HRRP sequence target radial length estimation method and device based on deep learning

Abstract

The invention relates to an HRRP sequence target radial length estimation method and device based on deep learning, and the method comprises the steps: carrying out the normalization processing of each sample one-dimensional high-resolution range profile in a data set, converting the normalized range profile into a corresponding sample two-dimensional image, carrying out the calculation of the radial length of a target in each sample one-dimensional high-resolution range profile, and carrying out the calculation of the radial length of a target in each sample one-dimensional high-resolution range profile. The method comprises the following steps: acquiring two-dimensional images of each sample, taking the two-dimensional images as labels, constructing a training data set according to the two-dimensional images of each sample and the corresponding labels, training a target radial length estimation neural network by using the training data set, performing normalization processing on a one-dimensional high-resolution distance image to be subjected to target estimation, and converting the one-dimensional high-resolution distance image into two-dimensional images. And estimating the radial length of the target in the one-dimensional high-resolution range profile by using the trained target radial length estimation neural network according to the two-dimensional image. By adopting the method, the estimation accuracy of the radial length of the target can be effectively improved, and the radial length of the target is used as priori knowledge to assist subsequent target detection work.

Description

Technical Field

[0001] The present application relates to the technical field of radar target recognition, and in particular to a method and device for estimating the radial length of a HRRP sequence target based on deep learning. Background Art

[0002] With the support of high-resolution imaging technology, radar targets are no longer simplified to single scattering points, but are presented as a collection of multiple scattering points, which are closely arranged in the distance dimension to form a series of adjacent peaks and produce the outline of a one-dimensional high-resolution range profile (HRRP). Therefore, HRRP is highly related to the physical structural characteristics of the target surface, and the target length information, as one of the target separability information, can be used to roughly classify the target or as a feature for target identification. Therefore, it is of great significance to study how to use HRRP to extract the length feature of the target.

[0003] Currently, the method of extracting the radial length of the target from the HRRP sequence mainly relies on detecting the boundary of the HRRP support area. The traditional threshold detection method is often used. However, the radial length estimation accuracy based on the existing method is low under low signal-to-noise ratio conditions, and it is easy to miss detection when the edge signal of the target support area is weak, making it difficult to accurately locate the scattering points at the edge of the support area. Summary of the invention

[0004] Based on this, it is necessary to provide a deep learning-based HRRP sequence target radial length estimation method and device that can effectively improve detection accuracy in response to the above technical problems.

[0005] The present application provides a method for estimating radial length of a HRRP sequence target based on deep learning, the method comprising:

[0006] Acquire a data set, wherein the data set includes a plurality of sample one-dimensional high-resolution range images of different targets;

[0007] After normalizing each of the one-dimensional high-resolution range images of the samples in the data set, the images are converted into corresponding two-dimensional images of the samples;

[0008] Based on the sample two-dimensional image, the radial length of the target in each of the sample one-dimensional high-resolution range images is calculated and used as a label of the corresponding sample two-dimensional image;

[0009] Constructing a training data set according to each of the sample two-dimensional images and the corresponding labels, and training a target radial length estimation neural network according to the training data set to obtain a trained target radial length estimation neural network;

[0010] A one-dimensional high-resolution range image of the target to be estimated is obtained, the one-dimensional high-resolution range image is normalized and converted into a corresponding two-dimensional image, and the radial length of the target in the one-dimensional high-resolution range image is estimated based on the two-dimensional image using a trained neural network for estimating the radial length of the target.

[0011] In one embodiment, after the sample one-dimensional high-resolution range image is normalized, the original one-dimensional high-resolution range image sequence is mapped to the interval [-1, 1], and the following formula is used for normalization:

[0012]

[0013] In the above formula, x i represents the i-th point in the original one-dimensional high-resolution range image sequence, By x i The normalized sequence, X min , X max They represent the minimum and maximum values ​​in the original one-dimensional high-resolution range image sequence respectively.

