An ultra-wideband impact fuze target recognition method based on heading information fusion

Abstract

This invention discloses an ultra-wideband impact fuze target recognition method based on heading information fusion, including the following steps: S100: UWB-IF target signal and heading information acquisition: For UWB-IF targets, the instantaneous heading information of the target is acquired while collecting the target radar echo signal; S200: DF-CBD network model design: A DF-CBD model is designed, including a radar signal recognition channel and a heading auxiliary channel, adopting a dual-channel form, using the target heading information to constrain the range of the basic target domain, thereby improving the recognition accuracy probabilistically; S300: DF-CBD model performance verification: For the same dataset, under different signal-to-noise ratio environments, the target recognition rate is compared with and without heading fusion.

Description

Technical Field

[0001] This invention relates to the field of ultra-wideband impact fuze target identification, and in particular to an ultra-wideband impact fuze target identification method based on heading information fusion. Background Technology

[0002] Ultra-wideband radar (UWBR) is a novel radar technology that has been widely applied in military and civilian fields such as gesture recognition, vital sign detection, human imaging, and autonomous driving, achieving good results. UWBRs are generally classified into two types based on the waveform of the transmitted signal: impulse pulse UWBR (UWB-IR) and non-impulse pulse UWBR. UWB-IF is an application of UWB-IR in fuzes, also known as a carrierless ultra-wideband fuze. It is a typical transient radar whose radiation typically consists of an impulse pulse sequence with an ultra-wide bandwidth.

[0003] Most commonly used UWB-IF target recognition methods utilize the one-dimensional echo signal of the target from radar, i.e., the high-resolution range profile (HRRP). Common HRRP-based target recognition methods include traditional HRRP feature extraction and analysis methods, HRRP two-dimensional processing methods, and deep learning-based HRRP target recognition methods. Besides HRRP target recognition methods, various UWB-IF target recognition methods based on multi-dimensional information fusion exist, including but not limited to imaging the target and fusing target range and angle information, sequential fusion of the target, and fusion of target information from slower time dimensions. In fact, in addition to the aforementioned UWB-IF target recognition methods, prior knowledge about the target can be used to eliminate impossible recognition results, thereby probabilistically improving target recognition performance. Compared to other methods, methods that narrow the target domain based on prior knowledge of the target, thus improving recognition results, have stronger stability and can adapt to various complex environments and conditions.

[0004] Based on the above research, this patent proposes a UWB-IF target recognition method based on heading information fusion. It attempts to use prior knowledge of radar target heading information to constrain the target domain of the target recognition task, and designs a neural network model DF-CBD (Direction Fused CNN BiLSTM DNN) to achieve heading information fusion through a heading auxiliary channel, ultimately improving the UWB-IF target recognition performance under different signal-to-noise ratios. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention provides an ultra-wideband impact fuze target recognition method based on heading information fusion. Under different signal-to-noise ratios, it can utilize prior knowledge of the target heading information to constrain the target domain of the recognition task, thereby improving the performance of target recognition probabilistically. This method can provide more assistance for UWB-IF target recognition in different application scenarios.

[0006] To achieve the aforementioned objectives of the invention, the technical solution adopted to solve its technical problems is as follows:

[0007] A target identification method for ultra-wideband impact fuze based on heading information fusion includes the following steps:

[0008] Step S100: Acquisition of UWB-IF target signal and heading information: For UWB-IF targets, while collecting the target radar echo signal, acquire the target's instantaneous heading information;

[0009] Step S200: DF-CBD Network Model Design: Design the DF-CBD model, including the radar signal identification channel and the heading auxiliary channel. The dual-channel approach is adopted, and the target heading information is used to constrain the range of the basic target domain, thereby improving the identification accuracy from a probabilistic perspective.

[0010] Step S300: DF-CBD model performance verification: For the same dataset, under different signal-to-noise ratio environments, compare the target recognition rate with and without heading fusion.

