An open set recognition method for high-resolution radar range profiles based on extreme value distribution

Abstract

The present invention discloses an open-set recognition method for high-resolution range profiles of radar based on extreme value distribution, comprising the following steps: establishing and preprocessing a training sample set and a test sample set; constructing a convolutional neural network; training the convolutional neural network using the training sample set; extracting high-dimensional features of the training sample set using the trained convolutional neural network; calculating the Euclidean distance of the high-dimensional features of the training sample set and fitting the Euclidean distance to an extreme value distribution; extracting high-dimensional features of the test sample set using the trained convolutional neural network; and performing open-set recognition on the high-dimensional features of the test sample set using the cumulative probability distribution function of the extreme value distribution. The method provided by the present invention can not only be used to recognize known category models of targets in a radar target recognition database, but also reject unknown category models of targets from outside the database, which not only improves the accuracy and practicability of the radar target recognition system, but also effectively improves the automation and intelligence level of the radar.

Description

Technical Field

[0001] The present invention belongs to the technical field of radar, and particularly relates to a method for open set recognition of high-resolution range profiles of radar based on extreme value distribution. Background Technique

[0002] The range resolution of a radar is proportional to the received pulse width after matched filtering, and the range cell length of the radar transmitted signal satisfies: ΔR = cτ / 2 = c / 2B, where ΔR is the range cell length of the radar transmitted signal, c is the speed of light, τ is the pulse width of the matched reception, and B is the bandwidth of the radar transmitted signal. A large radar transmitted signal bandwidth provides high range resolution. In fact, the level of radar range resolution is relative to the observed target. When the size of the observed target along the radar line of sight is L, if L << ΔR, the width of the corresponding radar echo signal is approximately the same as the radar transmitted pulse width (the received pulse after matched processing), which is usually called the "point" target echo, and such a radar is a low-resolution radar. If ΔR << L, the target echo becomes a "one-dimensional range profile" that extends in range according to the target characteristics, and such a radar is a high-resolution radar (<< means much less than).

[0003] A high-resolution radar transmits broadband coherent signals (linear frequency modulation or stepped frequency signals), and the radar receives echo data through the backscattering of the target to the transmitted electromagnetic wave. Usually, the echo characteristics are calculated using a simplified scattering point model, that is, the first-order Born approximation that ignores multiple scattering is adopted. The fluctuations and spikes presented in the high-resolution radar echo reflect the distribution of the radar cross-section (RCS) of the scatterers (such as the nose, wings, tail rudder, air intake, engine, etc.) on the target along the radar line of sight (RLOS) at a certain radar viewing angle, and reflect the relative geometric relationship of the scatterers in the radial direction, which is often called the high-resolution range profile (HRRP, high resolution range profile). Therefore, the HRRP samples contain important structural features of the target and are valuable for target recognition and classification.

[0004] Traditional target recognition methods for high-resolution range profile data mainly use support vector machines to directly classify the targets, or use feature extraction methods based on restricted Boltzmann machines to project the data into a high-dimensional space first and then use a classifier to classify the data. However, the above methods only utilize the time-domain features of the signals, and the target recognition accuracy is not high.

[0005] In recent years, the target recognition methods for radar high-resolution range profile (HRRP) data have mainly focused on closed-set recognition. This type of recognition requires that the data category models in the test sample set be consistent with those in the training sample set. However, in practical applications, in addition to capturing the HRRPs of targets in the target recognition database, radar also captures the HRRPs of many unknown category model targets outside the database. In this case, the existing closed-set recognition algorithms cannot reject the target data of unknown category models outside the library, but instead misclassify them as a certain category model target in the database, which greatly reduces the accuracy and practicality of radar target recognition.

[0006] Therefore, some researchers have started to study the open-set recognition of radar HRRP. For example, Chai Jing et al. proposed a multi-kernel support vector domain description (Multi kernel SVDD) model based on support vector domain description (SVDD) to more flexibly describe the multimodal distribution of HRRP data in the high-dimensional feature space, thereby improving the recognition and rejection performance of radar HRRP. Zhang Xuefeng et al. proposed a multi-classifier fusion algorithm based on the maximum correlation classifier (MCC), support vector machine (SVM), and relevance vector machine (RVM) to achieve the rejection and recognition functions of radar HRRP. However, both of the above two algorithms need to rely on specific forms of kernel functions to extract features, which limits the ability of the model to extract sufficient separable features, thus affecting the accuracy of target recognition and the intelligent level of radar. Summary of the Invention

[0007] To solve the above problems existing in the prior art, the present invention provides an open-set recognition method for radar high-resolution range profile based on extreme value distribution. The technical problems to be solved by the present invention are realized through the following technical solutions:

[0008] The first aspect of the embodiment of the present invention provides an open-set recognition method for radar high-resolution range profile based on extreme value distribution, including the following steps:

[0009] Establish a first training sample set and a first test sample set; wherein, the training sample set includes the radar high-resolution range profiles of several targets with known category models, and the test sample set includes the radar high-resolution range profiles of several targets with known category models and the radar high-resolution range profiles of targets with unknown categories outside the radar target recognition database;

[0010] Preprocess the radar high-resolution range images in the training sample set and the test sample set to obtain a second training sample set and a second test sample set;

[0011] Construct a convolutional neural network;

[0012] Train the convolutional neural network using the second training sample set to obtain a trained convolutional neural network;

[0013] Extract the high-dimensional features of the second training sample set using the trained convolutional neural network;

[0014] Calculate the feature center of the high-dimensional features of the second training sample set and the Euclidean distance from each high-dimensional feature to the corresponding feature center, and perform extreme value distribution fitting on all the Euclidean distances to obtain an extreme value distribution cumulative probability distribution function;

[0015] Extract the high-dimensional features of the second test sample set using the trained convolutional neural network;

[0016] Perform open-set recognition on the high-dimensional features of the second test sample set using the extreme value distribution cumulative probability distribution function to obtain the recognition result of the second test sample set.

[0017] In an embodiment of the present invention, the preprocessing of the radar high-resolution range images in the training sample set and the test sample set to obtain a preprocessed second training sample set and a second test sample set includes:

[0018] Perform centroid alignment and normalization processing on the radar high-resolution range images in the training sample set and the test sample set in sequence to obtain a preprocessed second training sample set and a second test sample set.

[0019] In an embodiment of the present invention, the convolutional neural network includes: three convolutional layers and a fourth fully connected layer; the three convolutional layers are the first convolutional layer, the second convolutional layer, and the third convolutional layer; the

[0020] The convolution step size of each convolutional layer is the same; each convolutional layer includes a number of convolutional kernels, and the size of each convolutional kernel is the same;

[0021] Among them, the expression of the loss function of the convolutional neural network is:

[0022]

[0023] Among them, Θ(x) and Θ(x k ) are the output results of the convolutional neural network, O i(i = 1, …, k, …, N) is the i-th prototype randomly initialized according to a Gaussian distribution; d(Θ(x), O k ) is the Euclidean distance from Θ(x) to O k , and λ is a hyperparameter.

[0024] In an embodiment of the present invention, training the convolutional neural network using the second training sample set to obtain a trained convolutional neural network includes:

[0025] Randomly divide the sample data of the second training sample set into q batches, and the data of each batch is an n×D-dimensional matrix data; where, floor() represents rounding down, and P represents the number of high-resolution range images in the second training sample set;

[0026] Sequentially input the sample data of each batch into the convolutional neural network for processing to obtain the output result of the convolutional neural network;

[0027] Calculate the value of the loss function according to the output result of the convolutional neural network and the loss function of the convolutional neural network, and update the parameter values of the convolutional neural network using the stochastic gradient method until the network converges to obtain a trained convolutional neural network.

[0028] In an embodiment of the present invention, the sequentially inputting the sample data of each batch into the convolutional neural network for processing to obtain the output result of the convolutional neural network includes:

[0029] Use the first convolutional layer to perform convolution and downsampling processing on the input sample data of the current batch to obtain a first feature map;

[0030] Use the second convolutional layer to perform convolution and downsampling processing on the first feature map to obtain a second feature map;

[0031] Use the third convolutional layer to perform convolution and downsampling processing on the second feature map to obtain a third feature map;

[0032] Use the fourth fully connected layer to perform non-linear transformation processing on the third feature map to obtain the output result of the current sample data;

[0033] Repeat the above steps until the processing of the sample data of q batches is completed to obtain the output result of the convolutional neural network.

[0034] In an embodiment of the present invention, the expression of the feature center of the high-dimensional features of the second training sample set is:

[0035]

[0036] Where, represents the i-th sample in the k-th category in the second training sample set, and the k-th category has a total of N k samples;

[0037] The cumulative probability distribution function expression of the extreme value distribution of the kth category is:

[0038]

[0039] in, represents the cumulative probability distribution function of extreme value distribution, t represents the Euclidean distance from each high-dimensional feature to the corresponding feature center, which is the function independent variable; ξ, α and β represent the parameters of the cumulative probability distribution function of extreme value distribution.

[0040] In one embodiment of the present invention, the step of performing open set recognition on the high-dimensional features of the second test sample set using the extreme value distribution cumulative probability distribution function to obtain a recognition result of the second test sample set includes:

[0041] when When it is established, the test samples corresponding to the high-dimensional features of the second test sample set are unknown categories outside the database;

[0042] when If it is not true, the category of the test sample corresponding to the high-dimensional feature of the second test sample set is

[0043] Among them, τ is the preset decision threshold.

