Pulse doppler radar target detection method and system based on background contrast attention mechanism

Abstract

The application discloses a pulse Doppler radar target detection method and system based on a background contrast attention mechanism, and relates to the field of radar target detection. The method comprises the following steps: data preprocessing, generating a range Doppler image; constructing a neural network based on a background contrast attention mechanism; training the neural network to obtain a trained neural network; and obtaining a detection result of a pulse Doppler image to be detected by using the trained neural network. The application focuses on learning the clutter distribution characteristics of the background in the range Doppler image and suppressing the background features and improving the saliency of the target features from the feature level based on the background contrast attention mechanism. A multi-scale background learner module based on random masking, an attention mechanism module based on background contrast and the introduction of JS divergence into a loss function are designed in a targeted manner, so that the background interference can be better suppressed from the feature level, and the detection performance in a low signal-to-clutter ratio scene can be improved.

Description

Technical Field

[0001] This invention belongs to the field of signal processing and target detection, specifically relating to a pulse Doppler radar target detection method and system based on background contrast attention mechanism. Background Technology

[0002] Radar is one of the most effective tools for detecting unmanned aerial vehicle (UAV) targets. Radar-based target detection refers to the detection and localization of targets by preprocessing and using target detection algorithms based on the echo signals received by the radar, thereby obtaining relevant target information. With its advantage of operating in all weather conditions and around the clock, radar can capture target distance and speed information, providing an effective means for target detection and tracking.

[0003] Current radar target detection methods include statistical model-based methods and feature-based methods. Statistical model-based methods analyze the presence of a target in the signal by making assumptions about clutter and the target, with constant false alarm rate (CFAR) algorithms being a representative example. However, these methods require prior assumptions about background clutter, leading to a sharp performance drop in performance in today's complex and dynamic environments. Feature-based methods distinguish targets by utilizing the differences in features between the target and the background, including fractal features, chaotic features, time-frequency features, and micro-Doppler features. Deep learning, through training, can automatically extract latent features of the target and detect it, exhibiting strong robustness and generalization ability. Therefore, deep neural networks are increasingly being used for radar target detection. The input to a neural network can be the original radar echo signal, the signal's time-frequency plot, or the signal's range-Doppler plot, with the range-Doppler plot being a common data format. Current range-Doppler plot-based deep learning algorithms often directly extract target feature information during target detection, neglecting to learn about the background. This leads to an increased false alarm rate in low-clutter scenarios, resulting in decreased detection performance.

[0004] Radar target detection algorithms model radar echo signals and detect targets. However, traditional algorithms have poor robustness, and current deep learning-based algorithms are significantly affected in low signal-to-clutter scenarios. Summary of the Invention

[0005] To address the problems in the prior art, this invention proposes a pulse Doppler radar target detection method and device system based on a background contrast attention mechanism.

[0006] The technical solution adopted in this invention is as follows:

[0007] In a first aspect, the present invention discloses a pulse Doppler radar target detection method based on a background contrast attention mechanism, comprising the following steps:

[0008] Step 1): Obtain the pulse Doppler image of the radar echo signal;

[0009] Step 2): Construct a neural network based on a background contrast attention mechanism. The neural network is used to obtain distance and velocity information of targets in the Doppler image. The neural network includes a multi-scale background learner module based on random mask, a Transformer encoder module based on a background contrast attention mechanism, a Transformer decoder, and a feedforward network module.

[0010] Step 3): Train the neural network based on the background contrast attention mechanism to obtain a trained neural network;

[0011] Step 4) Use the trained neural network to obtain the target category information and location information in the range Doppler image to be detected, so as to realize the detection of radar targets.

[0012] Secondly, the present invention discloses a pulse Doppler radar target detection system based on a background contrast attention mechanism according to the method described above, comprising:

[0013] The signal preprocessing module is used to generate pulse Doppler images corresponding to radar echoes;

[0014] A neural network construction module is used to construct a neural network based on a background contrast attention mechanism. The neural network is used to acquire the category information and location information of the target in the range Doppler image. The neural network includes a multi-scale background learner module based on random mask, a Transformer encoder module based on a background contrast attention mechanism, a Transformer decoder module, and a feedforward network module.

[0015] The neural network training module is used to train a neural network based on a background contrast attention mechanism to obtain a trained neural network.

