Spike convolution method for radar signals and radar target semantic segmentation network

Abstract

The present invention provides a peak convolution method for radar signals and a radar target semantic segmentation network, the method comprising: performing Fourier transform on radar echo signals to obtain tensors of several viewing angles; selecting any unit to be tested, obtaining a protection band and a reference band corresponding to the selected unit to be tested, and obtaining a peak receptive field; obtaining the total number of reference units in the reference band area; extracting all reference units; performing original peak convolution or noise suppression peak convolution processing to obtain radar echo signal characteristics after original / noise suppression peak convolution filtering and denoising; traversing all units to be tested in the tensors of all viewing angles to obtain radar echo signal characteristics after original / noise suppression peak convolution filtering and denoising corresponding to all units to be tested, thereby completing the peak convolution of the radar echo signal. The present invention can solve the technical problem that the convolution method in the prior art is not suitable for radar signals and cannot exert its due potential in the field of radar semantic segmentation.

Description

Technical Field

[0001] The present invention relates to the technical field of spike convolution, and in particular to a spike convolution method for radar signals and a radar target semantic segmentation network. Background Art

[0002] As a sensor with long-range perception, radar transmits modulated electromagnetic waves in a certain direction through an antenna in a specific working area to detect targets of interest. Radar is a type of active detection equipment. Compared with lidar, it can better cope with the adverse effects of extreme weather; compared with visible light sensors, radar's detection performance will not decrease due to extreme lighting conditions (extremely weak and extremely strong lighting conditions). In addition to providing real-world location information of the target, radar uses the Doppler phenomenon to further realize the perception of speed. Based on the above advantages, radar plays an irreplaceable role in many autonomous driving and confrontation tasks, such as autonomous driving environment perception and drone early warning.

[0003] Classic radar target detection is based on the CFAR method, which sets a guard band mechanism around the signal and uses the sampling points outside the guard band to estimate the noise, thereby providing a basis for setting the target discrimination threshold. However, in practical applications, the detection performance of this method depends on a large number of manual parameters, including threshold factors, shape and size of the detection area, etc. In addition, traditional radar detection methods cannot identify target types and lack the ability to predict target category semantic information. These hinder the automated development of radar target detection. Benefiting from the development of deep learning, radar target detection methods based on deep learning have pushed the interpretation of radar signals from the binary hypothesis of "target-clutter" of traditional radar target detection to modern machine learning for target semantic information, including radar target detection and recognition or radar semantic segmentation. Most of the existing studies are based on convolutional neural networks (CNNs), which use the frequency domain representation of radar signals as input and output prediction results on the corresponding frequency domain representations such as range-Doppler (RD) or range-azimuth (RA). However, radar target echo signals are not as easy for humans to intuitively understand as visible light images captured by cameras. Images have rich information such as color and texture, which makes it easier for convolution operations to efficiently extract semantic information in their receptive fields. Further combining human priors, more convolution methods such as multi-scale fusion convolution, dilated convolution, and deformable convolution have been proposed. These methods are directly applied to radar semantic segmentation or radar target detection tasks. For example, multi-scale convolution methods such as parallel inception or stacked pyramid pooling are used to extract radar signal features in receptive fields of different scales; dilated convolution is used to extract radar signal features in a larger receptive field; and deformable convolution is used to capture irregular signal patterns.

[0004] However, these convolution methods are actually proposed based on visible light image signals, not directly for radar signals. In the process of learning features, they do not simultaneously consider the impact of clutter on radar target signals. In the process of interpreting radar signals (such as detection or semantic segmentation), the reasonable processing of clutter is a key factor affecting the interpretation performance. This limitation directly leads to the inability of current convolutional neural networks to exert their due potential in the field of radar semantic segmentation. Summary of the invention

[0005] The present invention provides a peak convolution method for radar signals and a radar target semantic segmentation network, which can solve the technical problem that the convolution method in the prior art is not suitable for radar signals and cannot exert its due potential in the field of radar semantic segmentation.

