False alarm identification modeling method, identification method, equipment, storage medium and product

Abstract

The invention discloses a false alarm identification modeling method, an identification method, equipment, a storage medium and a product, and relates to the technical field of radars. The modeling method comprises the following steps: acquiring target trajectory data; extracting a target point location echo signal according to the target point location information, the obtained radar echo signal and the steering vector; performing micro-Doppler feature extraction on the target point location echo signal to obtain a micro-Doppler map; labeling the micro-Doppler maps, and constructing a sample data set according to the micro-Doppler maps and labels thereof; and constructing a deep learning network model, and training and testing the deep learning network model by using the sample data set to obtain a false alarm identification model. According to the invention, accurate detection of a real target and a false target under a complex background is realized, the detection precision is improved, the false alarm rate is reduced, and the anti-interference capability and robustness of radar equipment are enhanced.

Description

Technical Field

[0001] The present invention belongs to the field of radar technology, and in particular relates to a false alarm identification modeling method, an identification method, a device, a storage medium and a product. Background Art

[0002] A through-the-wall radar is a device that detects targets behind walls or other shelters through electromagnetic wave signals. It uses the strong penetration of ultra-wideband radar signals on non-metallic medium walls and other obstacles. After penetrating non-metallic obstacles such as walls, it encounters human targets and generates radar echoes. Weak movements such as human movement, breathing, heartbeat, body shaking, and limb swinging will form a Doppler effect on the echo. Based on this, the received echo is processed and analyzed to obtain the location, number, posture and other information of the human targets behind the wall, and display it on the control terminal client software interface.

[0003] False alarm means that the system incorrectly indicates that a target has appeared when there is no real target. For example, in a typical life scene, there are usually not only human targets, but also other non-human targets such as fans, air conditioners, curtains, etc. These non-human targets will also form a Doppler effect on the echo during use, causing the through-wall radar equipment to generate false alarms. The false alarm problem reduces the reliability and effectiveness of the radar system, and may lead to waste of resources, decision-making errors, and even security threats. For example, it will add unnecessary burdens to emergency rescue, delay rescue work, and cause loss of life and property.

[0004] False alarm suppression technology aims to reduce false alarms in the detection system, ensuring that the alarm is displayed or triggered only when the target is actually detected. This helps to improve detection accuracy, avoid waste of resources due to false detection, and ensure that operators can rely on the information provided by the system to make correct judgments. Therefore, false alarm suppression technology plays an important role in improving the overall performance and effectiveness of the system.

[0005] The development of micro-Doppler technology provides an effective way for radar target identification and false alarm suppression. Studies have shown that the micro-motion characteristics of human targets when stationary or moving are significantly different from those of non-human targets. The difference in micro-motion characteristics can provide effective information for radar target detection and identification. Micro-motion refers to the small periodic movement of an object outside of macroscopic movement. For example, human breathing or heartbeat can cause tiny vibrations in the body; a robotic arm may have slight vibrations when performing tasks, etc. These tiny movements are usually difficult to detect with the naked eye, but can be detected in sensor technologies such as radar. The Micro-Doppler Effect is a phenomenon in which the radar wave frequency of the echo signal changes slightly due to the micro-motion of an object. Different targets have different micro-Doppler characteristics due to their different materials, shapes, movement modes and other factors. For example, human breathing and heartbeat will produce specific micro-Doppler patterns, while rotating fan blades will have another set of different micro-Doppler patterns. The differences in these micro-Doppler patterns can be distinguished by time-frequency analysis of the echo signal. Common time-frequency analysis methods include short-time Fourier transform (STFT), Wigner-Ville distribution, etc. To distinguish the differences in micro-Doppler characteristics between different targets, a micro-motion model can be established, and machine learning algorithms can be used to learn the typical micro-Doppler signatures of different targets from a large amount of training data, and these signatures can be matched in practical applications to identify the target type.

[0006] In recent years, artificial intelligence technology represented by deep learning has developed rapidly. Deep learning technology has the advantages of strong learning ability, higher accuracy than traditional machine learning models, and strong generalization ability when solving practical problems. Deep learning technology has also been widely used in fields such as radar signal processing. For example, human posture recognition methods based on deep learning and Doppler radar have been used in gesture command recognition, gait recognition and other fields. Convolutional Neural Network (CNN) is the most commonly used algorithm in computer vision, especially in the field of image processing. CNN can achieve high-precision classification of input images, and no human intervention is required in the feature extraction process.

[0007] Therefore, combining deep learning with micro-Doppler images for intelligent extraction and recognition of target micro-motion features can not only make full use of the micro-motion information in the micro-Doppler images, but also give play to the advantages of deep learning in image processing, thereby realizing radar target recognition and false alarm suppression.