[0014] In one embodiment, when converting the normalized one-dimensional high-resolution range image of the sample into a two-dimensional image:

[0015] Map the normalized one-dimensional high-resolution range image of the sample into a polar coordinate system;

[0016] Based on the polar coordinate system, the cosine value corresponding to the sum of polar angles of two adjacent points on the normalized one-dimensional high-resolution range image sequence of the sample is calculated;

[0017] The amplitude of each pixel on the sample two-dimensional image is obtained according to the cosine value to obtain the sample two-dimensional image.

[0018] In one embodiment, the radial length of the target in each of the sample one-dimensional high-resolution range images is calculated based on the sample two-dimensional image using the following formula:

[0019]

[0020] In the above formula, L, W, and H represent the prior length, width, and height of the target in the one-dimensional high-resolution range image of the sample, respectively.

[0021] In one embodiment, the target radial length estimation neural network uses a pre-trained and optimized ResNet-101 neural network;

[0022] The MSE loss function is used when training the target radial length estimation neural network.

[0023] In one embodiment, the optimized ResNet-101 neural network optimizes the first convolutional layer to a single-channel input based on the original ResNet-101 neural network, and adds a fully connected layer to the output layer.

[0024] In one embodiment, after obtaining the estimated radial length of the target using the trained target radial length estimation neural network, the estimated radial length is calculated based on the range resolution of the radar system that obtains the one-dimensional high-resolution range image of the target to be estimated to obtain the final target radial length.

[0025] The present application also provides a device for estimating radial length of a HRRP sequence target based on deep learning, the device comprising:

[0026] A data set acquisition module, used to acquire a data set, wherein the data set includes a plurality of sample one-dimensional high-resolution range images of different targets;

[0027] A two-dimensional conversion module, used for normalizing each of the one-dimensional high-resolution range images of the samples in the data set and converting them into corresponding two-dimensional images of the samples;

[0028] A radial length label calculation module, used to calculate the radial length of the target in each of the sample one-dimensional high-resolution range images based on the sample two-dimensional image, and use it as a label of the corresponding sample two-dimensional image;

[0029] A neural network training module constructs a training data set according to each of the sample two-dimensional images and the corresponding labels, and trains the target radial length estimation neural network according to the training data set to obtain a trained target radial length estimation neural network;

[0030] The target radial length obtaining module is used to obtain a one-dimensional high-resolution range image of the target to be estimated, normalize the one-dimensional high-resolution range image, and convert it into a corresponding two-dimensional image, and use the trained target radial length estimation neural network to estimate the radial length of the target according to the two-dimensional image to obtain the radial length of the target in the one-dimensional high-resolution range image.

[0031] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0032] Acquire a data set, wherein the data set includes a plurality of sample one-dimensional high-resolution range images of different targets;

[0033] After normalizing each of the one-dimensional high-resolution range images of the samples in the data set, the images are converted into corresponding two-dimensional images of the samples;

[0034] Based on the sample two-dimensional image, the radial length of the target in each of the sample one-dimensional high-resolution range images is calculated and used as a label of the corresponding sample two-dimensional image;

[0035] Constructing a training data set according to each of the sample two-dimensional images and the corresponding labels, and training a target radial length estimation neural network according to the training data set to obtain a trained target radial length estimation neural network;

[0036] A one-dimensional high-resolution range image of the target to be estimated is obtained, the one-dimensional high-resolution range image is normalized and converted into a corresponding two-dimensional image, and the radial length of the target in the one-dimensional high-resolution range image is estimated based on the two-dimensional image using a trained neural network for estimating the radial length of the target.

[0037] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0038] Acquire a data set, wherein the data set includes a plurality of sample one-dimensional high-resolution range images of different targets;

[0039] After normalizing each of the one-dimensional high-resolution range images of the samples in the data set, the images are converted into corresponding two-dimensional images of the samples;

[0040] Based on the sample two-dimensional image, the radial length of the target in each of the sample one-dimensional high-resolution range images is calculated and used as a label of the corresponding sample two-dimensional image;

[0041] Constructing a training data set according to each of the sample two-dimensional images and the corresponding labels, and training a target radial length estimation neural network according to the training data set to obtain a trained target radial length estimation neural network;

[0042] A one-dimensional high-resolution range image of the target to be estimated is obtained, the one-dimensional high-resolution range image is normalized and converted into a corresponding two-dimensional image, and the radial length of the target in the one-dimensional high-resolution range image is estimated based on the two-dimensional image using a trained neural network for estimating the radial length of the target.