[0006] According to one aspect of the present invention, a peak convolution method for radar signals is provided, the method comprising:

[0007] Performing Fourier transform on the radar echo signal to obtain tensors of several viewing angles, wherein the tensor of each viewing angle includes multiple units to be tested;

[0008] Select any unit to be tested, obtain a guard band and a reference band corresponding to the selected unit to be tested, and obtain a peak receptive field based on the unit to be tested and the reference band corresponding to the unit to be tested;

[0009] Obtaining the total number of reference cells in the reference band region based on the width of the guard band and the width of the reference band;

[0010] All reference units are extracted from the corresponding visual angle tensor based on the unit under test, the spike receptive field, and the total number of reference units in the reference band area;

[0011] Performing original peak convolution or noise suppression peak convolution processing on all extracted reference units to obtain radar echo signal characteristics after original peak convolution filtering and denoising or radar echo signal characteristics after noise suppression peak convolution filtering and denoising;

[0012] All units to be tested in the tensors of all viewing angles are traversed to obtain the radar echo signal characteristics after original peak convolution filtering and denoising or radar echo signal characteristics after noise suppression peak convolution filtering and denoising corresponding to all units to be tested, thereby completing the peak convolution of the radar echo signal.

[0013] Preferably, the performing of original peak convolution processing on all extracted reference units to obtain radar echo signal features after original peak convolution filtering and denoising includes:

[0014] All the extracted reference units are input into the two-dimensional convolution module to obtain the clutter estimation value;

[0015] The clutter in the unit under test is removed based on the clutter estimation value to obtain the radar echo signal characteristics after the original peak convolution filtering and denoising.

[0016] Preferably, the clutter estimate is obtained by the following formula:

[0017]

[0018] The characteristics of the radar echo signal after the original peak convolution filter denoising are obtained by the following formula:

[0019]

[0020] In the formula, e r represents the clutter estimate, Vec(·) represents the vectorization operation, * represents the convolution operation, N r represents the total number of reference cells, represents the shared weights of the spiking convolution kernel of the jth output channel of the ith reference unit, represents the i-th reference unit, C out Indicates the total number of output channels, It represents the characteristics of radar echo signal after original peak convolution filtering and denoising. represents the spike receptive field, W represents the shared weight of the spike convolution kernel of the unit to be tested, and x c Indicates the unit under test.

[0021] Preferably, the noise suppression peak convolution processing is performed on all the extracted reference units to obtain the radar echo signal characteristics after noise suppression peak convolution filtering and denoising, which includes:

[0022] Duplicate the unit under test N times r times to align with the peak receptive field, where N r represents the total number of reference units;

[0023] Subtract the replicated unit under test from each reference unit to obtain the characteristics of the unit under test with different noise suppression levels;

[0024] The characteristics of the units under test with different noise suppression levels are input into the two-dimensional convolution module to obtain the characteristics of the radar echo signal after noise suppression spike convolution filtering and denoising.

[0025] Preferably, the characteristics of the unit under test with different noise suppression degrees are obtained by the following formula:

[0026]

[0027] The characteristics of the radar echo signal after denoising by noise suppression peak convolution filtering are obtained by the following formula:

[0028]

[0029] In the formula, represents the characteristics of the unit under test with different noise suppression levels, Vec(·) represents the vectorized operation, x c Indicates the unit under test, represents the i-th reference unit, N r represents the total number of reference cells, It represents the characteristics of radar echo signal after denoising by spike convolution filtering. represents the shared weights of the spiking convolution kernel of the jth output channel of the ith reference unit, C out Indicates the total number of output channels.

[0030] Preferably, the peak receptive field is obtained by the following formula:

[0031]

[0032] In the formula, represents the peak receptive field, x c Indicates the unit under test, represents the i-th reference unit, N r Represents the total number of reference cells.

[0033] Preferably, the total number of reference cells in the reference band area is obtained by the following formula:

[0034]

[0035] Where N r represents the total number of reference cells, Indicates the horizontal or vertical width of the protection band. Indicates the horizontal or vertical width of the reference band.

[0036] Preferably, the plurality of viewing angles include one or more viewing angles of a “range-Doppler” viewing angle, a “range-azimuth” viewing angle, and an “azimuth-range” viewing angle.

