Radar pulse stream panoramic cutting method, device and system and storage medium

Abstract

The invention discloses a radar pulse stream panoramic cutting method, device and system and a storage medium. The method comprises the following steps: S1, quantifying a radar pulse stream; and S2, carrying out panoramic cutting on the quantized radar pulse flow based on a PatchPST network. By adopting the technical scheme of the invention, the radar pulse stream panoramic cutting can be effectively realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal processing, and in particular relates to a radar pulse stream panoramic cutting method and device, a system and a storage medium. Background Art

[0002] Multi-function radar (MFR) can flexibly change the scanning direction of the beam by adjusting the phase and amplitude of each array unit. At the same time, it uses time division multiplexing (TDM) technology to realize the parallel processing of multiple tasks, showing flexible beam pointing capability, diverse functions, large-capacity target processing capability, low detectability and excellent anti-interference performance. When generating complex and changing signals, multi-function radars usually adopt a "divide and conquer" strategy, by gradually disassembling the tasks and converting them into a pulse combination with a specific structure emitted in a specific order. This pulse combination with a specific structure is called a radar word. As a basic component of a multi-function radar signal, the radar word is an important basis for multi-function radar analysis. Therefore, the accurate segmentation and recognition of radar words is a key technology for multi-function radar reconnaissance analysis.

[0003] The accurate segmentation and recognition task of radar characters can essentially be regarded as a panoramic cutting task of pulse streams. For the image cutting task, many scholars have conducted research from multiple aspects: Reference 1 (Jonathan Long, Evan Shelhamer, and Trevor Darrell. Fully convolutional networks for semantic segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 3431–3440, 2015.) designed an encoding-decoding cutting framework, laying the foundation for subsequent research in the field of image cutting. Reference 2 (Hengshuang Zhao, Jianping Shi, Xiaojuan Qi, Xiaogang Wang, and Jiaya Jia. Pyramid scene parsing network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2881–2890, 2017.) proposed a multi-scale pooling layer module to capture semantic information at different levels. Reference 3 (Huikai Wu, Junge Zhang, Kaiqi Huang, Kongming Liang, and Yizhou Yu. Fastfcn: Rethinking dilated convolution in the backbone for semantic segmentation. arXiv preprint arXiv: 1903.11816, 2019.) introduces this technology into the classic FCN framework, using a joint pyramid upsampling network structure to extract higher-dimensional image features. Reference 4 (A simple, efficient yet powerful semantic segmentation framework which unifies transformers with lightweight multilayer perceptron (MLP) decoders, proposed in NeurIPS 2021.) introduces the Transformer module in the cutting task to replace the traditional CNN module.Reference 5 (A simple mask classification model that formulates semantic segmentation as a per-pixel classification task, which is published in NeurIPS2021.) unifies instance cutting and semantic segmentation.

[0004] Most of the current research focuses on the field of image target processing, while there are still challenges in the panoramic cutting of radar pulse streams. Existing research mainly explores two methods for dealing with the needs of radar word cutting. The first method is to use change point detection technology to cut radar words. This method is suitable for the situation where the pulse parameters within the radar word are relatively consistent and there are significant differences between radar words. The second method is to determine the starting point of the radar word by traversing point by point, but this method has high computational complexity and long processing time. As the internal structure of multi-function radar radar words becomes increasingly complex and the demand for rapid reconnaissance and identification becomes increasingly urgent, how to achieve radar word cutting is an urgent problem to be solved. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a radar pulse stream panoramic cutting method and device, system and storage medium, quantize and encode the pulse sequence of the multi-function radar, and use the designed PatchPST network structure to extract features of the packaged pulse group, and through Header mapping, thereby realizing the panoramic cutting task of the pulse stream.

[0006] To achieve the above object, the present invention adopts the following technical solution:

[0007] A radar pulse stream panoramic cutting method, comprising:

[0008] Step S1, quantizing radar pulse stream;

[0009] Step S2: Perform panoramic cutting on the quantized radar pulse stream based on the PatchPST network.

[0010] Preferably, in step S1, TOA quantization coding technology is used to quantize and encode the radar pulse stream.