[0016] The range Doppler image detection module is used to obtain target category and location information in the range Doppler image to be detected using a trained neural network, thereby realizing target detection in the range Doppler image.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0018] 1) This invention proposes a DETR detection network (DetectionTransformer detection network, an end-to-end target detection network based on the Transformer architecture) for range Doppler images in the radar field. It is the first to introduce the DETR framework into the radar target detection domain, and during target detection, it suppresses background interference in the feature dimension by learning background features, rather than directly extracting target information. This method effectively solves the problem of high false alarm rate in low signal-to-clutter scenarios.

[0019] 2) This invention proposes a multi-scale background learning module based on random masks, which uses random masks to eliminate the influence of the target on the background when learning background features, and learns to fuse multi-scale information to effectively obtain background features.

[0020] 3) This invention proposes an attention mechanism module based on background contrast and integrates JS divergence into the loss function to measure the degree of difference between target features and background features. This module can suppress background interference and extract purer target information in the feature dimension. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the basic steps of an embodiment of the radar target detection method of the present invention.

[0022] Figure 2 This is a schematic diagram of the background contrast-based attention mechanism module of the present invention;

[0023] Figure 3 This is a schematic diagram of the radar target detection device of the present invention;

[0024] Figure 4 This is a partial structural diagram of the Transformer decoder and feedforward network module of the present invention;

[0025] Figure 5 The images show a simulated distance Doppler image dataset used in the experiment and the detection results after detection using the embodiments of the present invention and different methods. Detailed Implementation

[0026] The present invention will be further described and illustrated below with reference to specific embodiments. The embodiments described are merely examples of the content of this disclosure and do not limit the scope of the invention. The technical features of each embodiment in the present invention can be combined accordingly, provided that there is no mutual conflict.

[0027] The object of this invention is the down-converted radar echo signal. As a well-known technology in the field, down-conversion is the process of converting a high-frequency signal into a lower-frequency signal. By reducing the frequency, retaining bandwidth and phase information, the purpose is to facilitate subsequent processing and analysis.

[0028] like Figure 1 The diagram shown is a flowchart illustrating the basic steps of the pulse Doppler radar target detection method based on background contrast attention mechanism of the present invention. The method of the present invention mainly includes:

[0029] Step 1): Signal preprocessing

[0030] For the down-converted radar echo signal to be processed, this embodiment employs pulse compression and Doppler processing methods to obtain the target's range Doppler image. Let S... r1 (t) represents the one-dimensional echo sequence of the radar, where t represents time. The formula for pulse compression is as follows:

[0031] S rp (t)=S r1 (t)*h(t)

[0032] Where h(t) is the signal obtained by conjugating the radar transmitted signal after deconvolution, and S rp (t) represents the pulse-compressed signal; for multiple pulse echoes, time t is written as Where t m The slow time is represented by m, the number of received pulse echoes is represented by T, and the pulse repetition interval is represented by T. Describing fast time as a time series within a pulse repetition interval, then the one-dimensional target echo S... rp (t) represents the two-dimensional target echo.

[0033] Along the fast time dimension Perform a Fast Fourier Transform to obtain the range Doppler image matrix of the target echo. The calculation formula is as follows:

[0034]

[0035] Where N represents the number of sampling points. In this embodiment, N = 256, and k represents the frequency; the distance Doppler image matrix... As distance Doppler image I RD .

[0036] Step 2): Construct a neural network based on a background contrast attention mechanism. This neural network includes a multi-scale background learner module based on random masks, a Transformer encoder module based on the background contrast attention mechanism, a Transformer decoder module, and a feedforward network module. The multi-scale background learner module based on random masks learns the distribution characteristics of background clutter in the range-Doppler image. The Transformer encoder module extracts feature information of the target from the range-Doppler image and suppresses background interference along the feature dimension. The Transformer decoder and feedforward network module parse the information provided by the encoder and output the target's category and location information. This embodiment uses random parameters as the initial weights of the network.

[0037] The specific working process of each module within the neural network based on the background contrast attention mechanism is as follows:

[0038] Step 21): The multi-scale background learner module based on random masking aims to learn the distribution characteristics of background clutter, which is used to subsequently suppress background information in the target from the feature dimension. Its structure is shown in [link to module]. Figure 1 For the input distance Doppler image I RD First, the range-Doppler image is converted into range-Doppler image patches through slicing to facilitate random masking. The random mask is then used to minimize the contamination of the background by the target during background learning by randomly designing several blocks from the range-Doppler image patches. The calculation process is as follows:

[0039] I patch =Slice(I RD )

[0040] T patch =Random_Mask(I patch )

[0041] Among them, I patch Represents the distance Doppler image; Slice() represents the slicing function, Random_Mask() represents the random masking function, T patch This represents the sample obtained after processing.