[0037] According to another aspect of the present invention, a radar target semantic segmentation network is provided, wherein the radar target semantic segmentation network applies any of the above-mentioned methods to perform peak convolution processing to achieve radar semantic segmentation.

[0038] According to another aspect of the present invention, a computer device is provided, comprising a memory, a processor, and a peak convolution program for radar signals stored in the memory and executable on the processor, wherein the processor implements any of the above-mentioned methods when executing the peak convolution program for radar signals.

[0039] By applying the technical solution of the present invention, based on the peak response characteristics of the radar target signal itself, an efficient convolution mechanism suitable for radar target detection, namely peak convolution, is designed; the convolution mechanism continues the idea of ​​guard band setting and clutter estimation in the traditional radar target detection method, and realizes the learnability of clutter estimation and target screening by designing a novel "peak receptive field"; and according to the role of the unit under test in clutter estimation, the peak convolution adopts two implementation methods, namely the original peak convolution (PeakConv, PKC) and the noise suppression peak convolution (ResponseDifference-Aware PKC, ReDA-PKC), both of which contribute to the improvement of the radar semantic segmentation performance of the radar target semantic segmentation network (PKCIn-Net). The peak convolution method of the present invention solves the limitations of the current classic CFAR radar target detection method, such as strong dependence on manual parameter adjustment and lack of target category (semantic information) recognition ability, as well as the problems of low performance of existing deep learning-based radar target detection and recognition or semantic segmentation methods and no deep learning operators dedicated to radar signal processing. The peak convolution method of the present invention can be used for radar equipment with various detection systems such as continuous wave and pulse Doppler. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The included drawings are used to provide a further understanding of the embodiments of the present invention, which constitute a part of the specification, are used to illustrate the embodiments of the present invention, and together with the text description, explain the principles of the present invention. Obviously, the drawings in the following description are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0041] Figure 1 A flow chart of a peak convolution method for radar signals provided according to an embodiment of the present invention is shown;

[0042] Figure 2 A schematic diagram showing a protection band and a reference band provided according to an embodiment of the present invention is shown;

[0043] Figure 3 A schematic diagram of a peak receptive field provided according to an embodiment of the present invention is shown;

[0044] Figure 4 A schematic diagram showing the principle of the original spike convolution provided according to an embodiment of the present invention is shown;

[0045] Figure 5 A schematic diagram showing the principle of noise suppression spike convolution provided according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0046] It should be noted that, in the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0048] Unless otherwise specifically stated, the relative arrangement of the parts and steps described in these embodiments, numerical expressions and numerical values ​​do not limit the scope of the present invention. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the accompanying drawings are not drawn according to the actual proportional relationship. The technology, method and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but in appropriate cases, the technology, method and equipment should be regarded as a part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once a certain item is defined in an accompanying drawing, it does not need to be further discussed in subsequent drawings.

[0049] like Figure 1 As shown, the present invention provides a peak convolution method for radar signals, the method comprising:

[0050] S10, performing Fourier transform on the radar echo signal to obtain tensors of several viewing angles, wherein the tensor of each viewing angle includes multiple units to be tested;

[0051] S20, select any unit to be tested, and obtain the protection band and reference band corresponding to the selected unit to be tested, such as Figure 2 As shown, the peak receptive field is obtained based on the unit under test and the reference band corresponding to the unit under test, as shown in Figure 3 As shown;

[0052] S30, acquiring the total number of reference cells in the reference band area based on the width of the protection band and the width of the reference band;

[0053] S40, extracting all reference units from the corresponding visual angle tensor based on the unit to be tested, the spike receptive field, and the total number of reference units in the reference band area;

[0054] S50, performing original peak convolution or noise suppression peak convolution processing on all extracted reference units to obtain radar echo signal characteristics after original peak convolution filtering and denoising or radar echo signal characteristics after noise suppression peak convolution filtering and denoising;

[0055] S60, traversing all units to be tested in the tensors of all viewing angles, and repeating S20-S50 to obtain the radar echo signal characteristics after original peak convolution filtering and denoising or the radar echo signal characteristics after noise suppression peak convolution filtering and denoising corresponding to all units to be tested, thereby completing the peak convolution of the radar echo signal.

