A Method for Constructing an HRRP Recognition Database Based on Prototype Selection

Abstract

The present invention relates to a method for constructing an HRRP recognition database based on prototype selection, comprising the following steps: Step 1: extracting multiple radar HRRP data from a radar echo database to construct an initial HRRP database; Step 2: performing prototype selection on the radar HRRP data in the initial HRRP database using a probability distribution strategy and a constructed adaptive second-order optimizer to obtain a final radar HRRP prototype set; Step 3: using the final radar HRRP prototype set as an HRRP recognition database to provide learning data for machine learning classification and recognition. The method for constructing an HRRP recognition database based on prototype selection of the present invention selects the most valuable samples from the HRRP database to form a final radar HRRP prototype set when there are too many samples in the radar HRRP recognition database. The final radar HRRP prototype set serves as the HRRP recognition database to provide the most valuable learning data for subsequent machine learning classification and recognition, thereby saving storage resources and computing resources.

Description

Technical Field

[0001] The present invention belongs to the field of radar technology, and in particular relates to a method for constructing an HRRP recognition database based on prototype selection. Background Art

[0002] With the rapid development of radar software and hardware technology, the amount of radar high-resolution range profile (HRRP) data has surged unprecedentedly. Faced with such a massive amount of data, radar HRRP target recognition technology, centered around machine learning, particularly deep learning, has made significant progress. However, this quantitative expansion of HRRP data does not necessarily translate into qualitative improvements. While massive amounts of data provide ample room for data-driven machine learning methods to extract valuable information, the high dimensionality, excessive redundancy, and high noise characteristics of this complex data not only lead to a significant waste of storage and computing resources but also significantly increase the complexity of learning algorithms. Furthermore, this overwhelming amount of data can obscure truly valuable information, leading to a deterioration in machine learning performance. Summary of the Invention

[0003] In order to solve the above problems existing in the prior art, the present invention provides a method for constructing an HRRP recognition database based on prototype selection. The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0004] The present invention provides a method for constructing an HRRP recognition database based on prototype selection, comprising:

[0005] Step 1: Extract multiple radar HRRP data from the radar echo database to build an initial HRRP database;

[0006] Step 2: Using the probability distribution strategy and the constructed adaptive second-order optimizer, prototype selection is performed on the radar HRRP data in the initial HRRP database to obtain the final radar HRRP prototype set;

[0007] Step 3: Use the final radar HRRP prototype set as the HRRP recognition database to provide learning data for machine learning classification and recognition.

[0008] In one embodiment of the present invention, the plurality of radar HRRP data include a plurality of radar HRRP data of different identification categories.

[0009] In one embodiment of the present invention, step 2 includes:

[0010] Step 2.1: Initialize the size of the radar HRRP prototype set to K, assign a weight to each radar HRRP data sample in the initial HRRP database according to the initial sampling probability, and obtain the initial radar HRRP prototype set;

[0011] Step 2.2: Input the radar HRRP prototype set into a deep neural network, train the deep neural network until the network converges, and use the constructed adaptive second-order optimizer to update the network parameters during the training process;

[0012] Step 2.3: Input all radar HRRP data samples in the initial HRRP database into the converged deep neural network to obtain loss values, and perform backpropagation to update the sampling probability;

[0013] Step 2.4: assigning a weight to each radar HRRP data sample in the initial HRRP database according to the updated sampling probability to obtain a new radar HRRP prototype set;

[0014] Step 2.5: Repeat steps 2.2 to 2.4 until the preset number of iterations is reached, and use the updated sampling probability obtained in the last iteration to assign a weight to each radar HRRP data sample in the initial HRRP database to obtain the final radar HRRP prototype set.

[0015] In one embodiment of the present invention, the initial sampling probability of the weight of each radar HRRP data sample is D represents the size of the initial HRRP database.

[0016] In one embodiment of the present invention, the weight m of each radar HRRP data sample is i ∈{0,1},m i =1 means that radar HRRP data sample i is selected into the prototype set, m i =0 means that radar HRRP data sample i is not selected into the prototype set;

[0017] Among them, each weight m i Expressed as a Bernoulli random variable with probability s i and 1-s i Take 1 and 0 as m i The distribution function of the weight m of all radar HRRP data samples is:

[0018]

[0019] Where s i represents the sampling probability of radar HRRP data sample i, and n represents the size of the initial HRRP database.