[0044] Beneficial effects of the present invention:

[0045] 1. The radar high-resolution range profile open set recognition method provided by the present invention uses a convolutional neural network to combine the primary features of each layer, thereby obtaining higher-level features for recognition, so the recognition rate is significantly improved. It can not only be used to identify and classify known category targets in the library, but also reject unknown category targets outside the library, thereby improving the target recognition accuracy, thereby improving the automation and intelligence level of the radar;

[0046] 2. The present invention adopts a multi-layer convolutional neural network structure and performs energy normalization and alignment preprocessing on the data, so as to mine the high-level features of the high-resolution range image data and remove the amplitude sensitivity, translation sensitivity and attitude sensitivity of the radar high-resolution range image data. Compared with the traditional direct classification method, it has stronger robustness.

[0047] 3. In the decision-making stage, the present invention introduces the extreme value distribution established for the target features of known categories in the library, and determines the specific category of the target to be measured through the cumulative distribution function of the extreme value distribution. Compared with the previous distance metric-based methods, this strategy can effectively improve the recognition rate of samples at the edge of known category clusters (i.e., extreme value samples, which are often difficult to correctly identify in the recognition task), and has strong recognition robustness.

[0048] The following will further elaborate on the present invention in conjunction with the drawings and embodiments. Description of the Drawings

[0049] Figure 1 is a schematic flow chart of an open-set recognition method for radar high-resolution range profiles based on extreme value distribution provided by an embodiment of the present invention;

[0050] Figure 2 is the simulation test result provided by an embodiment of the present invention. Detailed Embodiments

[0051] The following further describes the present invention in detail with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0052] Embodiment 1

[0053] As Figure 1 shown, an open-set recognition method for radar high-resolution range profiles based on extreme value distribution includes the following steps:

[0054] Step 10, establish a first training sample set and a first test sample set; wherein, the training sample set includes the radar high-resolution range profiles of several targets of known category models, and the test sample set includes the radar high-resolution range profiles of several targets of known category models and the radar high-resolution range profiles of targets of unknown categories outside the radar target recognition database;

[0055] Step 20, preprocess the radar high-resolution range profiles in the training sample set and the test sample set to obtain a second training sample set and a second test sample set;

[0056] Step 30, construct a convolutional neural network;

[0057] Step 40, use the second training sample set to train the convolutional neural network to obtain a trained convolutional neural network;

[0058] Step 50, use the trained convolutional neural network to extract the high-dimensional features of the second training sample set;

[0059] Step 60: Calculate the feature center of the high-dimensional features of the second training sample set and the Euclidean distance from each high-dimensional feature to the corresponding feature center, and perform extreme value distribution fitting on all the Euclidean distances to obtain the extreme value distribution cumulative probability distribution function;

[0060] Step 70: Use the trained convolutional neural network to extract the high-dimensional features of the second test sample set;

[0061] Step 80: Use the extreme value distribution cumulative probability distribution function to perform open-set recognition on the high-dimensional features of the second test sample set to obtain the recognition result of the second test sample set.

[0062] In this embodiment, it can not only be used to identify and classify known-category targets in the library, but also reject unknown-category targets outside the library, improving the target recognition accuracy, and further enhancing the automation and intelligence level of the radar.

[0063] Embodiment 2

[0064] An open-set recognition method for radar high-resolution range profiles based on extreme value distribution, comprising the following steps:

[0065] Step 100: Establish a first training sample set and a first test sample set; wherein, the training sample set includes radar high-resolution range profile sample data of several known-category model targets, and the test sample set includes radar high-resolution range profile sample data of several known-category model targets and radar high-resolution range profile sample data of unknown-category targets outside the radar target recognition database.

[0066] Step 100 specifically includes:

[0067] Step 101: Obtain P radar high-resolution range profile raw data of N categories as the first training sample set, where N≥3 and P≥900;

[0068] Step 102: Obtain Q radar high-resolution range profile raw data of N categories and R radar high-resolution range profile raw data of M unknown categories as the second test sample set, where Q≥900, M≥1, and L≥300.

[0069] The P radar high-resolution range profile raw data and the Q radar high-resolution range profile raw data are both data of known categories in the database, and the L radar high-resolution range profile raw data are data of unknown categories outside the database.

[0070] Step 200: Preprocess the radar high-resolution range profile sample data in the training sample set and the test sample set to obtain a second training sample set and a second test sample set.

[0071] In this embodiment, the original data in the first training sample set and the first test sample set are sequentially subjected to centroid alignment and normalization processing to obtain the preprocessed training sample set and test sample set, that is, the second training sample set and the second test sample set are obtained after preprocessing.