[0008] At present, the false alarm identification methods of through-wall radar include traditional methods and methods based on feature extraction. Among them, the traditional method usually relies on threshold setting to filter false alarms. This method is simple, but is easily affected by environmental noise. Especially in complex scenarios, its performance will be greatly reduced; the feature extraction-based method manually designs features and then uses machine learning algorithms (such as support vector machines SVM, random forest RF, etc.) for classification. Although this method can improve the accuracy of target detection to a certain extent, it has poor adaptability to dynamically changing environments. Summary of the invention

[0009] The purpose of the present invention is to provide a false alarm identification modeling method, an identification method, a device, a storage medium and a product to solve the problems that the traditional false alarm suppression technology has poor environmental adaptability and low false alarm detection accuracy.

[0010] The present invention solves the above technical problems through the following technical solutions: a false alarm identification modeling method, comprising:

[0011] Acquire target trajectory data; wherein the target trajectory data includes target point information at different times, and the target point information includes a distance unit and an azimuth unit where the target is located;

[0012] Obtain radar echo signals and guidance vectors after clutter suppression;

[0013] Extracting the target point echo signal according to the target point information, the radar echo signal after clutter suppression and the guidance vector;

[0014] When the length of the target point echo signal reaches a set threshold, micro-Doppler feature extraction is performed on the target point echo signal to obtain a micro-Doppler map;

[0015] Annotating the micro-Doppler image, and constructing a sample data set according to the micro-Doppler image and its annotations;

[0016] A deep learning network model is constructed, and the deep learning network model is trained and tested using the sample data set to obtain a false alarm recognition model.

[0017] Furthermore, before extracting the target point echo signal, the modeling method also includes performing sliding average processing on the target point information. The specific processing formula is:

[0018]

[0019] Among them, N A and P AThey represent the distance unit and azimuth unit of the target after sliding average processing, round() represents the rounding function, K represents the size of the sliding average window, and N i and P i They respectively represent the distance unit and azimuth unit where the target is located at the i-th moment before sliding average processing.

[0020] Further, according to the target point information, the radar echo signal after clutter suppression and the guidance vector, the target point echo signal is extracted, which specifically includes:

[0021] Extract the Nth A Extract the echo signal of the Pth distance unit from the guidance vector A The steering vector of the azimuth unit;

[0022] According to the Nth A The echo signal of the first distance unit and the A The guidance vector of each azimuth unit is used to calculate the target point echo signal. The specific calculation formula is:

[0023] I(N A ,P A )=X N θ p ;

[0024] Among them, X N Indicates the Nth A The echo signal of the range unit, θ p Indicates P A The steering vector of the azimuth unit, I(N A ,P A ) indicates the Nth A The distance unit is in P A The signal strength of each azimuth unit, N A and P A They respectively represent the distance unit and azimuth unit where the target is located.

[0025] Furthermore, when the length of the target point echo signal reaches a set threshold, before extracting the micro-Doppler feature of the target point echo signal, the modeling method further includes correcting the target point echo signal, specifically including:

[0026] Performing first-order difference processing on the target point echo signal to obtain the difference between the data before and after each moment;

[0027] The one-dimensional constant false alarm rate method is used to detect the difference of the data before and after each moment to obtain the discontinuity point;

[0028] Calculate the amplitude and phase difference of the data before and after each discontinuous point, and use the amplitude and phase difference as the correction weight of the discontinuous point;

[0029] According to the correction weight, the data after the discontinuous point is weighted point by point to obtain a corrected target point echo signal.

[0030] Furthermore, the deep learning network model selects the MobileNetV3 model, and builds the MobileNetV3 model based on the pytorch deep learning framework, and changes the number of channels of the input layer of the MobileNetV3 model to 1.

[0031] Based on the same concept, the present application also provides a false alarm identification method, the identification method comprising:

[0032] Calling a false alarm recognition model; wherein the false alarm recognition model is constructed according to the false alarm recognition modeling method described above;

[0033] A micro-Doppler map of an unknown target is obtained, and false alarm identification is performed on the micro-Doppler map of the unknown target using the false alarm identification model.

[0034] Furthermore, a micro-Doppler map of an unknown target is obtained, including:

[0035] Acquire unknown target trajectory data; wherein the unknown target trajectory data includes unknown target point information at different times, and the unknown target point information includes the distance unit and the azimuth unit where the unknown target is located;

[0036] Obtain radar echo signals and guidance vectors after clutter suppression;

[0037] Extracting the unknown target point echo signal according to the unknown target point information, the radar echo signal after clutter suppression and the guidance vector;

[0038] When the length of the unknown target point echo signal reaches a set threshold, micro-Doppler feature extraction is performed on the unknown target point echo signal to obtain a micro-Doppler map of the unknown target.