[0011] Preferably, the PatchPST network includes: a PatchPST network basic module and a pulse stream panoramic segmentation head module; wherein the PatchPST network basic module is used to cut the intercepted pulse stream into non-overlapping blocks, the block sequence is input into the mapping module in sequence, and is input into the Transformer module to extract feature vectors; the pulse stream panoramic segmentation head module is used to convert the feature map into pixel-level classification predictions.

[0012] The present invention also provides a radar pulse stream panoramic cutting device, comprising:

[0013] A quantization module, used to quantize radar pulse stream;

[0014] The cutting module is used to perform panoramic cutting on the quantized radar pulse stream based on the PatchPST network.

[0015] Preferably, the quantization module uses TOA quantization coding technology to quantize and encode the radar pulse stream.

[0016] Preferably, the PatchPST network includes: a PatchPST network basic module and a pulse stream panoramic segmentation head module; wherein the PatchPST network basic module is used to cut the intercepted pulse stream into non-overlapping blocks, the block sequence is input into the mapping module in sequence, and is input into the Transformer module to extract feature vectors; the pulse stream panoramic segmentation head module is used to convert the feature map into pixel-level classification predictions.

[0017] The present invention also provides a radar pulse stream panoramic cutting system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a radar pulse stream panoramic cutting method when executed by the processor.

[0018] The present invention also provides a storage medium, on which a computer program is stored, and the computer program executes the radar pulse stream panoramic cutting method when running.

[0019] Aiming at the task of panoramic cutting of radar pulse stream, the present invention proposes a radar pulse stream panoramic cutting technology based on PatchPST. First, the radar pulse stream is quantized, and the radar word labels corresponding to the quantization points are marked. Then, aiming at the panoramic cutting problem of the quantized pulse sequence, the PatchPST network structure is designed to realize the feature extraction of the quantization points. Finally, the radar pulse stream panoramic cutting technology based on PatchPST is adopted to realize the panoramic cutting of the pulse stream. The present invention can effectively realize the panoramic cutting of radar pulse stream. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0021] Figure 1This is a flow chart of a radar pulse stream panoramic cutting method according to an embodiment of the present invention;

[0022] Figure 2 Schematic diagram of TOA quantization coding;

[0023] Figure 3 This is a schematic diagram of the panoramic cutting network structure based on PatchPST;

[0024] Figure 4 Schematic diagram of a low-complexity segmentation head;

[0025] Figure 5 It is a flow chart of another radar pulse stream panoramic cutting method according to an embodiment of the present invention;

[0026] Figure 6 This is a schematic diagram of the quantization and annotation of radar pulse streams in the training set;

[0027] Figure 7 This is the result of the radar pulse stream panorama cutting based on PatchPST under ideal conditions;

[0028] Figure 8 This is the result of the pulse stream panorama cutting based on PatchPST when the probability of missing pulses is 10% and the false pulse rate is 2000 times / s;

[0029] Fig. 9 This is the result of the pulse stream panorama cutting based on PatchPST when the probability of missing pulses is 20% and the false pulse rate is 4000 times / s;

[0030] Fig.10 This is the result of the pulse stream panorama cutting based on PatchPST when the probability of missing pulses is 30% and the false pulse rate is 6000 times / s;

[0031] Fig.11 This is the result of the pulse stream panorama cutting based on PatchPST when the probability of missing pulses is 40% and the false pulse rate is 8000 times / s;

[0032] Fig.12 It is the pixel prediction accuracy in 6 scenarios;

[0033] Fig.13 IoU in 6 scenarios. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0036] Embodiment 1:

[0037] like Figure 1 As shown, an embodiment of the present invention provides a radar pulse stream panoramic cutting method, comprising:

[0038] Step S1, quantizing radar pulse stream;

[0039] Step S2: Perform panoramic cutting on the quantized radar pulse stream based on the PatchPST network.

[0040] As an implementation of the embodiment of the present invention, in step S1, TOA quantization coding technology is used to quantize and code the radar pulse stream.