[0042] Then, T is processed using three convolutional kernels of different sizes. patch Feature extraction and fusion are performed to achieve multi-scale background feature learning. The convolution used here is depthwise separable convolution, which, compared to ordinary convolution, reduces the number of parameters and computational cost while maintaining performance. Figure 1 In Chinese, it is represented by "DWConv", and the calculation process is as follows:

[0043] T1=DWConv77(T paach)

[0044] T2=DWConv55(T patch )

[0045] T3=DWConv33(T patch )

[0046] T concat =Concat(T1,T2,T3)

[0047] T temp =DWConvBlock(T concat )

[0048] T B =LayerNorm(AvgPool(T temp ))

[0049] Where DWConv77, DWConv55, and DWConv33 represent depthwise separable convolution functions with kernel sizes of 7×7, 5×5, and 3×3, respectively; Concat() represents the concatenation function; DWConvBlock() is a feature extraction function composed of three depthwise separable convolution functions with kernel size of 3×3, and T1, T2, T3, T... concat T temp The intermediate results obtained from the corresponding operations; AvgPool() represents the average pooling function, LayerNorm() represents the layer normalization function, and the final features are obtained from the multi-scale background learner module based on random masks.

[0050] Step 22): The transformer encoder module based on the background contrast attention mechanism consists of a convolutional neural network, a positional encoder, and a transformer encoder. Its structure is shown in the figure below. Figure 1 In this embodiment, the convolutional neural network uses the ResNet-50 structure, therefore, for the input distance Doppler image I... RD After passing through CNN and positional encoding, a feature map F with positional encoding is obtained. RD The calculation formula is as follows:

[0051] F RD =ResNet(I RD Positional Encoding

[0052] in, c represents the number of channels, h represents the height, w represents the width, ResNet() represents a deep residual network, and Positional Encoding is the positional encoding.

[0053] Then, F RD The data is flattened and passed through an embedding layer to obtain the feature T, which is then input to the self-attention module. RD1 The calculation formula is as follows:

[0054] T RD1 =Embedding(Flatten(F RD ))

[0055] Here, Flatten() represents the flattening operation, and Embedding() represents the embedding layer operation. d model This indicates the number of channels in the embedding layer.

[0056] The Transformer encoder module based on background contrast attention mechanism includes multiple cascaded Transformer encoder blocks. Each Transformer encoder block includes a multi-head self-attention module, a background contrast attention module, and a multilayer perceptron module.

[0057] Transformer encoder block pair T RD1 The process of performing multi-head self-attention mechanism computation, residual connection, and normalization operations can be represented as follows:

[0058] Q = T RD1 W q K = T RD1 W k V = T RD1 W v

[0059] head1,…,head h =Split([Q,K,V])

[0060]

[0061] T = Concat(head') i W0

[0062] T RD2 =LayerNorm(T+T) RD1 )

[0063] Among them, W q W k W v These represent the projection matrices of the query matrix Q, the key matrix K, and the value matrix V, respectively; head i The feature vector T represents the distance to the Doppler image obtained after initial feature extraction. RDThe i-th attention head after splitting; in this embodiment, there are a total of h attention heads; Split([Q,K,V]) is the attention head splitting function; Softmax is the activation function, which can transform the input into weights that sum to 1 and weight them onto the value matrix V; head′ i Q represents the i-th attention head after attention weighting; i Let be the query matrix for the i-th attention head; V is the transpose of the key matrix of the i-th attention head; i Let `head` be the value matrix of the i-th attention head; `Concat(head')` i ) represents the attention head concatenation function; T represents the intermediate result; W0 is a mapping matrix that makes the output T RD2 Convert to input T RD1 Same dimensions; LayerNorm() is the layer normalization function.