[0072] Specifically, the original data in the first training sample set or the first test sample set is denoted as x0. First, the original data x0 is subjected to centroid alignment to obtain the data x′0 after centroid alignment; then, the data x′0 after centroid alignment is subjected to two-norm normalization processing to obtain the sample data x after normalization processing, and its expression is:

[0073]

[0074] where x can represent the sample data of the second training sample set x train or the sample data of the second test sample set x test The second training sample set and the second test sample set are P×D and (Q + R)×D-dimensional matrices respectively, where D represents the total number of range cells included in the original data of a radar high-resolution range profile.

[0075] Step 300, construct a convolutional neural network.

[0076] In this embodiment, it is set that the convolutional neural network includes three convolutional layers and one fully connected layer, which are denoted as the first convolutional layer, the second convolutional layer, the third convolutional layer, and the fourth fully connected layer respectively. Among them, each convolutional layer has the same convolutional stride, and each convolutional layer includes a number of convolutional kernels, and the size of each convolutional kernel is the same.

[0077] Specifically, for the first convolutional layer:

[0078] It is set to include C convolutional kernels, and the C convolutional kernels of the first convolutional layer are denoted as K, and the size of K is set to 1×w×1, where w represents the window of each convolutional kernel in the first convolutional layer, 1 < w < D; C is a positive integer greater than 0; the convolutional stride of the first convolutional layer is set to L; at the same time, the kernel window size of the downsampling process of the first convolutional layer is set to m×m, 1 < m < D, D represents the total number of range cells included in a radar high-resolution range imaging data in the second training sample set, and m is a positive integer greater than 0; the stride of the downsampling process of the first convolutional layer is set to I, and the value of I is equal to that of m.

[0079] The activation function of the first convolutional layer is set to x represents the preprocessed sample data (the sample data of the second training sample set or the second test sample set), represents the convolution operation, and b represents the all-1 bias of the third convolutional layer.

[0080] For the second convolutional layer:

[0081] Set it to include C' convolutional kernels, and denote the C' convolutional kernels of the second convolutional layer as K'. The size of K' is the same as that of the convolutional kernel K of the first convolutional layer; denote the convolutional stride of the second convolutional layer as L', where w ≤ L' ≤ D - w, and L' has the same value as the convolutional stride L of the first convolutional layer; at the same time, set the kernel window size of the downsampling process of the second convolutional layer to be m'×m', where 1 < m' < D and m' is a positive integer greater than 0; the stride of the second downsampling process is I', and I' has the same value as m'.

[0082] Set the activation function of the second convolutional layer to be represents the first feature map output by the first convolutional layer, represents the convolutional operation, and b' represents the all-ones bias of the second convolutional layer.

[0083] For the third convolutional layer:

[0084] Set it to include C″ convolutional kernels, and let the C″ convolutional kernels of the third convolutional layer be K″. The size of K″ is the same as that of each convolutional kernel window of the second convolutional layer; set the convolutional stride of the third convolutional layer to be L″, and it has the same value as the convolutional stride L' of the second convolutional layer; at the same time, set the kernel window size of the downsampling process of the third convolutional layer to be m″×m″, where 1 < m″ < D and m″ is a positive integer greater than 0; the stride of the third downsampling process is I″, and I″ has the same value as m″.

[0085] Set the activation function of the second convolutional layer to be represents the second feature map output by the second convolutional layer, represents the convolutional operation, and b″ represents the all-ones bias of the third convolutional layer.

[0086] For the fourth fully connected layer:

[0087] Set its randomly initialized weight matrix to be a B×U-dimensional matrix, floor() represents rounding down, D represents the total number of range cells included in a radar high-resolution range imaging data in the second training sample, B ≥ D, and B is a positive integer greater than 0; set the activation function to be represents the third feature map output by the third convolutional layer, represents the all-ones bias of the fourth fully connected layer, and is a U×1-dimensional.

[0088] After constructing the model of the convolutional neural network, it further includes constructing the loss function of the convolutional neural network, and its expression is:

[0089]

[0090] where Θ(x) and Θ(x k ) are the output results of the convolutional neural network, and x k represents the training sample data of the k-th category; O i (i = 1, …, k, …, N) are the i-th prototypes randomly initialized according to the Gaussian distribution, with a total of N; d(Θ(x), O k ) is the Euclidean distance from Θ(x) to O k , and λ is a hyperparameter.

[0091] Step 400: Train the convolutional neural network using the second training sample set to obtain the trained convolutional neural network, specifically including:

[0092] Step 401: Randomly divide the sample data of the second training sample set into q batches, and the data of each batch is an n×D-dimensional matrix data; where, floor() represents rounding down, and P represents the number of high-resolution range images in the second training sample set.