[0041] Furthermore, when the receiver collects the pulse stream, it usually measures the pulse arrival time (TOA), pulse arrival angle (DOA), signal carrier frequency (Carrier Frequency), pulse width (Pulse Width), and pulse amplitude (Pulse Amplitude) of the radar pulse. In order to reflect the pulse repetition interval characteristics of the pulse stream, the present invention adopts TOA quantization coding technology, such as Figure 2 The detailed calculation process is shown in reference 6 (Visnevski N A. Syntactic modeling of multi-function radars [D]., 2005.)

[0042] The quantized sequence S can be expressed as

[0043] S={s1 s2 L s N}

[0044] Where N is the total number of quantization points, s i is the quantized value at the i-th moment.

[0045] As an implementation method of the present invention, in step S2, a panoramic cutting network structure based on PatchPST is designed, such as Figure 3 shown.

[0046] The panoramic segmentation network structure based on PatchPST consists of two parts, namely the skeleton part composed of PatchPST (Patch Pulse Stream Transformer) and the pulse stream panoptic segmentation header module (Pulse stream panoptic segmentation header).

[0047] (1) PatchPST network basic module

[0048] The PatchPST network basic module first cuts the intercepted pulse stream into non-overlapping patches. The patch sequence is input into the mapping module in order and then into the Transformer module to extract its feature vector.

[0049] Patching divides the radar pulse stream data into multiple patches, each of which represents a time window. This processing method allows the model to capture local and global features more effectively, expressed as

[0050] S={S1 S2 LS Np}

[0051] Among them, N p is the total number of partitions, S j is the jth block, which can be expressed as

[0052] S j ={s j1 s j2 L s jM}

[0053] Where M is the number of quantization points in each block.

[0054] First, each block quantization point is mapped to a D-dimensional feature space through a linear layer to facilitate feature extraction by the subsequent Transformer structure, which is expressed as

[0055] x j =f(s j1 s j2 L s jM )

[0056] in, is a D-dimensional feature vector. Then, N p The feature matrix of the block The input is sent to a Transformer module containing H attention heads. The attention module divides its query, key, and value parameters into H parts, and then processes each part independently through a separate attention head. All these attention calculation results are finally fused together to form the final attention score. Through the self-attention layer, PatchTST can effectively identify the relationship between the parts in the sequence and capture the temporal dependency. The attention formula can be expressed as

[0057]

[0058] Among them, Q, K and V represent Query, Key and Value respectively. k Represents the dimension of each attention head, usually Through the feature extraction of the Transformer encoder, the d-dimensional feature map of the pulse stream is obtained, which is expressed as

[0059] (2) Pulse Stream Panoramic Segmentation Head Module

[0060] The segmentation head is a component in the model that is specifically used for image segmentation tasks. It is usually located at the end of the model, immediately following the feature extraction part. Its main function is to convert the feature map extracted by the previous network into pixel-level classification predictions. The design of the segmentation head has a direct impact on the segmentation performance and efficiency of the model. In order to meet the needs of pulse stream panorama cutting, the present invention designs a low-complexity segmentation head, such as Figure 4 shown.

[0061] Feature Map First, input it into the one-dimensional convolution model. Let N W represents the number of radar word types contained in the multi-function radar. In the case of considering the predicted class, the panoramic cutting task contains N W +1 label category. Therefore, the one-dimensional convolution formula can be expressed as

[0062]

[0063] In order to achieve panoramic cutting, the prediction results are expanded using a linear layer, which can be expressed as

[0064]

[0065] Where N is the total number of quantization points. Then it is the probability matrix of the label corresponding to each quantization point.

[0066] Example experiment:

[0067] like Figure 5 As shown, the present invention implements a radar pulse stream panoramic cutting method based on PatchPST, comprising:

[0068] Step 1: Build a training set for radar word cutting

[0069] The radar pulse train is quantized using the TOA quantization method. In the training set, the quantization points are marked, where the quantization points in the radar word are marked as W i (i=1,2,K,N W ), the background quantization point is marked as W0, such as Figure 6 shown.

[0070] In the present invention, a total of 9 different radar words are designed. Different radar words have the same time length, and the only difference is that the PRI of the pulse is different. The PRI set of 9 radar words can be expressed as {111μs, 99μs, 88μs, 77μs, 66μs, 22μs, 16.5μs, 11μs, 5.5μs}. All radar word backgrounds in the pulse stream are classified as 1 category, so a total of 10 labels are included in the panoramic cutting task.