[0056] Based on the peak response characteristics of the radar target signal itself, the present invention designs an efficient convolution mechanism suitable for radar target detection, namely, peak convolution; the convolution mechanism continues the idea of ​​guard band setting and clutter estimation in the traditional radar target detection method, and realizes the learnability of clutter estimation and target screening by designing a novel "peak receptive field"; and according to the role of the unit under test in clutter estimation, the peak convolution adopts two implementation methods, namely, the original peak convolution (PeakConv, PKC) and the noise suppression peak convolution (Response Difference-Aware PKC, ReDA-PKC), both of which contribute to the improvement of the radar semantic segmentation performance of the radar target semantic segmentation network (PKCIn-Net). The peak convolution method of the present invention solves the limitations of the current classic CFAR radar target detection method, such as strong dependence on manual parameter adjustment and lack of target category (semantic information) recognition ability, as well as the problems of low performance of the existing deep learning-based radar target detection and recognition or semantic segmentation methods and no deep learning operators dedicated to radar signal processing. The peak convolution method of the present invention can be used for radar equipment with various detection systems such as continuous wave and pulse Doppler.

[0057] According to an embodiment of the present invention, in S10 of the present invention, the several viewing angles include one or more viewing angles of a “range-Doppler” viewing angle, a “range-azimuth” viewing angle, and an “azimuth-range” viewing angle.

[0058] According to an embodiment of the present invention, in S20 of the present invention, the peak receptive field is obtained by the following formula:

[0059]

[0060] In the formula, represents the peak receptive field, x c Indicates the unit under test, represents the i-th reference unit, N r Represents the total number of reference cells.

[0061] in,

[0062]

[0063] Where b G represents the width of the guard band, Respectively represent the horizontal and vertical widths of the protection band, b R represents the width of the reference band, Respectively represent the horizontal and vertical widths of the reference band, p c represents the coordinates of the unit to be measured, Respectively represent the horizontal and vertical coordinates of the unit to be tested, represents the coordinates of the i-th reference unit, Respectively represent the horizontal and vertical coordinates of the i-th reference unit, is the real number field, and C is the number of channels of the view tensor.

[0064] According to an embodiment of the present invention, in S30 of the present invention, the total number of reference cells in the reference band area is obtained by the following formula:

[0065]

[0066] Where N r represents the total number of reference cells, Indicates the horizontal or vertical width of the protection band. Indicates the horizontal or vertical width of the reference band.

[0067] According to an embodiment of the present invention, Figure 4 As shown, in S50 of the present invention, the original peak convolution processing is performed on all the extracted reference units to obtain the radar echo signal characteristics after the original peak convolution filtering and denoising, including:

[0068] S51, inputting all the extracted reference units into a two-dimensional convolution module to obtain a clutter estimation value;

[0069] S52, removing clutter in the unit to be tested based on the clutter estimation value, and obtaining the radar echo signal characteristics after the original peak convolution filtering and denoising.

[0070] Specifically, in S51 of the present invention, the clutter estimation value is obtained by the following formula:

[0071]

[0072] In S52 of the present invention, the characteristic of the radar echo signal after the original peak convolution filtering and denoising is obtained by the following formula:

[0073]

[0074] In the formula, e r represents the clutter estimate, Vec(·) represents the vectorization operation, * represents the convolution operation, N r represents the total number of reference cells, represents the shared weights of the spiking convolution kernel of the jth output channel of the ith reference unit, represents the i-th reference unit, C out Indicates the total number of output channels, It represents the characteristics of radar echo signal after original peak convolution filtering and denoising. represents the spike receptive field, W represents the shared weight of the spike convolution kernel of the unit to be tested, and x c Indicates the unit under test.

[0075] in, And C in =C out ,

[0076] In the formula, C in Indicates the total number of input channels.

