A method for detecting external radiation source radar targets

Abstract

The invention discloses a target detection method for an external radiation source radar. The invention first obtains a range-Doppler matrix A of an external radiation source radar; saves A as a grayscale image B, inputs B into a network to detect whether an abnormal situation such as a reference channel containing a target echo occurs; if the output is 0, it indicates that no abnormal situation occurs, then a constant false alarm detection is directly performed to obtain a target detection result; if the output is 1, it indicates that an abnormal situation occurs, then a constant false alarm detection is first performed to obtain a detection matrix C, each element of C greater than 0 is regarded as a pending target, and a reference signal is reconstructed with each pending target, and then clutter suppression, range-Doppler processing, and constant false alarm detection are performed with the reconstructed reference signal to obtain a new detection matrix D; compare the number of elements greater than 0 in C and D, if the number decreases, the corresponding pending target is a true target and should be retained; if the number increases, the corresponding pending target is a false target and should be set to 0. After all pending targets are processed as described above, the final target detection result can be obtained.

Description

Technical Field

[0001] The invention belongs to the technical field of external radiation source radar target detection, and in particular relates to a target detection method when an external radiation source radar reference channel contains a target echo. Background Art

[0002] The external radiation source radar has at least two receiving channels. The channel that points to the base station location and is used to receive the direct wave signal emitted by the base station is called the reference channel; the other channels are used to receive the reflected echo of the target and are called echo channels.

[0003] In addition to receiving the target echo, the echo channel often also receives a small amount of direct wave signals, multipath signals and noise, and the energy of the target echo is much lower than the energy of the direct wave and multipath. Therefore, before performing target detection, clutter suppression and range-Doppler processing are required to improve the target detection signal-to-noise ratio. The premise of clutter suppression and range-Doppler processing is to have a "pure" direct wave signal. In the conventional processing of external radiation source radar target detection, it is a basic assumption that the reference channel is not mixed with target echo signals. However, in practice, when there is a site error in the base station or the main lobe of the antenna beam is wide, the reference channel may also receive the target echo. At this time, if the conventional method is continued, clutter suppression and range-Doppler processing are performed with the "impure" reference channel, many false targets will eventually be detected, which will affect the external radiation source radar target detection. Summary of the invention

[0004] The purpose of the present invention is to provide a target detection method for an exo-radiation source radar, which detects a target when a reference channel of the exo-radiation source radar is mixed with a target echo, reduces the number of false targets, and meets the target detection requirements of the exo-radiation source radar when the reference channel contains a target echo.

[0005] In order to achieve the above technical objectives, the present invention adopts the following technical solutions to achieve them.

[0006] A method for detecting a target by an external radiation source radar comprises the following steps:

[0007] Step 1: Obtain a reference signal matrix A and an echo signal matrix B of an external radiation source radar; perform clutter suppression and range-Doppler processing on the echo signal matrix B to obtain a range-Doppler matrix C; perform grayscale processing on the obtained range-Doppler matrix C to obtain a grayscale image D;

[0008] Step 2: Input the grayscale image D into the trained image binary classification network. The network outputs two results: 0 or 1. 0 means that the reference signal does not contain the target echo, and enters the normal processing branch, and then proceeds to step 3; 1 means that the reference signal contains the target echo, and enters the abnormal processing branch, and then proceeds to step 4;

[0009] Step 3: Multiply the average value of each row of the range-Doppler matrix C by a fixed detection coefficient to obtain the detection threshold value of each Doppler channel of the range-Doppler matrix C; retain the elements in each row of the range-Doppler matrix C that are greater than the detection threshold value, and set the elements that are less than or equal to the detection threshold value to 0 to obtain a new matrix E, which is the result of target detection when the reference channel does not contain target echo;

[0010] Step 4: Perform the same operation as step 3 to obtain a new matrix E. Traverse each row of E and record the row number of each row that is not all zero as row1…row k ; and row i Each non-zero element in a row is considered as a pending target, where 0<i≤k, and the number of pending targets in each row is recorded as t i ; Use row1…row k The reference signal of each target to be determined is reconstructed; the reconstructed reference signal is then used to perform clutter suppression and range-Doppler processing on the echo signal matrix B again, and a new range-Doppler matrix is ​​obtained;

[0011] Step 5: Transform the new range-Doppler matrix F ij The average value of each row of elements is multiplied by a fixed detection coefficient to obtain the range-Doppler matrix F ij The detection threshold value of each Doppler channel; the range-Doppler matrix F ij In each row, the elements greater than the detection threshold are retained, and the elements less than or equal to the detection threshold are set to 0, and a new matrix G is obtained. ij ;