[0093] Step 402: Input the sample data of each batch into the convolutional neural network for processing in sequence to obtain the output results of the convolutional neural network. Specifically, Step 402 includes Steps 402-1 to 402-5:

[0094] Step 402-1: After inputting the sample data into the convolutional neural network, use the first convolutional layer to perform convolution and downsampling processing on the currently input sample data to obtain the first feature map.

[0095] Specifically, use the convolution stride L of the first convolutional layer to perform convolution on the input sample data x with the C convolutional kernels of the first convolutional layer respectively to obtain the C convolution results of the first convolutional layer, and denote them as the C feature maps y of the first convolutional layer:

[0096]

[0097] Perform Gaussian normalization processing on the C feature maps y of the first convolutional layer to obtain the C feature maps of the first convolutional layer after Gaussian normalization processing

[0098] Perform downsampling processing on each feature map in respectively to obtain the C feature maps after downsampling processing of the first convolutional layer That is, the first feature map, denoted as:

[0099]

[0100] Among them, represents the maximum value of C feature maps of the first convolutional layer after Gaussian normalization within the kernel window size m×m of the first-layer downsampling process of, represents C feature maps of the first convolutional layer after Gaussian normalization.

[0101] Step 402-2: Use the second convolutional layer to perform convolution and downsampling on the first feature map to obtain a second feature map.

[0102] Specifically, use the convolution stride L′ of the second convolutional layer to convolve the C feature maps (i.e., the first feature map) after the downsampling process of the first convolutional layer with the C′ convolutional kernels K′ of the second convolutional layer respectively, to obtain C′ convolution results of the second convolutional layer, and denote them as C′ feature maps of the second convolutional layer

[0103] Perform Gaussian normalization on the C′ feature maps of the second convolutional layer to obtain C′ feature maps of the second convolutional layer after Gaussian normalization

[0104] For each of the feature maps in, perform downsampling respectively, and further obtain C′ feature maps after the downsampling process of the second convolutional layer i.e., the second feature map, denoted as:

[0105]

[0106] Among them, represents the maximum value of C′ feature maps of the second convolutional layer after Gaussian normalization within the kernel window size m′×m′ of the second-layer downsampling process of, represents C′ feature maps of the second convolutional layer after Gaussian normalization.

[0107] Step 402-3: Use the third convolutional layer to perform convolution and downsampling on the second feature map to obtain a third feature map.

[0108] Specifically, use the convolution stride L′ of the third convolutional layer to convolve the C′ feature maps (i.e., the second feature map) after the downsampling process of the second convolutional layer with the C″ convolutional kernels K″ of the third convolutional layer respectively, to obtain C″ convolution results of the third convolutional layer, and denote them as C″ feature maps of the third convolutional layer

[0109] For the C″ feature maps of the third convolutional layer Perform Gaussian normalization to obtain the C″ feature maps of the third convolutional layer after Gaussian normalization

[0110] For Each of the feature maps in is separately downsampled to obtain the C″ feature maps of the third convolutional layer after downsampling That is, the third feature map, denoted as:

[0111]

[0112] Wherein, Represents taking the maximum value of the C″ feature maps of the third convolutional layer after Gaussian normalization within the kernel window size m″×m″ of the third layer downsampling Of, Represents the C″ feature maps of the third convolutional layer after Gaussian normalization

[0113] Step 402-4, Use the fourth fully connected layer to perform non-linear transformation on the third feature map to obtain the processing result of the current sample data Its expression is:

[0114]

[0115] Wherein, Represents the randomly initialized weight matrix of the fourth fully connected layer Represents the all-ones bias of the fourth fully connected layer

[0116] Step 402-5, Repeat the above steps 402-1 to 402-4 until the processing of all batches of input data is completed to obtain the output result of the convolutional neural network, and this output result is also called the high-dimensional feature

[0117] Step 403, Calculate the value of the loss function according to the output result of the convolutional neural network, and use the stochastic gradient method to update the parameter values of the convolutional neural network until the network converges to obtain the trained convolutional neural network

[0118] Specifically, substitute the obtained in the above step 402 As the output result of the convolutional neural network into the above loss function expression of the convolutional neural network to obtain the value of the loss function, and use the existing stochastic gradient method to update the parameter values of the convolutional neural network until the network converges to obtain the trained convolutional neural network. Regarding the stochastic gradient method here, it belongs to the mature existing technology and will not be specifically introduced in this embodiment

[0119] In this embodiment, by adopting a multi-layer convolutional neural network structure and performing preprocessing of two-norm normalization and alignment on the data, high-level features of the high-resolution range image data can be mined, and the amplitude sensitivity, translation sensitivity, and pose sensitivity of the high-resolution range image data are removed. Compared with the traditional direct classification method, it has stronger robustness.

