial effects described above for the method are also applicable to the radar radiation source individual identification device 500 and its constituent units for low-altitude airport security, and will not be repeated here.

[0081] The following is for reference. Figure 6 It shows a schematic diagram of the structure of an electronic device 600 (e.g., a computing device) suitable for implementing some embodiments of the present disclosure. Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0082] like Figure 6As shown, electronic device 600 may include processing device 601 (e.g., central processing unit, graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 602 or a program loaded from storage device 608 into random access memory (RAM) 603. RAM 603 also stores various programs and data required for the operation of electronic device 600. Processing device 601, ROM 602, and RAM 603 are interconnected via bus 604. Input / output (I / O) interface 605 is also connected to bus 604.

[0083] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 608 including, for example, magnetic tapes, hard disks, etc.; and communication devices 609. Communication device 609 allows electronic device 600 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6 An electronic device 600 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 6 Each box shown can represent a device or multiple devices as needed.

[0084] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure 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 a communication device 609, or installed from a storage device 608, or installed from a ROM 602. When the computer program is executed by the processing device 601, it performs the functions defined above in the methods of some embodiments of this disclosure.

[0085] It should be noted that, in some embodiments of this disclosure, the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may 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 or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may 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 some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can 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: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0086] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0087] The aforementioned computer-readable medium may be included within the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: generate a set of radar signal segments based on the collected raw radar signals, wherein the raw radar signals are unmarked radar signals received by different receiving devices located in the airport area, including receiving stations and airborne receivers of patrol drones; perform self-supervised pre-training on a radiation source feature extraction network based on the aforementioned set of radar signal segments to update the model parameters of the radiation source feature extraction network; construct an individual radar radiation source identification model based on the aforementioned radiation source feature extraction network; and, through the known set of radiation source signals, ... The aforementioned radar source individual identification model is jointly optimized to generate a set of known radiation source center features, wherein the set of known radiation source signal groups is a labeled small sample dataset collected in the aforementioned airport area. Based on the aforementioned radar source individual identification model and the aforementioned set of known radiation source center features, a set of known radiation source boundary parameters is generated, wherein each known radiation source boundary parameter corresponds to a known radar source individual within the aforementioned airport area. Based on the aforementioned set of known radiation source boundary parameters and the aforementioned radar source individual identification model, the real-time radar signals received by each receiving device within the aforementioned airport area are used to identify the individual radiation source, thereby generating radar radiation source identification results.

[0088] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0089] 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 disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing 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 the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, 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.

[0090] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0091] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for identifying individual radar radiation sources for low-altitude security at airports, characterized in that, include: Based on the collected raw radar signals, a set of radar signal fragments is generated. The raw radar signals are unmarked radar signals received by different receiving devices located in the airport area. The receiving devices include receiving stations and airborne receivers of patrol drones. Based on the radar signal segment set, the radiation source feature extraction network is self-supervised pre-trained to update the model parameters of the radiation source feature extraction network. Based on the radiation source feature extraction network, a radar radiation source individual identification model is constructed; By using a known set of radiation source signal groups, the radar radiation source individual identification model is jointly optimized to generate a set of known radiation source center features, wherein the known set of radiation source signal groups is a labeled small sample dataset collected in the airport area; Based on the radar radiation source individual identification model and the known radiation source center feature set, a known radiation source boundary parameter set is generated, wherein each known radiation source boundary parameter corresponds to a known radar radiation source individual within the airport area; Based on the known set of radiation source boundary parameters and the radar radiation source individual identification model, the radar radiation source individual identification is performed on the real-time radar signals received by each receiving device within the airport area to generate radar radiation source identification results.

2. The method according to claim 1, characterized in that, The step of generating a set of radar signal fragments based on the collected raw radar signals includes: For each raw radar signal acquired, the following time-domain processing steps are performed: The original radar signal is subjected to time-frequency transformation processing to generate signal time-frequency energy information; The time-frequency energy information of the signal is used to perform noise estimation to generate a signal noise basis and a signal noise estimation scale; Based on the signal noise basis and the signal noise estimation scale, regional features are extracted from the time-frequency energy information of the signal to obtain a time-frequency regional feature set. Based on the time-frequency region feature set, the original radar signal is divided into time-domain segments to obtain various radar signal segments; Feature extraction is performed on each radar signal segment to generate radar signal segment features; Based on the characteristics of each generated radar signal segment, the radar signal segments are classified into groups of signals from the same source to obtain a set of radar signal segment groups.