[0011] Furthermore, in step S100, the acquisition of UWB-IF target signal and heading information not only requires acquiring the instantaneous echo signal of the target, but also the instantaneous heading information of the target, to ensure a one-to-one correspondence between the echo signal, the target type, and the target heading.

[0012] Furthermore, in step S200, the design of the DF-CBD network model includes a main channel and a heading auxiliary channel, each channel specifically including:

[0013] (1) Main passage:

[0014] Main channel one-dimensional convolutional layer 1: convolutional kernel length is 3, and the number of convolutional kernels is 48;

[0015] Heading Fusion Node 1: Fuses the output of the fully connected layer 1 with the feature constraints of the heading auxiliary channel and the output of the one-dimensional convolutional layer 1 of the main channel;

[0016] Main channel one-dimensional convolutional layer 2: convolutional kernel length is 3, and the number of convolutional kernels is 36;

[0017] Heading Fusion Node 2: Fuses the output of the fully connected layer 2 with the feature constraints of the heading auxiliary channel and the output of the one-dimensional convolutional layer 2 of the main channel;

[0018] Main channel one-dimensional convolutional layer 3: convolutional kernel length is 3, and the number of convolutional kernels is 16;

[0019] Heading fusion node 3: fuses the output of the fully connected layer 3 with the feature constraints of the heading auxiliary channel with the output of the one-dimensional convolutional layer 3 of the main channel;

[0020] Main channel BiLSTM layer 1: number of hidden nodes is 128;

[0021] Main channel BiLSTM layer 2: number of hidden nodes is 128;

[0022] Main channel Attention layer: self-attention mechanism;

[0023] Main channel fully connected layer 1: Number of nodes is 1024;

[0024] Heading Fusion Node 4: Fusion of the output of the heading auxiliary channel classification constraint fully connected layer 1 and the output of the main channel fully connected layer 1;

[0025] Main channel fully connected layer 2: Number of nodes is 128;

[0026] Heading Fusion Node 5: Fusion of the output of the heading auxiliary channel classification constraint fully connected layer 2 and the output of the main channel fully connected layer 2;

[0027] Main channel fully connected layer 3: Number of nodes is 3;

[0028] (2) Heading Auxiliary Channel:

[0029] Fully connected layer 1 with characteristic constraints for heading auxiliary channel: number of nodes is 48;

[0030] Fully connected layer 2 with characteristic constraints for heading auxiliary channel: number of nodes is 36;

[0031] Fully connected layer 3 with characteristic constraints for heading auxiliary channel: 16 nodes;

[0032] Fully connected layer 1 of the heading auxiliary channel classification constraint: number of nodes is 1024;

[0033] Fully connected layer 2 with heading auxiliary channel classification constraints: 128 nodes;

[0034] Except for the last fully connected layer whose output does not contain an activation function, all the other layers that do contain an activation function use ReLU, and the model loss function is the cross-entropy function.

[0035] Furthermore, the DF-CBD main channel is connected serially between each layer, while the heading auxiliary channel is connected in parallel according to feature constraint function and classification constraint function respectively. That is, the connection method is as follows:

[0036] Input -> fully connected layer 1 of heading auxiliary channel feature constraints -> fully connected layer 2 of heading auxiliary channel feature constraints -> fully connected layer 3 of heading auxiliary channel feature constraints, and input -> fully connected layer 1 of heading auxiliary channel classification constraints -> fully connected layer 2 of heading auxiliary channel classification constraints.

[0037] Furthermore, the DF-CBD main channel is used for classifying and identifying target radar echo signals, while the heading auxiliary channel corrects and constrains the output of the main channel through the heading fusion node.

[0038] Furthermore, the DF-CBD model uses prior information about the target heading to constrain the target domain of the main channel classification task, thereby improving the accuracy of the target recognition task probabilistically.