[0064] Then, T RD2 T and the output of the multi-scale background learning module based on random masks B Both are input into the background contrast-based attention mechanism (BCA) module, which suppresses the background from the feature dimension by fusing the feature vectors of the target and the background. RD2 First, the target feature tensor T is obtained through reshaping and average pooling operations for subsequent calculations. T2 The calculation formula is as follows:

[0065] T T1 =Reshape(T) RD2 )

[0066] T T2 =AvgPool(T T1 )

[0067] Reshape() is the reshaping function used to change the shape of the tensor; AvgPool() is the average pooling function. T representing the background B After a 1×1 convolution, we obtain the result with T. T2 For channels with the same dimension, the same space size is obtained after a copy operation. The calculation formula is as follows:

[0068] T B1 =Conv(T B )

[0069] T B2 =Repeat(T) B1 )

[0070] Where Conv() is the convolution function; Repeat() is the copy function; features feature

[0071] The processed T T2 and T B2 By splicing, we can obtain T. c The target and background features are fused using 1×1 convolution and pixel-dimensional max pooling to obtain T′. c The aim is to suppress background at the feature level, and the calculation formula is as follows:

[0072] T c =Concat(T T2 ,T B2 )

[0073] T′ c =Conv(T c )+MaxPool(T c )

[0074] Then, for T′ c Upsample to T T1 Using the same spatial size and passing through a sigmoid activation function, an attention weight map with strong target saliency is generated, which is then compared with T. T1 After performing a dot product, reshaping, and then residual join and normalization operations, the calculation formula is as follows:

[0075] T f1 =Reshape(σ(UpSample(T′)) c ))·T T1 )

[0076] T f2 =LayerNorm(T f +T RD2 )

[0077] Where Reshape() is the reshaping function, UpSample() is the upsampling function; σ() represents the sigmoid activation function, LayerNorm() represents the layer normalization function, and T f1 T f2 Features.

[0078] Finally T f2 After mapping using a multilayer perceptron, the calculation formula is as follows:

[0079] T M1 =MLP(T f2 )

[0080] T M2 =LayerNorm(T M1 +T f2 )

[0081] The output characteristic T of the upper-level cascaded Transformer encoder block M2 The input is fed into the next-level cascaded Transformer encoder block. After processing by m cascaded Transformer encoder blocks, the Transformer encoder module based on the background contrast attention mechanism finally outputs the feature V. encoder In this embodiment, m = 4.

[0082] Step 23): The Transformer decoder and feedforward network module aim to parse the target feature information output by the encoder module, progressively transforming target queries into the final output target category and location information. The Transformer decoder structure is as follows: Figure 4 As shown, the Transformer decoder consists of multiple serial Transformer decoder blocks;

[0083] The processing of each Transformer decoder block includes: taking a single target query as an example, it first initializes Q0 through position encoding, and then performs multi-head self-attention mechanism calculation, residual connection and normalization operation on Q0 to obtain Q2. The process is the same as the multi-head self-attention mechanism in step 22), and will not be described here.

[0084] Then, using the encoder output V encoder Q3 is obtained by performing a multi-head attention mechanism with Q2, and then Q4 is obtained through residual connection and normalization operations. The process is as follows:

[0085] Q = Q²W q K = (V encoder +Positional Encoding)W k V = V encoder W v

[0086] head1,…,head h =Split([Q,K,V])

[0087]

[0088] Q3 = Concat(head') i W0

[0089] Q4 = LayerNorm(Q3 + Q2)

[0090] Then, the mapping is performed using a multilayer perceptron, and the calculation formula is as follows:

[0091] Q5 = MLP(Q4)

[0092] Q6 = LayerNorm(Q5 + Q4)

[0093] The output Q6 of the upper-level Transformer decoder block is input into the lower-level Transformer decoder block, and the Transformer decoder finally outputs the feature vector V. decoder In this embodiment, the Transformer decoder includes four serial Transformer decoder blocks.

[0094] Then, the feature vector output by the Transformer decoder is finally passed through two feedforward networks to output target category information, namely a classification head and a regression head, calculated as follows:

[0095] V class =MLP_Class(V decoder )

[0096] V reg =MLP_Regression(V decoder )

[0097] Where MLP_Class() represents the classification head function, and MLP_Regression() represents the regression head function; V class This represents the category confidence score. In this example, since there are only two categories, target and background, therefore... V reg This represents the position coordinates; in this example,

[0098] Step 3): Training the neural network model

[0099] Using the entire distance Doppler image as training samples, the network employs supervised training. Samples are input into the neural network model. To ensure effective background suppression in newly generated target features, JS divergence is introduced to measure the similarity between background features and newly generated target features. This divergence is then inverted and added to the loss function. Ultimately, this invention uses the weighted sum of the Hungarian loss of the target detection bounding box and the JS divergence loss as the total loss function. It is expressed as follows:

[0100]

[0101] in, Indicates Hungary's loss. Let denot be the JS divergence loss, and λ represent the weight coefficient. In this embodiment, λ = 1.0. The Hungarian loss is a well-known loss function in the art, which finds the optimal match between the predicted bounding box and the ground truth bounding box using the Hungarian algorithm, minimizing the total matching cost. The matching cost is usually based on some distance metric between the predicted bounding box and the ground truth bounding box (such as IoU or L2 distance). For more information on the Hungarian loss, see: Carion, N., Massa, F., Synnaeve, G., Usunier, N., Kirillov, A., Zagoruyko, S. (2020). End-to-End Object Detection with Transformers. In: Vedaldi, A., Bischof, H., Brox, T., Frahm, JM. (eds) Computer Vision–ECCV 2020. ECCV 2020. Lecture Notes in Computer Science(), vol 12346.

[0102] The formula for calculating the JS divergence loss is as follows:

[0103]

[0104] Where C is a constant used to ensure It is a non-negative number, and

[0105]

[0106]

[0107] Where, P(x) = softmax(T) f1 Let T be the probability distribution of the target, and Q(x) = softmax(T) B1 The probability distribution with ) as the background.

[0108] This invention updates network weights using Adam gradient descent with adaptive learning rate adjustment. In this embodiment, the training is iterated 400 times.

[0109] Step 4): For the down-converted radar echo signal to be processed, the range Doppler image is obtained through step 1), and then input into the neural network model trained in step 3). The network outputs whether the target exists in the range Doppler image and the location of the target, thus obtaining the detection result.

[0110] Figure 3 This is a block diagram illustrating a pulse Doppler radar target detection system based on a background contrast attention mechanism, according to an embodiment. Figure 3 As shown, the system includes:

[0111] The signal preprocessing module is used to generate range Doppler images corresponding to radar echoes;

[0112] A neural network construction module is used to construct a neural network based on a background contrast attention mechanism. The neural network is used to acquire the category information and location information of the target in the range Doppler image. The neural network includes a multi-scale background learner module based on random mask, a Transformer encoder module based on a background contrast attention mechanism, a Transformer decoder module, and a feedforward network module.

[0113] The neural network training module uses a weighted sum of the Hungarian loss and JS divergence loss of the target detection box as the loss function to train the neural network based on the background contrast attention mechanism, thus obtaining a trained neural network.

[0114] The range Doppler image detection module is used to obtain target category and location information in the range Doppler image to be detected using a trained neural network, thereby realizing target detection in the range Doppler image.

[0115] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0116] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the description of the method embodiments. The device embodiments described above are merely illustrative. For example, the image preprocessing module can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules can be combined or integrated into another unit. Furthermore, the connection between the modules shown or discussed can be a communication connection through some interfaces, which can be electrical or other forms. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort. The following uses simulated distance Doppler images as an example to illustrate specific implementation methods to demonstrate the technical effects of the present invention. Specific steps in the embodiments will not be repeated.

[0117] Example

[0118] Next, using a simulation dataset as the research object, we will verify the radar target detection method of this invention. To comprehensively compare detection performance and present the detection results from both visual and quantitative perspectives, a redefined detection rate p is adopted. d False alarm rate p f and average minimum distance d min The evaluation is performed using the following formula:

[0119]

[0120] in, It is the set of targets to be detected, i.e., the true values; It is the collection of detection results, i.e., the predicted values; Calculate the Euclidean distance T between two targets in the Doppler image. dis This represents the distance threshold.

[0121] The radar and target parameters for the simulation dataset are set as follows: the transmitted signal is a linear frequency modulated signal with a carrier frequency of 3 GHz, a bandwidth of 5 MHz, a pulse width of 5 μs, a pulse repetition frequency of 2 kHz, and a sampling frequency of 30 MHz; the target distance is between 0.5 km and 5.0 km, and the target speed is between 0 m / s and 40 m / s. Figure 5 The first row contains the range Doppler images and their target ground truth frames obtained from simulations at signal-to-noise ratios of -25dB, -20dB, -15dB, and -10dB.

[0122] Figure 5 The second to fourth rows are the detection results obtained from the simulation dataset using the embodiments of the present invention and different radar target detection algorithms.