3. The method according to claim 1, characterized in that, The step of performing self-supervised pre-training on the radiation source feature extraction network based on the radar signal segment set to update the model parameters of the radiation source feature extraction network includes: For each radar signal segment group in the aforementioned radar signal segment group set, the following model training steps are performed: The radar signal segment group is subjected to sample signal enhancement processing to generate a set of signal segment sample pairs; Based on the set of signal segment sample pairs, the following self-supervised training steps are performed: A predetermined number of signal segment sample pairs are extracted from the set of signal segment sample pairs and input into the radiation source feature extraction network to determine the batch signal loss value; In response to the determination that the radiation source feature extraction network has not converged, the parameters of the radiation source feature extraction network are updated according to the batch signal loss value, and the self-supervised training step is executed again.

4. The method according to claim 3, characterized in that, The step of performing sample signal enhancement processing on the radar signal segment group to generate a set of signal segment sample pairs includes: For each radar signal segment in the radar signal segment group, the following sample enhancement steps are performed: The radar signal segment is subjected to noise injection processing to generate a first enhanced signal segment; The radar signal segment is time-shifted to generate a second enhanced signal segment; The first enhanced signal segment and the second enhanced signal segment are determined as a signal segment sample pair; The generated signal segment sample pairs are defined as a signal segment sample pair set.

5. The method according to claim 1, characterized in that, The step of generating a set of known radiation source boundary parameters based on the radar radiation source individual identification model and the known radiation source center feature set includes: For each known radiation source center feature in the set of known radiation source center features, perform the following steps: The spatial distance between each known radiation source feature and the center feature of the known radiation source is determined as the feature distance; Select at least one feature distance that satisfies the spacing condition from the various feature distances as the tail feature distance set; The tail feature distance set is fitted with a distribution to generate feature distribution location parameters, feature scale parameters, and feature shape parameters; The characteristic distribution location parameter, the characteristic scale parameter, and the characteristic shape parameter are determined as the boundary parameters of a known radiation source. The generated boundary parameters of each known radiation source are defined as the set of known radiation source boundary parameters.

6. The method according to claim 1, characterized in that, The step of identifying individual radar sources for each receiving device within the airport area based on the known set of radiation source boundary parameters and the radar radiation source individual identification model, to generate radar radiation source identification results, includes: The real-time radar signal is segmented to obtain a set of real-time radar signal segments; The radar radiation source individual identification model includes a radiation source feature extraction network, which is used to extract features from the set of real-time radar signal segments to generate features for each real-time radar signal segment. Based on the characteristics of each real-time radar signal segment, generate the mean characteristics of the real-time radar signal; The mean features of the real-time radar signal are input into the classification output head of the individual radar radiation source identification model to obtain the individual identification probability. Based on the individual identification probability, the target known radiation source center features and target known radiation source boundary parameters are determined respectively from the known radiation source center feature set and the known radiation source boundary parameter set; The feature distance between the known radiation source center features of the target and the mean features of the real-time radar signal is generated as the feature distance to be verified. Based on the known radiation source boundary parameters of the target, the credibility of the feature distance to be verified is verified to obtain the credibility verification result; The individual identification probability and credibility verification results are determined as the radar radiation source identification results.

7. A radar radiation source individual identification device for low-altitude security at airports, characterized in that, include: The first generation unit is configured to generate a set of radar signal fragments based on the collected raw radar signals, wherein the raw radar signals are unmarked radar signals received by different receiving devices located in the airport area, including receiving stations and airborne receivers of patrol drones. The self-supervised pre-training unit is configured to perform self-supervised pre-training on the radiation source feature extraction network based on the radar signal segment set, so as to update the model parameters of the radiation source feature extraction network. The building unit is configured to build an individual radar radiation source identification model based on the radiation source feature extraction network; The joint optimization unit is configured to jointly optimize the radar radiation source individual identification model using a known set of radiation source signal groups to generate a set of known radiation source center features, wherein the known set of radiation source signal groups is a labeled small sample dataset collected in the airport area; The second generation unit is configured to generate a set of known radiation source boundary parameters based on the radar radiation source individual identification model and the known radiation source center feature set, wherein each known radiation source boundary parameter corresponds to a known radar radiation source individual within the airport area; The individual identification unit is configured to perform individual radar source identification on the real-time radar signals received by each receiving device within the airport area based on the known set of radiation source boundary parameters and the radar radiation source individual identification model, so as to generate radar radiation source identification results.

8. An electronic device, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 6.

9. A computer-readable medium, characterized in that, It stores a computer program thereon, wherein the computer program, when executed by a processor, implements the method as described in any one of claims 1 to 6.

Images