[0012] Step 6: For the new matrix G ij row i The rows are traversed, and the number of elements greater than 0 is recorded as s i , if s i Less than the row of matrix E in step 4 i The number of pending targets for the row t i , then keep the corresponding elements in the matrix E; if s i Equal to or greater than the row in step 4 i The number of pending targets for the row t i , then set the corresponding elements in the matrix E to 0, and after performing the above reference signal reconstruction and other processing on each pending target in the detection matrix E in step 4, a new target detection matrix H will be obtained. This matrix is ​​the target detection result when the reference channel contains target echo.

[0013] Preferably, the grayscale image processing of the range-Doppler matrix C comprises the following steps: obtaining the value max of the maximum element and the value min of the minimum element of the range-Doppler matrix, subtracting each element of the range-Doppler matrix from the minimum value min and dividing by the difference between the maximum value and the minimum value, to obtain a range-Doppler matrix in which each element value is between [0,1]; and then saving the range-Doppler matrix in the form of an image to obtain a grayscale image corresponding to the range-Doppler matrix C.

[0014] Preferably, the external radiation source radar obtains the echo signal through the main antenna and obtains the reference signal through the auxiliary antenna; the clutter suppression of the echo signal matrix B is to cancel the clutter of the echo signal and the reference signal; the range-Doppler matrix C is an m×n-order matrix, m is the number of Doppler channels of the external radiation source radar, and n is the number of range channels of the external radiation source radar.

[0015] Preferably, in step 2, the image binary classification network uses an automatic machine learning framework to perform hyperparameter search on batch_size, drop_out, and learning rate, and optimizes the kernel_size of the convolutional layer of the classification network and the features of the fully connected layer. A set of network parameters that gives the highest F1 value on the test set can be obtained.

[0016] Preferably, in step 4, the matrix E row i The jth non-zero element in the row is recorded as the pending target tar ij ; Using the undetermined target to reconstruct the reference signal includes the following steps:

[0017] Record the Doppler unit of the target to be determined as fd and the distance unit as rd; construct a temporary matrix T using the Doppler unit fd and the reference signal matrix A;

[0018]

[0019] Wherein, T(j) is the value of the j-th row of the temporary matrix, A(j) is the value of the j-th row of the reference signal matrix, and LS is the number of rows of the reference signal matrix. Then, the T matrix is ​​delayed by rd-1, rd, and rd+1 respectively to form the target subspace matrix Q. The matrix Q is shown as follows:

[0020]

[0021] Where T(i) is the element value of the i-th row of matrix T, LS is the number of rows of the reference signal matrix; then the projection coefficient matrix W can be obtained using the following formula

[0022] W=(Q H *Q) -1 *QH *Err

[0023] Among them, Err is the residual echo signal matrix after clutter suppression of the echo signal matrix B in step 1; finally, the reconstructed reference signal matrix A can be obtained using the following formula: new

[0024] A new =AQ*W

[0025] Where A is the original reference signal matrix, Q is the target subspace matrix, and W is the projection coefficient matrix.

[0026] The beneficial effects of the present invention are as follows: 1) The present invention uses an automatic machine learning method to perform a hyperparameter search on the image classification network, and automatically finds a set of parameters that optimize the network performance. The image classification network is used to detect whether the reference channel is mixed with target echoes, thereby realizing frame-level detection without the need to perform matching detection row by row in the range-Doppler matrix, thereby improving the efficiency of detection. 2) The present invention uses a method of reconstructing a reference signal to remove false targets from the range-Doppler matrix when the reference channel is mixed with target echoes. Compared with the traditional external radiation source target detection method, the false targets generated by the reference channel containing target echoes are effectively eliminated. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a schematic flow chart of the external radiation source radar target detection method of the present invention;

[0028] Figure 2 Schematic diagram of the range-Doppler matrix saved as a grayscale image;

[0029] Figure 3a This is a schematic diagram of the results of directly performing constant false alarm detection on the range-Doppler matrix when the reference channel contains a target echo using the traditional method;

[0030] Figure 3b It is a schematic diagram of the detection result after the range-Doppler matrix of the reference channel containing the target echo is processed by the present invention; DETAILED DESCRIPTION

[0031] The present invention will be further described below in conjunction with the accompanying drawings:

[0032] Reference Figure 1 , is a flow chart of a method for detecting a target by an external radiation source radar of the present invention. The method for detecting a target by an external radiation source radar comprises the following steps:

[0033] Step 1: Obtain a reference signal matrix A and an echo signal matrix B of an external radiation source radar; perform clutter suppression and range-Doppler processing on the echo signal matrix B to obtain a range-Doppler matrix C of the external radiation source radar; the specific description is as follows: the echo channel of the external radiation source radar receives the echo signal reflected by the target, and the reference channel receives the reference signal irradiated by the base station; then use the reference signal to perform clutter cancellation and range-Doppler processing on the echo signal to obtain the range-Doppler matrix C (the number of Doppler channels of the external radiation source radar is m, and the number of range channels of the external radiation source radar is n, then the range-Doppler matrix C is an m×n matrix. The range-Doppler matrix C can be expressed as:

[0034]

[0035] Among them, c ij is the th element in matrix C, i ranges from 1 to m, and j ranges from 1 to n.

[0036] After obtaining the range-Doppler matrix C, the range-Doppler matrix C is converted into a grayscale image to obtain a grayscale image D. The specific operations are as follows:

[0037] Traverse the range-Doppler matrix C, obtain the value of the maximum element of the range-Doppler as max, and the value of the minimum element as min. Use the following formula to update the value of each element of the matrix C, save the new matrix in the form of an image, and you can get the grayscale image D corresponding to the range-Doppler matrix C.

[0038]

[0039] Among them, c ij is the element in the i-th row and j-th column of the range-Doppler matrix C.

[0040] Step 2: Input the grayscale image D into the trained image binary classification network. The network will output two results, 0 or 1, where 0 means that there is no target echo in the reference channel and enters the normal processing branch, which includes step 3; 1 means that there is a target echo in the reference channel and enters the abnormal processing branch, which includes steps 4, 5, and 6. The image binary classification network uses the automatic machine learning framework NNI to perform hyperparameter search for batch_size, drop_out, and learning rate, and optimizes the kernel_size of the convolution layer of the VGG19 network and the features of the fully connected layer. The optimal network is trained using 10017 images, and the F1 value reaches 0.889527 on 1292 test images.

[0041] Step 3: Perform constant false alarm detection on the range-Doppler matrix C in step 1 to obtain the target detection matrix E. The specific instructions are as follows:

[0042] The average value of each Doppler channel of the range-Doppler matrix C is calculated in sequence using the following formula.

[0043]

[0044] Among them, c ij represents the element in the i-th row and j-th column of the range-Doppler matrix C, avg i Represents the average value of the elements in the i-th row of the range-Doppler matrix C.

[0045] Then, the threshold value of each Doppler channel of the range-Doppler matrix C is calculated using the following formula.

[0046] gate i =avg i *coeff

[0047] Where coeff is a set detection coefficient, gate i Represents the detection threshold of the i-th Doppler channel of the range-Doppler matrix C.

[0048] Then each element of the i-th Doppler channel of the range-Doppler matrix C is compared with the threshold value gate of this channel i For comparison, elements greater than the threshold are retained, and elements less than or equal to the threshold are set to 0.

[0049] After performing the above operations on all Doppler channels of the range-Doppler matrix C in turn, the target detection matrix E is obtained, which is the target detection result when the reference channel does not contain the target echo.

[0050] Step 4: When the output result of the image binary classification network is 1, it means that the matrix C in step 1 is the range-Doppler matrix obtained when the reference channel contains the target echo. Therefore, additional processing is required. First, the detection matrix E is obtained according to step 3. Each row of the matrix E is traversed, and the row number of each row that is not all 0 is recorded as row1…row k . And row i Each non-zero element in the (1≤i≤k) row is considered as a pending target, and the number of pending targets in each row is recorded as t i . Use row1…row k For each pending target in the row, the reference signal is reconstructed. i The jth target to be determined in the row is reconstructed to obtain a new range-Doppler matrix denoted as F ij . The detection matrix E row i The j-th non-zero element in the (1≤i≤k) row is recorded as the pending target tar ij , the specific description of using the pending target to reconstruct the reference signal is as follows:

[0051] The Doppler unit where the target is located is recorded as fd, and the distance unit where the target is located is recorded as rd. The temporary matrix T is constructed using the following formula.