[0120] Step 500, use the trained convolutional neural network to extract the high-dimensional features Θ(x train ) of the second training sample set x train ) according to steps 401 - 402.

[0121] Step 600, calculate the feature center of the high-dimensional features of the second training sample set and the Euclidean distance from each high-dimensional feature to the corresponding feature center, and perform extreme value distribution fitting on all the Euclidean distances to obtain the extreme value distribution cumulative probability distribution function.

[0122] Specifically, the calculation expression of the feature center (where k represents the kth category) of the high-dimensional features of the sample data of each category in the second training sample set is:

[0123]

[0124] where, represents the ith sample in the kth category in the second training sample set, and there are a total of N k samples in this category. Then calculate the Euclidean distance from each high-dimensional feature in each category to the corresponding feature center of that category, and perform extreme value distribution fitting on these Euclidean distances. The extreme value distribution cumulative probability distribution function of the kth category obtained by fitting is as follows:

[0125]

[0126] where t is the independent variable of the function, representing the Euclidean distance from the high-dimensional feature to the feature center; ξ, α, and β are the relevant parameters of the extreme value distribution cumulative probability distribution function. In this embodiment, an extreme value distribution established for the target features of known categories in the library is introduced in the decision-making stage, and the specific category of the target to be measured is judged through the cumulative distribution function of the extreme value distribution, which can eliminate the dimensional sensitivity introduced by the distance criterion. Compared with the traditional classification method, it has stronger robustness. In the decision-making stage, an extreme value distribution established for the target features of known categories in the library is introduced, and the specific category of the target to be measured is judged through the cumulative distribution function of the extreme value distribution. This strategy can effectively improve the recognition rate of samples (i.e., extreme value samples, which are often difficult to correctly identify in the recognition task) at the edge of the known category cluster compared with the previous methods based on distance measurement, and has strong recognition robustness.

[0127] Step 700: Use the trained convolutional neural network to extract the high-dimensional features Θ(x test ) of the second test sample set x according to Steps 401 and 402. test )

[0128] Step 800: Use the extreme value distribution cumulative probability distribution function to perform open-set recognition on each high-dimensional feature of the second test sample set to obtain the recognition result of the second test sample set.

[0129] Specifically, calculate the Euclidean distance from the high-dimensional feature Θ(x test ) to the feature center . When holds, it is determined that the test sample corresponding to the high-dimensional feature of the second test sample set is an unknown category outside the database;

[0130] When does not hold, it is determined as a known category in the library, and the category of the test sample corresponding to the high-dimensional feature of the second test sample set is

[0131] where τ represents a preset decision threshold.

[0132] Embodiment III

[0133] An open-set recognition device for radar high-resolution range profiles based on extreme value distribution according to an embodiment of the present invention includes:

[0134] A data acquisition module, configured to establish a first training sample set and a first test sample set; wherein, the training sample set includes radar high-resolution range profiles of several targets of known categories, and the test sample set includes radar high-resolution range profiles of several targets of known categories and radar high-resolution range profiles of targets of unknown categories outside the database;

[0135] A preprocessing module, configured to preprocess the radar high-resolution range profiles in the training sample set and the test sample set to obtain a second training sample set and a second test sample set;

[0136] A model construction module, configured to construct a convolutional neural network;

[0137] A training module, configured to use the second training sample set to train the convolutional neural network to obtain a trained convolutional neural network;

[0138] A first extraction module, configured to use the trained convolutional neural network to extract the high-dimensional features of the second training sample set;

[0139] The extreme value distribution fitting module is used to calculate the feature center of the high-dimensional features of the second training sample set and the Euclidean distance from each high-dimensional feature to the feature center, and perform extreme value distribution fitting on all the Euclidean distances to obtain the extreme value distribution cumulative probability distribution function;

[0140] The second extraction module is used to extract the high-dimensional features of the second test sample set by using the trained convolutional neural network;

[0141] The target recognition module is used to perform open-set recognition on the high-dimensional features of the second test sample set by using the extreme value distribution cumulative probability distribution function to obtain the recognition result of the second test sample set.

[0142] The radar high-resolution range image open-set recognition device provided in this embodiment can implement the radar high-resolution range image open-set recognition method provided in the above Embodiment 1, and the detailed process will not be elaborated here.

[0143] Therefore, the radar high-resolution range image open-set recognition device provided in this embodiment also has the advantages of not only being able to identify and classify known category targets in the library, but also being able to reject unknown category targets outside the library, and having a high target recognition accuracy.

[0144] Embodiment 4

[0145] Next, the beneficial effects of the present invention will be verified and explained through simulation experiments.

[0146] 1. Simulation conditions

[0147] The hardware platform of the simulation experiment in this embodiment is:

[0148] Processor: Intel(R) Core(TM) i9-10980XE, main frequency 3.00GHz, memory 256GB.