[0043] The above-mentioned deep learning-based HRRP sequence target radial length estimation method and device, after normalizing each sample one-dimensional high-resolution range image in the data set, converts it into a corresponding sample two-dimensional image, calculates the radial length of the target in each sample one-dimensional high-resolution range image based on the sample two-dimensional image, and uses it as the label of the corresponding sample two-dimensional image, constructs a training data set according to each of the sample two-dimensional images and the corresponding label, uses the training data set to train the target radial length estimation neural network to obtain a trained target radial length estimation neural network, and converts the one-dimensional high-resolution range image to be estimated into a corresponding two-dimensional image after normalizing the one-dimensional high-resolution range image, and uses the trained target radial length estimation neural network to estimate the radial length of the target in the one-dimensional high-resolution range image according to the two-dimensional image. The use of this method can effectively improve the accuracy of target radial length estimation, and use it as prior knowledge to assist subsequent target detection work. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 1 is a flow chart of a method for estimating radial length of a HRRP sequence target based on deep learning in one embodiment;

[0045] Figure 2 is a flowchart of a method for estimating radial length of a HRRP sequence target based on deep learning in another embodiment;

[0046] Figure 3 A schematic diagram of a process of fine-tuning a ResNet-101 in another embodiment;

[0047] Figure 4 This is a schematic diagram comparing the changes in loss function during training and testing of this method and one-dimensional CNN in an experiment.

[0048] Figure 5 is a structural block diagram of a device for estimating radial length of a HRRP sequence target based on deep learning in one embodiment;

[0049] Figure 6 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0051] In the prior art, the method of extracting the radial length of the target from the HRRP (High Resolution Range Profile) sequence mainly relies on detecting the boundary of the HRRP support area, and the traditional threshold detection method is often used. However, the radial length estimation accuracy based on the existing method is low under low signal-to-noise ratio conditions, and it is easy to miss detection when the edge signal of the target support area is weak, and it is difficult to accurately locate the scattering points at the edge of the support area. Figure 1 As shown, a method for estimating the radial length of a HRRP sequence target based on deep learning is provided, comprising the following steps:

[0052] Step S100: acquiring a data set, wherein the data set includes a plurality of sample one-dimensional high-resolution range images of different targets.

[0053] Step S110 , after normalizing each sample one-dimensional high-resolution range image in the data set, convert it into a corresponding sample two-dimensional image.

[0054] Step S120 , based on the sample two-dimensional image, the radial length of the target in each sample one-dimensional high-resolution range image is calculated and used as a label of the corresponding sample two-dimensional image.

[0055] Step S130 , constructing a training data set according to each sample two-dimensional image and the corresponding label, and training the target radial length estimation neural network according to the training data set to obtain a trained target radial length estimation neural network.

[0056] Step S140, obtaining a one-dimensional high-resolution range image of the target to be estimated, normalizing the one-dimensional high-resolution range image, and converting it into a corresponding two-dimensional image, using a trained target radial length estimation neural network to estimate the radial length of the target based on the two-dimensional image, and obtaining the radial length of the target in the one-dimensional high-resolution range image.

[0057] In this embodiment, a deep neural network is used to estimate the radial length of the target. At the same time, in view of the fact that the one-dimensional HRRP sequence contains time series information, the one-dimensional HRRP sequence is converted into a two-dimensional image as the input of the deep neural network, and the radial length is used as a label to fine-tune its parameters. Then, the radial length is estimated by the readjusted network, which solves the current problem of poor accuracy in extracting the radial length of the target based on the HRRP sequence. Steps S100 to S130 are the process of training the neural network, and step S140 is to use the trained neural network to estimate the radial length of the target in the one-dimensional HRRP.