[0039] Furthermore, in step S300, the DF-CBD model performance verification is to compare the performance and results obtained by adding heading fusion and not adding heading fusion under different signal-to-noise ratio environments for the same dataset.

[0040] By employing the above technical solutions, this invention has the following advantages and positive effects compared with the prior art:

[0041] This invention proposes a UWB-IF target recognition method based on heading information fusion. By utilizing prior knowledge of target heading information, the target domain of the target recognition task is constrained, thereby improving the accuracy of the target recognition task from a probabilistic perspective. This provides a new technical approach for UWB-IF target recognition in different scenarios. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0043] Figure 1 This is a flowchart illustrating a UWB-IF target recognition method based on heading information fusion according to the present invention.

[0044] Figure 2 The correspondence between the three types of target signals (cars, SUVs, and trucks) and heading information collected;

[0045] Figure 3 To compare the recognition performance of the DF-CBD model under different signal-to-noise ratio conditions before and after incorporating heading information. Detailed Implementation

[0046] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] This embodiment discloses a target identification method for ultrawide band impulse fuzes (UWB-IF) based on heading information fusion, and its flowchart is shown below. Figure 1 As shown, it includes the following steps:

[0048] Step S100: Acquisition of UWB-IF target signal and heading information: For UWB-IF targets, while collecting the target radar echo signal, acquire the target's instantaneous heading information;

[0049] Step S200: DF-CBD (Direction Fused CNN-BiLSTM-DNN, Convolutional Neural Network with Fusion Heading Information - Bidirectional Long Short-Term Memory Network - Deep Neural Network) Network Model Design: Design the DF-CBD model, including radar signal recognition channel and heading auxiliary channel, adopting a dual-channel form, utilizing target heading information to constrain the range of the basic target domain, thereby improving the recognition accuracy from a probabilistic perspective;

[0050] Step S300: DF-CBD model performance verification: For the same dataset, under different signal-to-noise ratio environments, compare the target recognition rate with and without heading fusion.

[0051] The steps described above are explained in detail below:

[0052] In step S100, the acquisition of UWB-IF target signal and heading information requires not only acquiring the instantaneous echo signal of the target, but also acquiring the instantaneous heading information of the target to ensure a one-to-one correspondence between the echo signal, the target type, and the target heading.

[0053] Preferably, in step S100, the UWB-IF target signal and heading information are acquired as follows: Figure 2 As shown, the dataset used consists of measured data from a P440 ultra-wideband radar module against three different vehicle targets. Figure 2In the figures, from left to right, the data represent targets such as cars, SUVs, and trucks. The horizontal direction of each figure represents the dimension of the radar received signal, and the vertical direction represents different observation angle ranges. Due to manual operation errors, the speed at which the radar observation angle is changed varies in each experiment, resulting in different amounts of data for each measurement. However, all data follow a uniform distribution within the range of 0° to 360°. Different rectangles in the figures indicate data at different observation angles.

[0054] As can be seen from the figure: 1) When the observation conditions of the other radars are kept consistent, the radar echo signals of the same type of vehicle show significant differences under different observation angles; 2) When the observation conditions of the other radars are kept consistent, the differences between the three types of vehicles also change considerably with the change of observation angle. Therefore, it can be seen that the data collected by the radar for targets at different observation angles varies considerably, and the differences between the three types of vehicle targets also differ significantly under different observation angles. Thus, if the target's heading information can be used to obtain the radar's observation angle relative to the target, it will be of great help in ultra-wideband fuze target identification tasks.

[0055] In step S200, an ultra-wideband target recognition network model DF-CBD based on heading information fusion was designed. Specifically, the overall structure of the model is as follows: Figure 1 As shown in Table 1, the design of each layer and dimensional transformation of the DF-CBD model are as follows.