[0077] According to an embodiment of the present invention, Figure 5 As shown, in S50 of the present invention, the noise suppression peak convolution processing is performed on all the extracted reference units to obtain the radar echo signal characteristics after noise suppression peak convolution filtering and denoising, which include:

[0078] S53, copy the unit to be tested to N r times to align with the peak receptive field, where N r represents the total number of reference units;

[0079] S54, subtracting the replicated unit under test from each reference unit to obtain characteristics of the unit under test with different noise suppression levels;

[0080] S55, inputting the characteristics of the unit to be tested with different noise suppression degrees into a two-dimensional convolution module to obtain the characteristics of the radar echo signal after noise suppression spike convolution filtering and denoising.

[0081] Specifically, in S54 of the present invention, the characteristics of the unit under test with different noise suppression degrees are obtained by the following formula:

[0082]

[0083] In S55 of the present invention, the radar echo signal characteristics after noise suppression peak convolution filtering and denoising are obtained by the following formula:

[0084]

[0085] In the formula, represents the characteristics of the unit under test with different noise suppression levels, Vec(·) represents the vectorized operation, x c Indicates the unit under test, represents the i-th reference unit, N r represents the total number of reference cells, It represents the characteristics of radar echo signal after denoising by spike convolution filtering. represents the shared weights of the spiking convolution kernel of the jth output channel of the ith reference unit, C out Indicates the total number of output channels.

[0086] in,

[0087] The present invention also provides a radar target semantic segmentation network, which applies any of the above-mentioned methods to perform peak convolution processing to achieve radar semantic segmentation.

[0088] The present invention also provides a computer device, comprising a memory, a processor, and a peak convolution program for radar signals stored in the memory and executable on the processor, wherein the processor implements any of the above-mentioned methods when executing the peak convolution program for radar signals.

[0089] In summary, the present invention provides a peak convolution method for radar signals and a radar target semantic segmentation network. Based on the peak response characteristics of the radar target signal itself, an efficient convolution mechanism suitable for radar target detection, namely, peak convolution, is designed. The convolution mechanism continues the idea of ​​guard band setting and clutter estimation in the traditional radar target detection method, and realizes the learnability of clutter estimation and target screening by designing a novel "peak receptive field". According to the role of the unit under test in clutter estimation, the peak convolution adopts two implementation methods, namely, original peak convolution (PeakConv, PKC) and noise suppression peak convolution (Response Difference-Aware PKC, ReDA-PKC). Both implementation methods contribute to the improvement of the radar semantic segmentation performance of the radar target semantic segmentation network (PKCIn-Net). The peak convolution method of the present invention solves the limitations of the current classic CFAR radar target detection method, such as strong dependence on manual parameter adjustment and lack of target category (semantic information) recognition ability, as well as the low performance of existing deep learning radar target detection and recognition or semantic segmentation methods and the lack of deep learning operators dedicated to radar signal processing. The peak convolution method of the present invention can be used for radar equipment with various detection systems such as continuous wave and pulse Doppler.

[0090] Parts of the present invention that are not described in detail are well known to those skilled in the art.

[0091] In the description of the present invention, it is necessary to understand that the directions or positional relationships indicated by directional words such as "front, back, up, down, left, right", "lateral, vertical, perpendicular, horizontal" and "top, bottom" are usually based on the directions or positional relationships shown in the drawings. They are only for the convenience of describing the present invention and simplifying the description. Unless otherwise specified, these directional words do not indicate or imply that the devices or elements referred to must have a specific direction or be constructed and operated in a specific direction. Therefore, they cannot be understood as limiting the scope of protection of the present invention. The directional words "inside and outside" refer to the inside and outside relative to the contours of each component itself.

[0092] For ease of description, spatially relative terms such as "above", "above", "on the upper surface of", "above", etc. may be used here to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" can include both "above" and "below". The device can also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptions used here are interpreted accordingly.

[0093] In addition, it should be noted that the use of terms such as "first" and "second" to limit components is only for the convenience of distinguishing the corresponding components. If not otherwise stated, the above terms have no special meaning and therefore cannot be understood as limiting the scope of protection of the present invention.