[0052]

[0053] Wherein, T(j) is the value of the j-th row of the temporary matrix, A(j) is the value of the j-th row of the reference signal matrix, and LS is the number of rows of the reference signal matrix. Then, the T matrix is ​​delayed by rd-1, rd, and rd+1 respectively to form the target subspace matrix Q. The matrix Q is shown as follows:

[0054]

[0055] Where T(i) is the element value of the i-th row of matrix T, and LS is the number of rows in the reference signal matrix. Then the projection coefficient matrix W can be obtained using the following formula

[0056] W=(Q H *Q) -1 *Q H *Err

[0057] Where Err is the remaining echo signal matrix after clutter suppression of the echo signal matrix B in step 1. Finally, the reconstructed reference signal matrix A can be obtained using the following formula new

[0058] A new =AQ*W

[0059] Where A is the original reference signal matrix, Q is the target subspace matrix, and W is the projection coefficient matrix. Then, the reconstructed reference signal matrix A is used new , perform clutter suppression and range-Doppler processing on the echo signal matrix B again to obtain a new range-Doppler matrix. i The new range-Doppler matrix is ​​obtained by adding the j-th non-zero element in the (1≤i≤k) row, so the new range-Doppler matrix is ​​recorded as F ij .

[0060] Step 5: Use the new range-Doppler matrix F ij Perform constant false alarm detection. The constant false alarm detection steps are as described in step 3, and a new detection matrix G is obtained. ij .

[0061] Step 6: For the new matrix G ij row i The rows are traversed, and the number of elements greater than 0 is recorded as s i , if s i Less than the row of matrix E in step 4i The number of pending targets for the row t i , then keep the corresponding elements in the matrix E; if s i Equal to or greater than the row of matrix E in step 4 i The number of pending targets for the row t i , then set the corresponding elements in the matrix E to 0. After performing the above reference signal reconstruction and other processing on each undetermined target in the detection matrix E in step 4, a new target detection matrix H will be obtained. This matrix is ​​the target detection result when the reference channel contains target echo.

[0062] After obtaining the target detection matrix, subsequent processing can be performed. For example, the target detection matrix is ​​transmitted to the processing module of the radar backend for analysis and processing.

[0063] The present invention will be further described below by a specific embodiment:

[0064] A frame of measured external radiation source radar data is detected, including two targets, target 1 and target 2. The echo signal of target 1 exists in both the reference channel and the echo channel, while the echo signal of target 2 exists only in the echo channel. Other parameters are set as follows: the number of Doppler channels m is 400, the number of range channels n is 200, the number of Doppler channel shielding is 4, the number of range channel shielding is 5, and the constant false alarm empirical value is set to 3.5.

[0065] Using the external radiation source radar reference channel to perform clutter cancellation and range-Doppler processing on the echo channel, the range-Doppler matrix can be obtained, and then the range-Doppler matrix is ​​saved as a grayscale image. Figure 2 .Will Figure 2 The image is sent to the binary classification network, and the network finally outputs 1, which means that the reference channel contains the target echo signal. The method proposed by the present invention is used to process the case where the reference channel contains the target echo, and the processing result is compared with the traditional processing method that does not consider this abnormal situation. Figure 3a , is the result of directly performing constant false alarm detection using the traditional method; refer to Figure 3b , is the result of using the abnormal situation processing method proposed by the present invention. Figure 3a and Figure 3b By comparison, we can find that Figure 3a When the constant false alarm detection is directly performed without considering abnormal conditions, the echo signal of target 1 is mixed into the reference channel, resulting in several false targets appearing after target 1, and all the false targets are mistaken for real targets by passing the threshold. However, the false targets generated by target 1 can be eliminated by using the method proposed by the present invention without affecting normal targets. This shows the effectiveness and superiority of the target detection method of the present invention.

[0066] In summary, the present invention processes the range-Doppler matrix grayscale image and sends it to the image binary classification network to identify whether the reference channel contains the target echo signal, thereby realizing frame-level recognition and improving recognition efficiency. For the problem that the range-Doppler matrix contains multiple false targets due to the abnormal situation, the present invention reconstructs the reference signal using each pending target, and then determines whether each pending target is a true target through the change in the number of targets in the two detection matrices before and after, so that all false targets can be removed.