[0071] The second step is to train the radar pulse stream panorama segmentation network based on PatchPST. The BiLSTM-Filt network structure is as follows: Figure 7 shown.

[0072] The third step is to build a test set for panoramic cutting of radar pulse stream.

[0073] Step 4: Test the radar pulse stream panoramic cutting network based on PatchPST.

[0074] In an ideal situation, the radar word recognition result is as follows Figure 7 As shown, considering the radar word cutting in the case of missing pulses, the result is as follows Figure 8 , 9 , 10, 11, the recognition results are shown in Table 1, and the visualization results are shown in Fig.12 , 13 shown.

[0075] Table 1

[0076]

[0077]

[0078] Embodiment 2:

[0079] The embodiment of the present invention further provides a radar pulse stream panoramic cutting device, comprising:

[0080] A quantization module, used to quantize radar pulse stream;

[0081] The cutting module is used to perform panoramic cutting on the quantized radar pulse stream based on the PatchPST network.

[0082] As an implementation method of the embodiment of the present invention, the quantization module uses TOA quantization coding technology to quantize and encode the pulse stream of the radar.

[0083] As an implementation method of an embodiment of the present invention, the PatchPST network includes: a PatchPST network basic module and a pulse stream panoramic segmentation head module; wherein the PatchPST network basic module is used to cut the intercepted pulse stream into non-overlapping blocks, and the block sequence is sequentially input into the mapping module in a front-to-back order, and is input into the Transformer module to extract feature vectors; the pulse stream panoramic segmentation head module is used to convert the feature map into a pixel-level classification prediction

[0084] Embodiment 3:

[0085] An embodiment of the present invention further provides a radar pulse stream panoramic cutting system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a radar pulse stream panoramic cutting method when executed by the processor.

[0086] Embodiment 4:

[0087] An embodiment of the present invention further provides a storage medium, on which a computer program is stored, and the computer program executes a radar pulse stream panoramic cutting method when running.

[0088] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A radar pulse stream panoramic cutting method, characterized in that: include: Step S1, quantizing radar pulse stream; Step S2: Perform panoramic cutting on the quantized radar pulse stream based on the PatchPST network.

2. The radar pulse stream panoramic cutting method according to claim 1, characterized in that: In step S1, the radar pulse stream is quantized and encoded using TOA quantization encoding technology.

3. The radar pulse stream panoramic cutting method according to claim 2, characterized in that: The PatchPST network includes: a PatchPST network basic module and a pulse stream panoramic segmentation head module; wherein, the PatchPST network basic module is used to cut the intercepted pulse stream into non-overlapping blocks, and the block sequence is input into the mapping module in sequence, and is input into the Transformer module to extract feature vectors; the pulse stream panoramic segmentation head module is used to convert the feature map into pixel-level classification prediction.

4. A radar pulse stream panoramic cutting device, characterized in that: include: A quantization module, used to quantize radar pulse stream; The cutting module is used to perform panoramic cutting on the quantized radar pulse stream based on the PatchPST network.

5. The radar pulse stream panoramic cutting device according to claim 4, characterized in that: The quantization module uses TOA quantization coding technology to quantize the radar pulse stream.

6. The radar pulse stream panoramic cutting device according to claim 5, characterized in that: The PatchPST network includes: a PatchPST network basic module and a pulse stream panoramic segmentation head module; wherein, the PatchPST network basic module is used to cut the intercepted pulse stream into non-overlapping blocks, and the block sequence is input into the mapping module in sequence, and is input into the Transformer module to extract feature vectors; the pulse stream panoramic segmentation head module is used to convert the feature map into pixel-level classification prediction.

7. A radar pulse stream panoramic cutting system, characterized in that: include: A memory and a processor, wherein the memory stores a computer program executed by the processor, and when the computer program is executed by the processor, the radar pulse stream panoramic cutting method according to any one of claims 1 to 3 is executed.

8. A storage medium, characterized in that: The storage medium stores a computer program, which, when running, executes the radar pulse stream panoramic cutting method according to any one of claims 1 to 3.

Images