[0123] Table 1 Evaluation Indicators of Detection Results from Simulated Radar Dataset

[0124]

[0125]

[0126] The comparative method CFAR comes from: S.Watts, "Cell-averaging CFAR gain in spatiallycorrelated K-distributed clutter", IEE Proceedings-Radar, Sonar and Navigation, vol.143, pp.321–327, 1996.

[0127] The comparison method DCNN is from: L. Wang, J. Tang and Q. Liao, "A Study on Radar Target Detection Based on Deep Neural Networks", in IEEE Sensors Letters, vol. 3, no. 3, pp. 1-4, March 2019.

[0128] The contrastive method DETR is from: Carion, N., Massa, F., Synnaeve, G., Usunier, N., Kirillov, A., Zagoruyko, S. (2020). End-to-End Object Detection with Transformers. In: Vedaldi, A., Bischof, H., Brox, T., Frahm, JM. (eds) Computer Vision–ECCV 2020. ECCV2020. Lecture Notes in Computer Science(), vol 12346.

[0129] The detection results of the simulation dataset are as follows Figure 5 As shown in Table 1, the quantization results demonstrate that this invention achieves the best performance under various signal-to-noise ratio (SNR) conditions. While all methods perform well under high SNR conditions, this method maintains a good detection rate and a low false alarm rate even when the SNR decreases. Furthermore, the d-value of this method... min The fact that there are always the best results indicates that the target was set more accurately.

[0130] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. Those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A pulse Doppler radar target detection method based on background contrast attention mechanism, characterized in that, Includes the following steps: Step 1) Obtain the pulse Doppler image of the radar echo signal; Step 2) Construct a neural network based on a background contrast attention mechanism. This neural network is used to acquire distance and velocity information of the target in the range-Doppler image. The neural network includes a multi-scale background learner module based on random masking, a Transformer encoder module based on the background contrast attention mechanism, a Transformer decoder, and a feedforward network module. The multi-scale background learner module based on random masking is used to acquire background clutter distribution features from the range-Doppler image. The Transformer encoder module based on the background contrast attention mechanism includes multiple cascaded Transformer encoder blocks. Each Transformer encoder block includes a multi-head self-attention module, a background contrast attention module, and a multilayer perceptron module. The Transformer encoder module based on the background contrast attention mechanism is used to encode and learn target features, thereby suppressing background clutter interference at the feature level. Step 3) Train the neural network based on the background contrast attention mechanism to obtain a trained neural network; wherein, the weighted sum of Hungarian loss and JS divergence loss is used as the loss function, and the weights of the neural network are updated based on gradient descent with adaptive learning rate adjustment. Step 4) For the radar echo signal to be processed, the pulse Doppler image of the radar echo signal is obtained in step 1), and the target category information and location information in the range Doppler image are obtained by using the trained neural network to realize the detection of radar targets.

2. The pulse Doppler radar target detection method based on background contrast attention mechanism according to claim 1, characterized in that, Step 1) includes: For radar echo signals, pulse compression and Doppler processing methods are used to obtain the target's range-Doppler image; let This represents the two-dimensional target echo sequence after pulse compression, where Indicates slow time. This indicates the number of received pulse echoes. Representing fast time, that is, the time sequence within one pulse repetition interval; along the fast time dimension... Perform a Fast Fourier Transform to obtain the range Doppler image matrix of the target echo. The calculation formula is as follows: Where N represents the number of sampling points, and k represents the frequency; the distance Doppler image matrix As a distance Doppler image .

3. The pulse Doppler radar target detection method based on background contrast attention mechanism according to claim 1, characterized in that, In step 2), the Transformer decoder and feedforward network module parse the encoder output, convert the initialized target queries into target category information and location information, and output them.

4. The pulse Doppler radar target detection method based on background contrast attention mechanism according to claim 1, characterized in that, A multi-scale background learner module based on random masks extracts background clutter distribution features from range-Doppler images; including: For the input distance Doppler image First, a slicing operation is performed to convert the range Doppler image into range Doppler image blocks. , Then, a random mask is used to randomly design several blocks in the distance Doppler image patch to reduce the contamination of the background by the target during background learning, thus obtaining samples. ; Using three convolution kernels of different sizes Feature extraction and fusion are performed to achieve multi-scale background feature learning; the process is as follows: in These represent depthwise separable convolution functions with kernel sizes of 7×7, 5×5, and 3×3, respectively. Indicates the concatenation function; It is a feature extraction function composed of three depthwise separable convolutions with a kernel size of 3×3. , , , , This refers to the intermediate result obtained from the corresponding operation; This represents the average pooling function. The layer normalization function is represented. The output features are those of the multi-scale background learner module.