[0067] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for detecting external radiation source radar targets, characterized in that: The following steps are involved: Step 1: Obtain a reference signal matrix A and an echo signal matrix B of an external radiation source radar; perform clutter suppression and range-Doppler processing on the echo signal matrix B to obtain a range-Doppler matrix C; perform grayscale processing on the obtained range-Doppler matrix C to obtain a grayscale image D; Step 2: Input the grayscale image D into the trained image binary classification network. The network outputs two results: 0 or 1. 0 means that the reference signal does not contain the target echo and enters the normal processing branch, then proceed to step 3. 1 means that the reference signal contains the target echo, enters the abnormal processing branch, and goes to step 4; Step 3: Multiply the average value of each row of the range-Doppler matrix C by a fixed detection coefficient to obtain the detection threshold value of each Doppler channel of the range-Doppler matrix C; retain the elements in each row of the range-Doppler matrix C that are greater than the detection threshold value, and set the elements that are less than or equal to the detection threshold value to 0 to obtain a new matrix E, which is the result of target detection when the reference channel does not contain target echo; Step 4: Perform the same operation as step 3 to obtain a new matrix E. Traverse each row of E and record the row number of each row that is not all zero as row1…row k ; and row i Each non-zero element in a row is considered as a pending target, where 0<i≤k, and the number of pending targets in each row is recorded as t i ; Use row1…row k The reference signal of each target to be determined is reconstructed; the reconstructed reference signal is then used to perform clutter suppression and range-Doppler processing on the echo signal matrix B again, and a new range-Doppler matrix is ​​obtained; Step 5: Transform the new range-Doppler matrix F ij The average value of each row of elements is multiplied by a fixed detection coefficient to obtain the range-Doppler matrix F ij The detection threshold value of each Doppler channel; the range-Doppler matrix F ij In each row, the elements greater than the detection threshold are retained, and the elements less than or equal to the detection threshold are set to 0, and a new matrix G is obtained. ij ; Step 6: For the new matrix G ij row i The rows are traversed, and the number of elements greater than 0 is recorded as s i , if s i Less than the row of matrix E in step 4 i The number of pending targets for the row t i , then keep the corresponding elements in the matrix E; if s i Equal to or greater than the row in step 4 i The number of pending targets for the row t i , then set the corresponding elements in the matrix E to 0, and after performing the above reference signal reconstruction and other processing on each pending target in the detection matrix E in step 4, a new target detection matrix H will be obtained. This matrix is ​​the target detection result when the reference channel contains target echo.

2. The method for detecting a target using an external radiation source radar as claimed in claim 1, wherein: The grayscale image processing of the range-Doppler matrix C comprises the following steps: obtaining the value max of the maximum element and the value min of the minimum element of the range-Doppler matrix, subtracting each element of the range-Doppler matrix from the minimum value min and dividing by the difference between the maximum value and the minimum value, to obtain a range-Doppler matrix in which each element value is between [0,1]; and then saving the range-Doppler matrix in the form of an image, to obtain a grayscale image corresponding to the range-Doppler matrix C.

3. The method for detecting a target using an external radiation source radar as claimed in claim 1, wherein: The external radiation source radar obtains the echo signal through the main antenna and obtains the reference signal through the auxiliary antenna; the clutter suppression of the echo signal matrix B is to cancel the clutter of the echo signal and the reference signal; the range-Doppler matrix C is an m×n-order matrix, m is the number of Doppler channels of the external radiation source radar, and n is the number of range channels of the external radiation source radar.

4. The method for detecting external radiation source radar target according to claim 1, characterized in that: In step 2, The image binary classification network uses an automatic machine learning framework to perform hyperparameter search on batch_size, drop_out, and learning rate. It also optimizes the kernel_size of the convolutional layer of the classification network and the features of the fully connected layer. A set of network parameters with the highest F1 value on the test set is obtained.

5. The method for detecting external radiation source radar target according to claim 1, characterized in that: In step 4, the row of matrix E is i The jth non-zero element in the row is recorded as the pending target tar ij ; Using the undetermined target to reconstruct the reference signal includes the following steps: Record the Doppler unit of the target to be determined as fd and the distance unit as rd; construct a temporary matrix T using the Doppler unit fd and the reference signal matrix A; Wherein, T(j) is the value of the j-th row of the temporary matrix, A(j) is the value of the j-th row of the reference signal matrix, and LS is the number of rows of the reference signal matrix. Then, the T matrix is ​​delayed by rd-1, rd, and rd+1 respectively to form the target subspace matrix Q. The matrix Q is shown as follows: Where T(i) is the element value of the i-th row of matrix T, LS is the number of rows of the reference signal matrix; then the projection coefficient matrix W can be obtained using the following formula W=(Q H *Q) -1 *Q H *Err Among them, Err is the residual echo signal matrix after clutter suppression of the echo signal matrix B in step 1; finally, the reconstructed reference signal matrix A can be obtained using the following formula: new A new =A-Q*W Where A is the original reference signal matrix, Q is the target subspace matrix, and W is the projection coefficient matrix.

Images