[0058] In step S100, the target type in the sample one-dimensional high-resolution range image in the data set can be consistent with the type of target to be estimated in step S140, that is, for different target detection tasks, sample images of corresponding target types can be used when training the neural network to obtain a target radial length estimation neural network with targeted target classes. Alternatively, sample images related to multiple different categories of targets can be used to train the neural network to obtain a target radial length estimation neural network that can simultaneously target multiple different target detection tasks.

[0059] It should be noted here that in the process of training the neural network and in the process of preprocessing the image of the target radial length, normalization processing and two-dimensional image conversion are included. In these two stages, the normalization processing means and two-dimensional image conversion means used are consistent. Therefore, here these two technical means are only explained in detail in the process of explaining the neural network training process.

[0060] In step S110, after the sample one-dimensional high-resolution range image is normalized, the original one-dimensional high-resolution range image sequence is mapped to the interval [-1, 1], and the following formula is used for normalization:

[0061]

[0062] In formula (1), x i represents the i-th point in the original one-dimensional high-resolution range image sequence, By x i The normalized sequence, X min , X max They represent the minimum and maximum values ​​in the original one-dimensional high-resolution range image sequence respectively.

[0063] Furthermore, the normalized one-dimensional high-resolution range image of the sample is converted into a two-dimensional Gram Angular Field (GAF) image. Specifically, the normalized one-dimensional high-resolution range image of the sample is first mapped to a polar coordinate system, and based on the polar coordinate system, a cosine value corresponding to the sum of polar angles of two adjacent points on the normalized one-dimensional high-resolution range image sequence of the sample is calculated. Finally, the amplitude of each pixel on the sample two-dimensional image is obtained according to the cosine value to obtain the sample two-dimensional image.

[0064] In this embodiment, the normalized one-dimensional high-resolution range image of the sample is mapped to a polar coordinate system using the following formula:

[0065]

[0066] In formula (2), xi The arc cosine value is the corresponding polar angle φ in the polar coordinate system i , and the time step i / N is used as the polar radius. According to the monotonicity of the inverse cosine function on the domain, it can be calculated that the interval [-1,1] is mapped one-to-one to [0,π], which is φ i The value range of .

[0067] Next, by calculating the cosine value corresponding to the sum of the polar angles of two adjacent points on the HRRP sequence, the amplitude of each pixel on the GAF image can be obtained, retaining the timing information to the greatest extent, that is:

[0068]

[0069] At this point, the one-dimensional HRRP sequence of length n is converted into a two-dimensional GAF image of size n×n.

[0070] In step S120, based on the sample two-dimensional image, the azimuth angle θ and the pitch angle corresponding to each two-dimensional GAF image are calculated. The radial length label corresponding to the HRRP is obtained. Since the original data set often does not provide the radial length label D, in this embodiment, the radial length label D is obtained based on the prior data of the length, width and height of the target, the azimuth angle θ, the pitch angle The radial length of the target in the sample one-dimensional high-resolution range image is calculated using the following formula:

[0071]

[0072] In formula (4), L, W, and H represent the prior length, width, and height of the target in the one-dimensional high-resolution range image of the sample, respectively.

[0073] Next, the sample two-dimensional image and the corresponding target radial length label are input into the target radial length estimation neural network for training.

[0074] In this embodiment, the target radial length estimation neural network uses a pre-trained and optimized ResNet-101 neural network. In the process of fine-tuning the ResNet-101 network parameters, that is, in the process of training the neural network, the back propagation (BP) and Adam optimizer of Pytorch are used to update the parameters, and the MSE loss function is used, which is expressed as:

[0075]

[0076] In formula (5), N represents the number of samples involved in training in one iteration. j , Represents the true value of the estimated j-th radial length, i.e., the label, and its estimated value.

[0077] In this embodiment, in order to make the ResNet-101 neural network adapt to the GAF two-dimensional image as input, since the two-dimensional image is a single-channel feature, the first convolutional layer of the ResNet-101 neural network is adjusted to a single-channel input. At the same time, since the ResNet-101 neural network is applied to target recognition, the output is the target position detection box and category probability, and this method estimates the radial length of the target, so a fully connected layer is connected at the end of the ResNet-101 neural network to adapt to the regression task requirements.