[0149] The software platform of the simulation experiment in this embodiment is: Ubuntu 20.04 operating system and python 3.9.

[0150] The data used in this simulation test are high-resolution range image measured data of 13 types of aircraft, including An-26, Cessna, Yak-42, A319, A320, A330-2, A330-3, B737-8, CRJ-900, A321, A350-941, B737-7 and B747-89L. The first three types of aircraft are regarded as known in-library target categories, and the last 10 types of aircraft are regarded as unknown out-library target categories, and training sample sets and test sample sets are produced. Among them, the training sample set has a total of about 50,000 samples, and each in-library category has about 15,000 samples; the test sample set includes 3 known categories in the library, totaling about 30,000 samples, and 10 unknown categories outside the library, totaling 30,000 samples, and each category has about 3,000 samples.

[0151] Before the experiment is conducted, all the original data are pre-processed according to step 200 in the above-mentioned second embodiment, and then the open set recognition experiment is conducted using the present invention.

[0152] 2. Simulation content and result analysis

[0153] This simulation experiment compares the method of the present invention with the traditional two-stage rejection recognition method and the SoftMax threshold method.

[0154] The traditional two-stage rejection recognition method mainly uses SVDD, OCSVM and other methods to reject out-of-library targets, and then uses SVM and other methods to further classify the targets that have been judged as in-library. The SoftMax threshold rule determines the category of the target to be tested based on the final classification output size of the convolutional neural network. If the final network classification output is greater than the specified threshold, it is judged as in-library, otherwise it is judged as out-of-library.

[0155] This simulation experiment uses the area under the receiver operating characteristic curve (ROC) AUC to evaluate the ability of different methods to reject out-of-library targets. The larger the AUC value, the stronger the ability to reject out-of-library targets.

[0156] See also Figure 2 , Figure 2 is a comparison chart of simulation test results provided by an embodiment of the present invention, Figure 2 It can be seen that in this simulation experiment, the method of the present invention has the strongest ability to reject out-of-stock targets, followed by the SoftMax threshold method, and the three traditional methods have average abilities to reject out-of-stock targets.

[0157] Since there are many types of data used in this simulation experiment, the Macro Average F1-Score is used to comprehensively evaluate the open-set recognition ability of different methods. The larger the value of the F1-Score, the stronger the open-set recognition ability. The results of the simulation experiment are shown in the following table.

[0158]

[0159]

[0160] It can be seen that in this simulation experiment, the comprehensive open-set recognition ability of the present invention is the strongest, significantly better than the other three methods.

[0161] In summary, the present invention has achieved the best results both in terms of the rejection ability for out-of-library targets and the comprehensive consideration of the open-set recognition ability, which proves the effectiveness of the present invention.

[0162] The open-set recognition method for radar high-resolution range profiles provided in this embodiment can combine the primary features of each layer by using convolutional neural network technology, so as to obtain higher-level features for recognition. Therefore, the recognition rate has been significantly improved. It can not only be used to recognize and classify known-category targets in the library, but also reject out-of-library unknown-category targets, improving the target recognition accuracy, and further enhancing the automation and intelligence level of the radar.

[0163] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0164] In addition, the terms "first" and "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality" means two or more, unless otherwise specifically defined.

[0165] In the present invention, unless otherwise clearly defined and limited, terms such as "installation", "connection", "coupling", "fixation" and the like shall be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral body; it may be a mechanical connection or an electrical connection; it may be a direct connection or an indirect connection through an intermediate medium, and it may be the communication inside two elements or the interaction relationship between two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0166] In the present invention, unless otherwise clearly defined and limited, the first feature being "above" or "below" the second feature may include the direct contact between the first and second features, or may include the situation where the first and second features are not in direct contact but in contact through additional features therebetween. Moreover, the first feature being "above", "over" and "on top of" the second feature includes that the first feature is directly above and obliquely above the second feature, or merely means that the horizontal height of the first feature is higher than that of the second feature. The first feature being "below", "under" and "beneath" the second feature includes that the first feature is directly below and obliquely below the second feature, or merely means that the horizontal height of the first feature is lower than that of the second feature.

[0167] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic descriptions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine the different embodiments or examples described in this specification.

[0168] The above content is a further detailed description of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is only limited to these descriptions. For those of ordinary skill in the technical field to which the present invention pertains, without departing from the concept of the present invention, several simple deductions or substitutions can still be made, and all of them should be regarded as belonging to the protection scope of the present invention.