[0056] Table 1. Comparison of Layer Design and Dimensional Transformation in the DF-CBD Model

[0057]

[0058] Except for the last fully connected layer, which does not contain an activation function, all other layers that do contain an activation function use ReLU, and the model loss function is the cross-entropy function. The settings of some parameters of the above model can be adjusted according to the experimental results.

[0059] Furthermore, the DF-CBD main channel is connected serially between each layer, while the heading auxiliary channel is connected in parallel according to feature constraint function and classification constraint function respectively. That is, the connection method is as follows:

[0060] Input -> fully connected layer 1 of heading auxiliary channel feature constraints -> fully connected layer 2 of heading auxiliary channel feature constraints -> fully connected layer 3 of heading auxiliary channel feature constraints, and input -> fully connected layer 1 of heading auxiliary channel classification constraints -> fully connected layer 2 of heading auxiliary channel classification constraints.

[0061] Preferably, the main channel of the DF-CBD is used for classifying and identifying target radar echo signals, while the heading auxiliary channel corrects and constrains the output of the main channel through a heading fusion node. In general, the purpose of the heading auxiliary channel is to utilize prior knowledge of target heading information to constrain and correct the main channel in two steps: target echo signal feature extraction and target feature classification.

[0062] Furthermore, the DF-CBD model attempts to use prior information about the target heading to constrain the target domain of the main channel classification task, thereby improving the accuracy of the target recognition task probabilistically.

[0063] In step S300, the DF-CBD model performance is verified by dividing the data in step S100 into 36 groups according to the heading, i.e., specifying 36 different headings, and designing the training parameters of the DF-CBD model as shown in Table 2.

[0064] Table 2 Training Parameter Settings

[0065] parameter set up Batch-size 128 Maximum iteration period 50 Learning rate 0.0003

[0066] Figure 3 Table 3 shows the recognition performance of the DF-CBD model with added heading information and the CBD model without added heading information at different signal-to-noise ratios under the same set of data.

[0067] Table 3. Recognition performance of DF-CBD and CBD models as a function of SNR

[0068] Signal-to-noise ratio (dB) 10 5 0 -5 -10 CBD 99.57％ 98.63％ 95.29％ 86.01％ 71.23％ DF-CBD 100.00％ 98.91％ 96.47％ 90.51％ 78.59％

[0069] As shown in the charts, the performance of both the CBD and DF-CBD models declines with decreasing signal-to-noise ratio (SNR), and the rate of decline accelerates. Compared to the CBD model, the DF-CBD model exhibits a much slower decline and achieves better recognition performance across all SNR levels. These results demonstrate that the DF-CBD model, by incorporating heading information and constraining the target domain, probabilistically improves the accuracy of target recognition, thus enhancing the performance of the original CBD model across various SNR levels.

[0070] In summary, the UWB-IF target recognition method based on heading information fusion provided in this embodiment can constrain the target domain of the recognition task under different signal-to-noise ratios by utilizing prior knowledge of the target heading information, thereby improving the performance of target recognition probabilistically and providing more assistance for UWB-IF target recognition in different application scenarios.