[0020] In one embodiment of the present invention, in step 2.2, updating the network parameters using the constructed adaptive second-order optimizer during the training process includes the following steps:

[0021] Step a: Input the radar HRRP prototype set into a deep neural network and use the back propagation algorithm to obtain the first-order gradient of the network parameters;

[0022] Step b: Use Rademacher distribution to generate a random vector with the same dimension as the first-order gradient, and use the following formula to calculate the diagonal of the second-order gradient of the network parameters:

[0023]

[0024] Where θ represents the network parameter, g represents the first-order gradient of the network parameter, D represents the diagonal of the second-order gradient of the network parameter, H represents the second-order gradient of the network parameter, diag() represents the diagonal matrix function, ⊙ represents the element-by-element point multiplication, z represents a random vector, and T represents the transpose;

[0025] Step c: Use the momentum method to process the diagonals of the first-order gradient and second-order gradient of the network parameters:

[0026]

[0027]

[0028] Where m t represents the first-order moment estimate of the t-th network update, v t represents the second-order moment estimate of the t-th network update, β1 = 0.9, β2 = 0.999, k = 1, and t represents the back propagation process of the t-th network;

[0029] Step d: Update the network parameters using the following formula:

[0030] θ t+1 =θ t -ηm t / v t ;

[0031] Where η represents the learning rate, and its value is 0.01.

[0032] In one embodiment of the present invention, in step 2.3, the sampling probability is back-propagated and updated according to the following formula:

[0033]

[0034] Where s k+1 represents the sampling probability after the k+1th iteration, s krepresents the sampling probability after the kth iteration, η represents the learning rate, its value is 0.01, L represents the loss value, p(m|s k ) represents the distribution function of the weight m of all radar HRRP data samples.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] The prototype selection-based HRRP recognition database construction method of the present invention utilizes probability distribution and an adaptive second-order optimizer when there are too many samples in the radar HRRP database. Compared with traditional methods, it can ensure lower computational complexity while mining the most valuable information in the HRRP data to generate a final radar HRRP prototype set. The final radar HRRP prototype set serves as the HRRP recognition database, providing the most valuable learning data for subsequent machine learning classification and recognition. On the one hand, it can save storage resources and computing resources, and on the other hand, it avoids the problem of deterioration of machine learning performance due to the complexity of learning data.

[0037] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following preferred embodiments are specifically cited and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 Schematic diagram of a method for constructing an HRRP identification database based on prototype selection provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the following describes in detail a method for constructing an HRRP identification database based on prototype selection proposed by the present invention in conjunction with the accompanying drawings and specific embodiments.

[0040] The aforementioned and other technical contents, features, and effects of the present invention are clearly presented in the following detailed description of the specific embodiments in conjunction with the accompanying drawings. Through the description of the specific embodiments, a deeper and more specific understanding of the technical means and effects adopted by the present invention to achieve the intended purpose can be obtained. However, the accompanying drawings are provided for reference and illustration purposes only and are not intended to limit the technical solutions of the present invention.

[0041] Example 1

[0042] See Figure 1 , Figure 1FIG. 1 is a schematic diagram of a method for constructing an HRRP identification database based on prototype selection provided by an embodiment of the present invention. As shown in the figure, the method for constructing an HRRP identification database based on prototype selection in this embodiment includes:

[0043] Step 1: Extract multiple radar HRRP data from the radar echo database to build an initial HRRP database;

[0044] In an optional embodiment, the plurality of radar HRRP data include a plurality of radar HRRP data of different identification categories.

[0045] Exemplarily, N radar HRRP data are extracted from radar echo databases of ten types of civil aircraft to construct an initial HRRP database, where N is an integer greater than 60,000, and the number of radar HRRP data of each type of civil aircraft is an integer greater than 6,000.

[0046] Step 2: Use the probability distribution strategy and the constructed adaptive second-order optimizer to select the prototypes of the radar HRRP data in the initial HRRP database to obtain the final radar HRRP prototype set;

[0047] In an optional embodiment, step 2 includes:

[0048] Step 2.1: Initialize the size of the radar HRRP prototype set to K, assign a weight to each radar HRRP data sample in the initial HRRP database according to the initial sampling probability, and obtain the initial radar HRRP prototype set;

[0049] In an optional embodiment, the initial sampling probability of the weight of each radar HRRP data sample is D represents the size of the initial HRRP database.

[0050] The weight m of each radar HRRP data sample i ∈{0,1},m i =1 means that radar HRRP data sample i is selected into the prototype set, m i =0 indicates that radar HRRP data sample i is not selected into the prototype set.

[0051] Among them, each weight m i Expressed as a Bernoulli random variable with probability s i and 1-s i Take 1 and 0 as m i The value of , each element in the weight m of all radar HRRP data samples is independent and identically distributed, so the distribution function of m is:

[0052]

[0053] Where s i represents the sampling probability of radar HRRP data sample i, and n represents the size of the initial HRRP database.