Claims

1. An open set recognition method for radar high-resolution range profiles based on extreme value distribution, characterized in that Including the following steps: Establish a first training sample set and a first test sample set; wherein, the training sample set includes radar high-resolution range profiles of a number of targets with known category models, and the test sample set includes radar high-resolution range profiles of a number of targets with known category models and radar high-resolution range profiles of targets with unknown categories outside the radar target recognition database; Preprocess the radar high-resolution range profiles in the training sample set and the test sample set to obtain a second training sample set and a second test sample set; Construct a convolutional neural network; Train the convolutional neural network using the second training sample set to obtain a trained convolutional neural network; Extract high-dimensional features of the second training sample set using the trained convolutional neural network; Calculate the feature centers of the high-dimensional features of the second training sample set and the Euclidean distances from each high-dimensional feature to the corresponding feature center, and perform extreme value distribution fitting on all the Euclidean distances to obtain an extreme value distribution cumulative probability distribution function; Extract high-dimensional features of the second test sample set using the trained convolutional neural network; Perform open-set recognition on the high-dimensional features of the second test sample set using the extreme value distribution cumulative probability distribution function to obtain the recognition results of the second test sample set.

2. The open set recognition method for radar high-resolution range profiles based on extreme value distribution according to claim 1, characterized in that, The preprocessing of the radar high-resolution range profiles in the training sample set and the test sample set to obtain a preprocessed second training sample set and a second test sample set includes: Perform centroid alignment and normalization processing on the radar high-resolution range profiles in the training sample set and the test sample set in sequence to obtain a preprocessed second training sample set and a second test sample set.

3. The open set recognition method for radar high-resolution range profiles based on extreme value distribution according to claim 1, wherein The convolutional neural network includes: three convolutional layers and a fourth fully connected layer; the three convolutional layers are the first convolutional layer, the second convolutional layer, and the third convolutional layer respectively; the The convolutional step size of each convolutional layer is the same; each convolutional layer includes a number of convolutional kernels, and the size of each convolutional kernel is the same; Among them, the expression of the loss function of the convolutional neural network is: Among them, Θ(x) and Θ(x k ) are the output results of the convolutional neural network, and O i (i = 1, …, k, …, N) is the i-th prototype randomly initialized according to the Gaussian distribution; d(Θ(x), O k ) is the Euclidean distance from Θ(x) to O k , and λ is a hyperparameter.

4. A radar high-resolution range profile open set recognition method based on extreme value distribution according to claim 1, characterized in that The training of the convolutional neural network using the second training sample set to obtain a trained convolutional neural network includes: Randomly divide the sample data of the second training sample set into q batches, and the data of each batch is an n×D-dimensional matrix data; among them, floor() represents rounding down, and P represents the number of high-resolution range profiles in the second training sample set; Input the sample data of each batch into the convolutional neural network in sequence for processing to obtain the output result of the convolutional neural network; Calculate the value of the loss function according to the output result of the convolutional neural network and the loss function of the convolutional neural network, and update the parameter values of the convolutional neural network using the stochastic gradient method until the network converges to obtain a trained convolutional neural network.

5. A method for open set recognition of high-resolution radar range profiles based on extreme value distribution according to claim 4, characterized in that, The inputting the sample data of each batch into the convolutional neural network in sequence for processing to obtain the output result of the convolutional neural network includes: Use the first convolutional layer to perform convolution and downsampling processing on the input sample data of the current batch to obtain a first feature map; Use the second convolutional layer to perform convolution and downsampling processing on the first feature map to obtain a second feature map; Use the third convolutional layer to perform convolution and downsampling processing on the second feature map to obtain a third feature map; The fourth fully connected layer is used to perform a non-linear transformation on the third feature map to obtain the output result of the current sample data; Repeat the above steps until the processing of q batches of sample data is completed to obtain the output result of the convolutional neural network.

6. The open set recognition method for high-resolution radar range profiles based on extreme value distribution according to claim 5, wherein The expression of the feature center of the high-dimensional features of the second training sample set is: Among them, represents the i-th sample in the k-th category of the second training sample set, and there are a total of N k samples in the k-th category; The expression of the extreme value distribution cumulative probability distribution function of the k-th category is: Among them, represents the extreme value distribution cumulative probability distribution function, t represents the Euclidean distance from each high-dimensional feature to the corresponding feature center, which is the independent variable of the function; ξ, α, and β represent the parameters of the extreme value distribution cumulative probability distribution function.

7. A method for open set recognition of high-resolution radar range profiles based on extreme value distribution according to claim 5, characterized in that The use of the extreme value distribution cumulative probability distribution function to perform open-set recognition on the high-dimensional features of the second test sample set to obtain the recognition result of the second test sample set includes: When is established, the test sample corresponding to the high-dimensional feature of the second test sample set is an unknown category outside the database; When does not hold, the category of the test sample corresponding to the high-dimensional feature of the second test sample set is where τ is a preset decision threshold.

Images