[0071] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A target identification method for ultra-wideband impact fuze based on heading information fusion, characterized in that, Includes the following steps: Step S100: Acquisition of UWB-IF target signal and heading information: For UWB-IF targets, while collecting the target radar echo signal, acquire the target's instantaneous heading information; Step S200: DF-CBD Network Model Design: Design the DF-CBD model, including the radar signal identification channel and the heading auxiliary channel. The dual-channel approach is adopted, and the target heading information is used to constrain the range of the basic target domain, thereby improving the identification accuracy from a probabilistic perspective. In step S200, the design of the DF-CBD network model includes a main channel and a heading auxiliary channel, each channel specifically including: (1) Main passage: Main channel one-dimensional convolutional layer 1: convolutional kernel length is 3, and the number of convolutional kernels is 48; Heading Fusion Node 1: Fuses the output of the fully connected layer 1 with the feature constraints of the heading auxiliary channel and the output of the one-dimensional convolutional layer 1 of the main channel; Main channel one-dimensional convolutional layer 2: convolutional kernel length is 3, and the number of convolutional kernels is 36; Heading Fusion Node 2: Fuses the output of the fully connected layer 2 with the feature constraints of the heading auxiliary channel and the output of the one-dimensional convolutional layer 2 of the main channel; Main channel one-dimensional convolutional layer 3: convolutional kernel length is 3, and the number of convolutional kernels is 16; Heading fusion node 3: fuses the output of the fully connected layer 3 with the feature constraints of the heading auxiliary channel with the output of the one-dimensional convolutional layer 3 of the main channel; Main channel BiLSTM layer 1: number of hidden nodes is 128; Main channel BiLSTM layer 2: number of hidden nodes is 128; Main channel Attention layer: self-attention mechanism; Main channel fully connected layer 1: Number of nodes is 1024; Heading Fusion Node 4: Fusion of the output of the heading auxiliary channel classification constraint fully connected layer 1 and the output of the main channel fully connected layer 1; Main channel fully connected layer 2: Number of nodes is 128; Heading Fusion Node 5: Fusion of the output of the heading auxiliary channel classification constraint fully connected layer 2 and the output of the main channel fully connected layer 2; Main channel fully connected layer 3: Number of nodes is 3; (2) Heading Auxiliary Channel: Fully connected layer 1 with characteristic constraints for heading auxiliary channel: number of nodes is 48; Fully connected layer 2 with characteristic constraints for heading auxiliary channel: number of nodes is 36; Fully connected layer 3 with characteristic constraints for heading auxiliary channel: 16 nodes; Fully connected layer 1 of the heading auxiliary channel classification constraint: number of nodes is 1024; Fully connected layer 2 with heading auxiliary channel classification constraints: 128 nodes; Except for the fully connected layer that outputs to the last heading fusion node 5, which does not contain an activation function, all other layers that do contain an activation function are ReLU, and the model loss function is the cross-entropy function. Step S300: DF-CBD model performance verification: For the same dataset, under different signal-to-noise ratio environments, compare the target recognition rate with and without heading fusion.

2. The ultra-wideband impact fuze target identification method based on heading information fusion according to claim 1, characterized in that, In step S100, the acquisition of UWB-IF target signal and heading information requires not only the acquisition of the instantaneous echo signal of the target, but also the acquisition of the instantaneous heading information of the target, in order to ensure a one-to-one correspondence between the echo signal, the target type, and the target heading.

3. The ultra-wideband impact fuze target identification method based on heading information fusion according to claim 1, characterized in that, The DF-CBD main channel is connected serially between each layer, while the heading auxiliary channel is connected in parallel according to feature constraint function and classification constraint function respectively. That is, the connection method is as follows: Input → Fully connected layer 1 for heading auxiliary channel feature constraints → Fully connected layer 2 for heading auxiliary channel feature constraints → Fully connected layer 3 for heading auxiliary channel feature constraints, and Input → Fully connected layer 1 for heading auxiliary channel classification constraints → Fully connected layer 2 for heading auxiliary channel classification constraints.

4. The ultra-wideband impact fuze target identification method based on heading information fusion according to claim 1, characterized in that, The DF-CBD main channel is used for classifying and identifying target radar echo signals, while the heading auxiliary channel corrects and constrains the output of the main channel through the heading fusion node.

5. The ultra-wideband impact fuze target identification method based on heading information fusion according to claim 4, characterized in that, The DF-CBD model uses prior information about the target heading to constrain the target domain of the main channel classification task, thereby improving the accuracy of the target recognition task from a probabilistic perspective.

6. The ultra-wideband impact fuze target identification method based on heading information fusion according to claim 1, characterized in that, In step S300, the DF-CBD model performance verification is to compare the performance and results obtained by adding heading fusion and not adding heading fusion under different signal-to-noise ratio environments for the same dataset.

Images