5. The pulse Doppler radar target detection method based on background contrast attention mechanism according to claim 1, characterized in that, The Transformer encoder module, based on a background contrast attention mechanism, learns to encode target features, including: For the input distance Doppler image After passing through ResNet-50 and positional encoding, a feature map with positional encoding is obtained. ; Will The features are flattened and passed through an embedding layer to obtain the features input to the Transformer encoder block. ; Features Perform multi-head self-attention mechanism computation, residual connection, and normalization operations to obtain the output features. ; and the output of the multi-scale background learning module based on random masks They are input together into the background contrast-based attention mechanism module. The target feature tensor is first obtained through reshaping and average pooling operations. ; background After a 1×1 convolution, we obtain... Features with the same channel dimension and the same spatial size ; The result after processing and The features are concatenated by using 1×1 convolution and pixel-level max pooling to fuse target and background features, resulting in the desired feature set. ; right Upsampling to With the same spatial size, and after passing through the sigmoid activation function, an attention weight map with strong target saliency is generated, and... After performing dot product, reshaping, and then performing residual join and normalization operations, the features are obtained. ; at last After mapping using a multilayer perceptron, the output features of the Transformer encoder block are obtained. The calculation formula is as follows: After processing by multiple cascaded Transformer encoder blocks, the Transformer encoder module based on the background contrast attention mechanism finally outputs features. .

6. The pulse Doppler radar target detection method based on background contrast attention mechanism according to claim 5, characterized in that, The Transformer decoder and feedforward network module parse the encoder output, convert the initialized target queries into target category information and location information, and output them; wherein, the Transformer decoder includes multiple cascaded Transformer decoder blocks; The processing steps for each Transformer decoder block include: For a single target query, it is initialized using positional encoding. Subsequently The multi-head self-attention mechanism is calculated, residual connections are performed, and normalization operations are performed to obtain the results. ; Using the encoder output and Obtaining multi-head attention mechanism Then, after residual join and normalization operations, the result is obtained. ; After mapping by a multilayer perceptron, we obtain The calculation formula is as follows: Output of the upper Transformer decoder block The input is fed into the next-level Transformer decoder block, and the Transformer decoder finally outputs the feature vector. ; The feature vector output by the Transformer decoder is finally passed through two feedforward networks to output target category information, namely a classification head and a regression head; the calculation formula is as follows: in, This represents the classification header function. Represents the regression head function; Indicates category confidence. Indicates the position coordinates.

7. The pulse Doppler radar target detection method based on background contrast attention mechanism according to claim 5, characterized in that, In step 3), the formula for calculating the JS divergence loss function is as follows: in It is a constant used to ensure It is a non-negative number, and in, The probability distribution of the target. The probability distribution is the background.

8. A pulse Doppler radar target detection system based on a background contrast attention mechanism that implements the method of claim 1, characterized in that, include: The signal preprocessing module is used to generate range Doppler images corresponding to radar echoes; A neural network construction module is used to construct a neural network based on a background contrast attention mechanism. This neural network is used to acquire the category and location information of targets in range-Doppler images. The neural network includes a multi-scale background learner module based on random masks, a Transformer encoder module based on the background contrast attention mechanism, a Transformer decoder, and a feedforward network module. The multi-scale background learner module based on random masks is used to acquire background clutter distribution features from the range-Doppler images. The Transformer encoder module based on the background contrast attention mechanism includes multiple cascaded Transformer encoder blocks. Each Transformer encoder block includes a multi-head self-attention module, a background contrast attention module, and a multilayer perceptron module. The Transformer encoder module based on the background contrast attention mechanism is used to encode and learn target features, thereby suppressing background clutter interference at the feature level. The neural network training module is used to train a neural network based on a background contrast attention mechanism to obtain a trained neural network. The loss function is a weighted sum of Hungarian loss and JS divergence loss, and the weights of the neural network are updated based on gradient descent with adaptive learning rate adjustment. The range Doppler image detection module is used to obtain target category and location information in the range Doppler image to be detected using a trained neural network, thereby realizing target detection in the range Doppler image.

Images