[0078] In step S140, after the estimated radial length of the target is obtained using the trained target radial length estimation neural network, the projection length of the target in the radar line of sight direction is obtained by the radar range resolution. Therefore, the estimated radial length needs to be calculated based on the range resolution of the radar system that obtains the one-dimensional high-resolution range image of the target to be estimated, so as to obtain the final target radial length.

[0079] In this embodiment, the final target radial length is calculated using the following formula:

[0080]

[0081] In the above formula, ΔR x represents the range resolution of the broadband radar system, RL represents the actual target radial length, represents the target radial length estimated by the target radial length estimation neural network.

[0082] In this embodiment, another embodiment is provided to further illustrate the method for estimating the radial length of a HRRP sequence target based on deep learning. The method steps are as follows: Figure 2 shown.

[0083] like Figure 2 As shown in FIG, the radial length estimation method of the HRRP sequence target based on deep learning includes the following steps: first, the obtained HRRP sequence is normalized; second, the normalized one-dimensional HRRP sequence is mapped to the polar coordinate system; third, the amplitude of each pixel point of the GAF image is calculated; then, each azimuth angle θ and pitch angle The radial length label corresponding to HRRP is obtained; then, the pre-trained ResNet-101 network provided by Pytorch is fine-tuned; finally, the output of the fine-tuned ResNet-101 is used to estimate the radial length.

[0084] The fifth step is an implementation method for fine-tuning the pre-trained ResNet-101 provided by Pytorch, such as Figure 3 shown.

[0085] In the process of fine-tuning the pre-trained ResNet-101, based on the electromagnetic simulation software, the X-band was selected for simulation, and the radar signal bandwidth was set to 8.5GHz~11.5GHz with a step size of 5MHz; the azimuth angle range was 75°-105° with a step size of 3°; the pitch angle range was 0°~60° with a step size of 0.05°. Six different types of aviation target models were selected for simulation to obtain a HRRP data set of size 6×10×1200. The data set was randomly divided into a training set, a validation set, and a test set in a ratio of 8:1:1. Then, the divided one-dimensional HRRP was converted into a two-dimensional GAF image, and the corresponding label was obtained by the above formula (4) to obtain the network input.

[0086] Furthermore, in the fine-tuning of the ResNet-101 network, Pytorch's back propagation (BP) and Adam optimizer are used to update parameters, and the loss function used is MSE (Mean Square Error).

[0087] In order to verify the performance of the method proposed in this paper (this method), two simulation experiments are carried out to illustrate it.

[0088] Simulation experiment 1 compares the effects of the method based on simple neural network and this method. Figure 4 (a)(b) shows that the method has the characteristics of fast convergence and low convergence value, and is stable on both the training set and the test set. Table 1 also shows that the method proposed in the present invention is better than the one-dimensional CNN under different SNR conditions.

[0089] The structure of the one-dimensional CNN involved contains three convolution modules, all with a step size of 1, a convolution kernel size of 3×3, and output channels of 32, 64, and 128, respectively. Each convolution layer is followed by a corresponding BN (Batch Normalization) layer to reduce overfitting, and at the end of the network are two fully connected (FC) layers for outputting regression results. In the experiment, the batch size is set to 32, the learning rate is set to 0.001, and the number of iterations is set to 100.

[0090] Table 1 Radial length estimation results of traditional threshold method, one-dimensional CNN and this method under different SNR conditions ( Mean Relative Error)

[0091]

[0092] Simulation experiment 2 compares the effects of the traditional threshold method and the method based on deep learning. In Table 1, the performance of the traditional threshold method is not as good as the one-dimensional CNN based on deep learning and this method. Figure 4 (c) It can be seen that under the condition of SNR=10, the traditional threshold method is greatly affected by noise, and the one-dimensional CNN method is also insufficient in accuracy. However, the detection method proposed in the present invention still has a good detection effect, indicating that the detection effect of this algorithm is robust in different noise environments.

[0093] In the above-mentioned deep learning-based HRRP sequence target radial length estimation method, the one-dimensional HRRP sequence is converted into a two-dimensional Gram Angular Field (GAF) image as input in view of the fact that the one-dimensional HRRP sequence contains temporal information. The pre-network ResNet-101 is fine-tuned with the radial length as the label, and then the radial length is estimated by the readjusted network, which solves the current problem of poor accuracy in radial length extraction of HRRP sequence-based targets.