[0054] Step 2.2: Input the radar HRRP prototype set into the deep neural network and train the deep neural network until the network converges. During the training process, the network parameters are updated using the constructed adaptive second-order optimizer;

[0055] In an optional embodiment, updating the network parameters using the constructed adaptive second-order optimizer during the training process includes the following steps:

[0056] Step a: Input the radar HRRP prototype set into the deep neural network and use the back propagation algorithm to obtain the first-order gradient of the network parameters;

[0057] Step b: Use Rademacher distribution to generate a random vector with the same dimension as the first-order gradient, and use the following formula to calculate the diagonal of the second-order gradient of the network parameters:

[0058]

[0059] Where θ represents the network parameter, g represents the first-order gradient of the network parameter, D represents the diagonal of the second-order gradient of the network parameter, H represents the second-order gradient of the network parameter, diag() represents the diagonal matrix function, ⊙ represents the element-by-element point multiplication, z represents a random vector, and T represents the transpose;

[0060] Step c: Use the momentum method to process the diagonals of the first-order gradient and second-order gradient of the network parameters:

[0061]

[0062]

[0063] Where m t represents the first-order moment estimate of the t-th network update, v t represents the second-order moment estimate of the t-th network update, β1 = 0.9, β2 = 0.999, k = 1, and t represents the back propagation process of the t-th network;

[0064] Step d: Update the network parameters using the following formula:

[0065] θ t+1 =θ t -ηm t / v t (5);

[0066] Where η represents the learning rate, and its value is 0.01.

[0067] In this embodiment, when training the deep neural network, a training number is preset. When the preset training number is reached, the network is considered to have converged.

[0068] Step 2.3: Input all radar HRRP data samples in the initial HRRP database into the converged deep neural network to obtain the loss value and perform backpropagation to update the sampling probability;

[0069] In this embodiment, the sampling probability is back-propagated and updated according to the following formula:

[0070]

[0071] Where s k+1 represents the sampling probability after the k+1th iteration, s k represents the sampling probability after the kth iteration, η represents the learning rate, its value is 0.01, L represents the loss value, p(m|s k ) represents the distribution function of the weight m of all radar HRRP data samples. In this embodiment, the loss value is calculated by the cross entropy loss function.

[0072] Step 2.4: Assign a weight to each radar HRRP data sample in the initial HRRP database according to the updated sampling probability to obtain a new radar HRRP prototype set;

[0073] Step 2.5: Repeat steps 2.2 to 2.4 until the preset number of iterations is reached. Use the updated sampling probability obtained in the last iteration to assign a weight to each radar HRRP data sample in the initial HRRP database to obtain the final radar HRRP prototype set.

[0074] In this embodiment, the preset number of iterations is 1000.

[0075] Step 3: Use the final radar HRRP prototype set as the HRRP recognition database to provide learning data for machine learning classification and recognition.

[0076] In this embodiment, the obtained final radar HRRP prototype set is used to train the deep neural network until the network converges. The trained deep neural network can then be used to classify and identify the radar HRRP data to be tested to obtain civil aircraft identification results.

[0077] When there are too many samples in the radar HRRP database, the method of the present invention utilizes probability distribution and an adaptive second-order optimizer. Compared with traditional methods, it can ensure lower computational complexity while ensuring that the most valuable information in the HRRP data is mined to generate a final radar HRRP prototype set. The final radar HRRP prototype set serves as an HRRP recognition database, providing the most valuable learning data for subsequent machine learning classification and recognition. On the one hand, it can save storage resources and computing resources, and on the other hand, it avoids the problem of deterioration of machine learning performance due to the complexity of learning data.

[0078] Example 2

[0079] This embodiment further illustrates the effect of the method of embodiment 1 of the present invention through specific experiments.

[0080] 1. Experimental conditions

[0081] The physical parameters of ten types of civil aircraft are shown in Table 1. The sizes of the prototype sets are set to 1%, 10%, and 50% of the initial HRRP database size. M radar HRRP data sets are extracted from the HRRP data of ten types of civil aircraft received by the radar as the test dataset, where M is an integer greater than 0.

[0082] Table 1 Physical parameters of ten types of civil aircraft

[0083]

[0084] 2. Experimental Content

[0085] Based on the above aircraft information, three prototype sets were selected from the initial HRRP database using the method of the present invention and the traditional method. These sets were input into the deep neural network for training. The test dataset was then fed into the trained deep neural network to obtain three sets of recognition results using the method of the present invention and three sets of recognition results using the traditional method. The results are shown in Table 2.

[0086] Table 2 Comparison of classification accuracy between traditional method and the method of the present invention

[0087]

[0088] 3. Results Analysis

[0089] From the perspective of classification accuracy, the prototype set extracted by the present invention is superior to the traditional method.