[0094] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A peak convolution method for radar signals, characterized in that: The method comprises: Performing Fourier transform on the radar echo signal to obtain tensors of several viewing angles, wherein the tensor of each viewing angle includes multiple units to be tested; Select any unit to be tested, obtain a guard band and a reference band corresponding to the selected unit to be tested, and obtain a peak receptive field based on the unit to be tested and the reference band corresponding to the unit to be tested; Obtaining the total number of reference cells in the reference band region based on the width of the guard band and the width of the reference band; All reference units are extracted from the corresponding visual angle tensor based on the unit under test, the spike receptive field, and the total number of reference units in the reference band area; Performing original peak convolution or noise suppression peak convolution processing on all extracted reference units to obtain radar echo signal characteristics after original peak convolution filtering and denoising or radar echo signal characteristics after noise suppression peak convolution filtering and denoising; All units to be tested in the tensors of all viewing angles are traversed to obtain the radar echo signal characteristics after original peak convolution filtering and denoising or radar echo signal characteristics after noise suppression peak convolution filtering and denoising corresponding to all units to be tested, thereby completing the peak convolution of the radar echo signal.

2. The method according to claim 1, characterized in that The original peak convolution processing is performed on all the extracted reference units to obtain the radar echo signal characteristics after the original peak convolution filtering and denoising, including: All the extracted reference units are input into the two-dimensional convolution module to obtain the clutter estimation value; The clutter in the unit under test is removed based on the clutter estimation value to obtain the radar echo signal characteristics after the original peak convolution filtering and denoising.

3. The method according to claim 2, characterized in that The clutter estimate is obtained by the following formula: The characteristics of the radar echo signal after the original peak convolution filter denoising are obtained by the following formula: In the formula, e r represents the clutter estimate, Vec(·) represents the vectorization operation, * represents the convolution operation, N r represents the total number of reference cells, represents the shared weights of the spiking convolution kernel of the jth output channel of the ith reference unit, represents the i-th reference unit, C out Indicates the total number of output channels, It represents the characteristics of radar echo signal after original peak convolution filtering and denoising. represents the spike receptive field, W represents the shared weight of the spike convolution kernel of the unit to be tested, and x c Indicates the unit under test.

4. The method according to claim 1, characterized in that: The noise suppression peak convolution processing is performed on all the extracted reference units to obtain the radar echo signal characteristics after noise suppression peak convolution filtering and denoising, which include: Duplicate the unit under test N times r times to align with the peak receptive field, where N r represents the total number of reference units; Subtract the replicated unit under test from each reference unit to obtain the characteristics of the unit under test with different noise suppression levels; The characteristics of the units under test with different noise suppression levels are input into the two-dimensional convolution module to obtain the characteristics of the radar echo signal after noise suppression spike convolution filtering and denoising.

5. The method according to claim 4, characterized in that The characteristics of the unit under test with different noise suppression levels are obtained by the following formula: The characteristics of the radar echo signal after denoising by noise suppression peak convolution filtering are obtained by the following formula: In the formula, represents the characteristics of the unit under test with different noise suppression levels, Vec(·) represents the vectorized operation, x c Indicates the unit under test, represents the i-th reference unit, N r represents the total number of reference cells, It represents the characteristics of radar echo signal after denoising by spike convolution filtering. represents the shared weights of the spiking convolution kernel of the jth output channel of the ith reference unit, C out Indicates the total number of output channels.

6. The method according to any one of claims 1 to 5, characterized in that: The peak receptive field is obtained by the following formula: In the formula, represents the peak receptive field, x c Indicates the unit under test, represents the i-th reference unit, N r Represents the total number of reference cells.

7. The method according to any one of claims 1 to 5, characterized in that: The total number of reference cells in the reference band area is obtained by the following formula: Where N r represents the total number of reference cells, Indicates the horizontal or vertical width of the protection band. Indicates the horizontal or vertical width of the reference band.

8. The method according to claim 1, characterized in that The multiple viewing angles include one or more viewing angles of a “range-Doppler” viewing angle, a “range-azimuth” viewing angle, and an “azimuth-range” viewing angle.

9. A computer device, characterized in that: The invention comprises a memory, a processor and a peak convolution program for radar signals which is stored in the memory and can be run on the processor. When the processor executes the peak convolution program for radar signals, the method according to any one of claims 1 to 8 is implemented.

Images