[0094] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0095] In one embodiment, Figure 5 As shown, a device for estimating the target radial length of a HRRP sequence based on deep learning is provided, comprising: a data set acquisition module 200, a two-dimensional conversion module 210, a radial length label calculation module 220, a neural network training module 230 and a target radial length acquisition module 240, wherein:

[0096] The data set acquisition module 200 is used to acquire a data set, wherein the data set includes a plurality of sample one-dimensional high-resolution range images of different targets;

[0097] A two-dimensional conversion module 210 is used to perform normalization processing on each of the sample one-dimensional high-resolution range images in the data set and convert it into a corresponding sample two-dimensional image;

[0098] A radial length label calculation module 220 is used to calculate the radial length of the target in each of the sample one-dimensional high-resolution range images based on the sample two-dimensional image, and use it as a label of the corresponding sample two-dimensional image;

[0099] A neural network training module 230 constructs a training data set according to each of the sample two-dimensional images and the corresponding labels, and trains a target radial length estimation neural network according to the training data set to obtain a trained target radial length estimation neural network;

[0100] The target radial length obtaining module 240 is used to obtain a one-dimensional high-resolution range image of the target to be estimated, normalize the one-dimensional high-resolution range image, and convert it into a corresponding two-dimensional image, and use the trained target radial length estimation neural network to estimate the radial length of the target based on the two-dimensional image to obtain the radial length of the target in the one-dimensional high-resolution range image.

[0101] For the specific limitations of the HRRP sequence target radial length estimation device based on deep learning, please refer to the limitations of the HRRP sequence target radial length estimation method based on deep learning above, which will not be repeated here. Each module in the above-mentioned HRRP sequence target radial length estimation device based on deep learning can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0102] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for estimating the radial length of a HRRP sequence target based on deep learning is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0103] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0104] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0105] Acquire a data set, wherein the data set includes a plurality of sample one-dimensional high-resolution range images of different targets;

[0106] After normalizing each of the one-dimensional high-resolution range images of the samples in the data set, the images are converted into corresponding two-dimensional images of the samples;

[0107] Based on the sample two-dimensional image, the radial length of the target in each of the sample one-dimensional high-resolution range images is calculated and used as a label of the corresponding sample two-dimensional image;

[0108] Constructing a training data set according to each of the sample two-dimensional images and the corresponding labels, and training a target radial length estimation neural network according to the training data set to obtain a trained target radial length estimation neural network;

[0109] A one-dimensional high-resolution range image of the target to be estimated is obtained, the one-dimensional high-resolution range image is normalized and converted into a corresponding two-dimensional image, and the radial length of the target in the one-dimensional high-resolution range image is estimated based on the two-dimensional image using a trained neural network for estimating the radial length of the target.

[0110] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0111] Acquire a data set, wherein the data set includes a plurality of sample one-dimensional high-resolution range images of different targets;

[0112] After normalizing each of the one-dimensional high-resolution range images of the samples in the data set, the images are converted into corresponding two-dimensional images of the samples;

[0113] Based on the sample two-dimensional image, the radial length of the target in each of the sample one-dimensional high-resolution range images is calculated and used as a label of the corresponding sample two-dimensional image;

[0114] Constructing a training data set according to each of the sample two-dimensional images and the corresponding labels, and training a target radial length estimation neural network according to the training data set to obtain a trained target radial length estimation neural network;

[0115] A one-dimensional high-resolution range image of the target to be estimated is obtained, the one-dimensional high-resolution range image is normalized and converted into a corresponding two-dimensional image, and the radial length of the target in the one-dimensional high-resolution range image is estimated based on the two-dimensional image using a trained neural network for estimating the radial length of the target.