[0090] The prototype selection-based HRRP recognition database construction method of the present invention assigns a sampling probability to each radar HRRP data sample and searches for the prototype set from a global scope, instead of using a greedy search algorithm like the traditional method, which only focuses on the local scope and causes erroneous search. At the same time, the present invention uses the proposed adaptive second-order optimizer to optimize the sampling probability and the network. Compared with the first-order optimization method, it can converge to the optimal point faster and better to reduce the calculation time.

[0091] It should be noted that, in this document, the terms "comprises," "comprising," or any other variations thereof are intended to encompass non-exclusive inclusion, such that an article or device comprising a list of elements includes not only those elements but also other elements not explicitly listed. Without further limitation, an element defined by the phrase "comprising a ..." does not preclude the presence of additional identical elements in the article or device comprising the element. The terms "connected," "connected," and the like are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect.

[0092] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. A method for constructing an HRRP recognition database based on prototype selection, characterized in that: include: Step 1: Extract multiple radar HRRP data from the radar echo database to build an initial HRRP database; Step 2: Prototype selection is performed on the radar HRRP data in the initial HRRP database using a probability distribution strategy and a constructed adaptive second-order optimizer to obtain a final radar HRRP prototype set; Step 2 includes: Step 2.1: Initialize the size of the radar HRRP prototype set to K, assign a weight to each radar HRRP data sample in the initial HRRP database according to the initial sampling probability, and obtain the initial radar HRRP prototype set; Step 2.2: Input the radar HRRP prototype set into a deep neural network, train the deep neural network until the network converges, and use the constructed adaptive second-order optimizer to update the network parameters during the training process; in step 2.2, using the constructed adaptive second-order optimizer to update the network parameters during the training process includes the following steps: Step a: Input the radar HRRP prototype set into a deep neural network and use the back propagation algorithm to obtain the first-order gradient of the network parameters; Step b: Use Rademacher distribution to generate a random vector with the same dimension as the first-order gradient, and use the following formula to calculate the diagonal of the second-order gradient of the network parameters: Where θ represents the network parameter, g represents the first-order gradient of the network parameter, D represents the diagonal of the second-order gradient of the network parameter, H represents the second-order gradient of the network parameter, diag() represents the diagonal matrix function, ⊙ represents the element-by-element point multiplication, z represents a random vector, and T represents the transpose; Step c: Use the momentum method to process the diagonals of the first-order gradient and second-order gradient of the network parameters: Where m t represents the first-order moment estimate of the t-th network update, v t represents the second-order moment estimate of the t-th network update, β1 = 0.9, β2 = 0.999, k = 1, and t represents the back propagation process of the t-th network; Step d: Update the network parameters using the following formula: i t+1 =θ t -m t / v t ; Where η represents the learning rate, and its value is 0.01; Step 2.3: Input all radar HRRP data samples in the initial HRRP database into the converged deep neural network to obtain loss values, and perform backpropagation to update the sampling probability; Step 2.4: assigning a weight to each radar HRRP data sample in the initial HRRP database according to the updated sampling probability to obtain a new radar HRRP prototype set; Step 2.5: Repeat steps 2.2 to 2.4 until a preset number of iterations is reached, and use the updated sampling probability obtained in the last iteration to assign a weight to each radar HRRP data sample in the initial HRRP database to obtain the final radar HRRP prototype set; Step 3: Use the final radar HRRP prototype set as the HRRP recognition database to provide learning data for machine learning classification and recognition.

2. The method for constructing an HRRP identification database based on prototype selection according to claim 1, characterized in that: The plurality of radar HRRP data includes a plurality of radar HRRP data of different identification categories.

3. The method for constructing an HRRP identification database based on prototype selection according to claim 1, characterized in that: The initial sampling probability of the weight of each radar HRRP data sample is D represents the size of the initial HRRP database.

4. The method for constructing an HRRP identification database based on prototype selection according to claim 3, characterized in that: The weight m of each radar HRRP data sample i ∈{0,1},m i =1 means that radar HRRP data sample i is selected into the prototype set, m i =0 means that radar HRRP data sample i is not selected into the prototype set; Among them, each weight m i Expressed as a Bernoulli random variable with probability s i and 1-s i Take 1 and 0 as m i The distribution function of the weight m of all radar HRRP data samples is: Where s i represents the sampling probability of radar HRRP data sample i, and n represents the size of the initial HRRP database.

5. The method for constructing an HRRP identification database based on prototype selection according to claim 1, characterized in that: In step 2.3, the sampling probability is back-propagated and updated according to the following formula: Where s k+1 represents the sampling probability after the k+1th iteration, s k represents the sampling probability after the kth iteration, η represents the learning rate, its value is 0.01, L represents the loss value, p(m|s k ) represents the distribution function of the weight m of all radar HRRP data samples.

Images