[0116] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0117] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0118] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A deep learning-based method for estimating radial length of HRRP sequence targets. It is characterized in that The method comprises: Acquire a data set, wherein the data set includes a plurality of sample one-dimensional high-resolution range images of different targets; After normalizing each of the one-dimensional high-resolution range images of the samples in the data set, the images are converted into corresponding two-dimensional images of the samples; Based on the sample two-dimensional image, the radial length of the target in each of the sample one-dimensional high-resolution range images is calculated and used as a label of the corresponding sample two-dimensional image; Constructing a training data set according to each of the sample two-dimensional images and the corresponding labels, and training a target radial length estimation neural network according to the training data set to obtain a trained target radial length estimation neural network; A one-dimensional high-resolution range image of the target to be estimated is obtained, the one-dimensional high-resolution range image is normalized and converted into a corresponding two-dimensional image, and the radial length of the target in the one-dimensional high-resolution range image is estimated based on the two-dimensional image using a trained neural network for estimating the radial length of the target.

2. The HRRP sequence target radial length estimation method according to claim 1, It is characterized in that After the sample one-dimensional high-resolution range image is normalized, the original one-dimensional high-resolution range image sequence is mapped to the interval [-1, 1], and the following formula is used for normalization: In the above formula, x i represents the i-th point in the original one-dimensional high-resolution range image sequence, By x i The normalized sequence, X min , X max They represent the minimum and maximum values ​​in the original one-dimensional high-resolution range image sequence respectively.

3. The HRRP sequence target radial length estimation method according to claim 1, It is characterized in that When converting the normalized sample one-dimensional high-resolution range image into a two-dimensional image: Map the normalized one-dimensional high-resolution range image of the sample into a polar coordinate system; Based on the polar coordinate system, the cosine value corresponding to the sum of polar angles of two adjacent points on the normalized one-dimensional high-resolution range image sequence of the sample is calculated; The amplitude of each pixel on the sample two-dimensional image is obtained according to the cosine value to obtain the sample two-dimensional image.

4. The method for estimating radial length of a HRRP sequence target according to claim 1, It is characterized in that Based on the sample two-dimensional image, the radial length of the target in each sample one-dimensional high-resolution range image is calculated using the following formula: In the above formula, L, W, and H represent the prior length, width, and height of the target in the one-dimensional high-resolution range image of the sample, respectively.

5. The method for estimating radial length of a HRRP sequence target according to claim 1, It is characterized in that The target radial length estimation neural network adopts a pre-trained and optimized ResNet-101 neural network; The MSE loss function is used when training the target radial length estimation neural network.

6. The method for estimating radial length of a HRRP sequence target according to claim 5, It is characterized in that The optimized ResNet-101 neural network optimizes the first convolutional layer to a single-channel input based on the original ResNet-101 neural network, and adds a fully connected layer to the output layer.

7. The method for estimating radial length of a HRRP sequence target according to any one of claims 1 to 5, It is characterized in that After obtaining the estimated radial length of the target using the trained target radial length estimation neural network, the estimated radial length is calculated according to the range resolution of the radar system that obtains the one-dimensional high-resolution range image of the target to be estimated to obtain the final target radial length.

8. A deep learning-based HRRP sequence target radial length estimation device, It is characterized in that The device comprises: A data set acquisition module, used to acquire a data set, wherein the data set includes a plurality of sample one-dimensional high-resolution range images of different targets; A two-dimensional conversion module, used for normalizing each of the one-dimensional high-resolution range images of the samples in the data set and converting them into corresponding two-dimensional images of the samples; A radial length label calculation module, used to calculate the radial length of the target in each of the sample one-dimensional high-resolution range images based on the sample two-dimensional image, and use it as a label of the corresponding sample two-dimensional image; A neural network training module constructs a training data set according to each of the sample two-dimensional images and the corresponding labels, and trains the target radial length estimation neural network according to the training data set to obtain a trained target radial length estimation neural network; The target radial length obtaining module is used to obtain a one-dimensional high-resolution range image of the target to be estimated, normalize the one-dimensional high-resolution range image, and convert it into a corresponding two-dimensional image, and use the trained target radial length estimation neural network to estimate the radial length of the target according to the two-dimensional image to obtain the radial length of the target in the one-dimensional high-resolution range image.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program. It is characterized in that When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images