[0039] Furthermore, the identification method also includes abnormal target management, and the abnormal target management specifically includes:

[0040] When the length of the unknown target point echo signal does not reach the set threshold, it is determined whether the target corresponding to the unknown target point echo signal matches the target in the abnormal target library. If so, the micro-Doppler feature extraction is performed on the echo signal of the target in the abnormal target library to obtain the micro-Doppler map of the unknown target; if not, the unknown target point echo signal is discarded;

[0041] When an unknown target is identified as a false target, it is determined whether the false target matches the target in the abnormal target library. If so, the information of the target in the abnormal target library is updated; if not, the false target is entered into the abnormal target library.

[0042] Based on the same concept, the present application also provides an electronic device, including a memory, a processor, and a computer program / instruction stored in the memory, wherein the processor executes the computer program / instruction to implement the false alarm identification modeling method as described above.

[0043] Based on the same concept, the present application also provides a computer-readable storage medium on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the false alarm identification modeling method as described above is implemented.

[0044] Based on the same concept, the present application also provides a computer program product, including a computer program / instruction, which implements the false alarm identification modeling method as described above when executed by a processor.

[0045] Beneficial Effects

[0046] Compared with the prior art, the advantages of the present invention are:

[0047] The present invention distinguishes human targets from non-human targets (i.e., false targets) through micro-Doppler features. The micro-Doppler features reflect the frequency changes caused by the different movement speeds of different parts of the target. This frequency change can provide physical property information of the target. The strong learning ability of the deep learning network model is used to learn the association between the micro-Doppler features and the target category, so that the difference between human targets and false targets can be more accurately identified, thereby realizing the accurate detection of real targets and false targets in complex backgrounds, improving the detection accuracy, reducing the false alarm rate, and enhancing the anti-interference ability and robustness of the radar equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only one embodiment of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 is a flow chart of a false alarm identification modeling method in an embodiment of the present invention;

[0050] Figure 2 is a schematic diagram of discontinuous point detection of a target point echo signal in an embodiment of the present invention;

[0051] Figure 3is a time-frequency diagram of the target point echo signal before correction in an embodiment of the present invention;

[0052] Figure 4 is a time-frequency diagram of the target point echo signal after correction in an embodiment of the present invention;

[0053] Figure 5 is a schematic diagram of extracting micro-Doppler characteristics using short-time Fourier transform in an embodiment of the present invention;

[0054] Figure 6 is a micro-Doppler image of a false target in an embodiment of the present invention;

[0055] Figure 7 is a micro-Doppler image of a human target in an embodiment of the present invention;

[0056] Figure 8 It is a flow chart of the false alarm identification method in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The following is a clear and complete description of the technical solutions in the present invention 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. Based on the embodiments of 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.

[0058] The technical solution of the present application is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0059] Example 1

[0060] Figure 1 FIG. 4 shows a flow chart of the false alarm identification modeling method provided by the present invention. Figure 1 As shown, the modeling method includes the following steps:

[0061] Step 1: Get target trajectory data.

[0062] When the through-the-wall radar is detecting, it receives the radar echo signal, processes the radar echo signal, and obtains the target point. The target point refers to the distance unit and azimuth unit where the target is located. By tracking the target point, the target trajectory data can be obtained. Therefore, the target trajectory data includes the target point information at different times, that is, the distance unit and azimuth unit where the target is located at different times.

[0063] Step 2: Obtain the radar echo signal and guidance vector after clutter suppression.

[0064] In order to prevent clutter from affecting the detection accuracy, the radar echo signal is subjected to clutter suppression processing to obtain the radar echo signal after clutter suppression. The steering vector is a vector used to describe the directional characteristics of the signal received by the antenna array. The steering vector is determined according to the antenna array type of the radar and the signal incident angle. The acquisition of the steering vector is an existing technology.

[0065] Step 3: Perform sliding average processing on the target point information of the target trajectory data in step 1.

[0066] Since a static target may not be completely stationary, the target point information recorded in the target trajectory data will change within a small range. This change is not conducive to the extraction of the echo signal of the target point. It is necessary to perform sliding average processing on the target point information to eliminate short-term fluctuations in the target point information. In a specific embodiment of the present invention, the sliding average processing formula is:

[0067]

[0068] Among them, N A and P A They represent the distance unit and azimuth unit of the target after sliding average processing, round() represents the rounding function, K represents the size of the sliding average window, and N i and P i They respectively represent the distance unit and azimuth unit where the target is located at the i-th moment before sliding average processing.

[0069] The size K of the sliding average window represents the number of target point information considered when calculating the average value. The advantage of sliding average processing is that it is simple and easy to implement, but the disadvantage is that it is easily affected by outliers, so it is necessary to select a suitable window size. The present invention determines the size of the sliding average window based on the sampling frequency of the target point information and the fluctuation period of the target point information. Exemplarily, assuming that the sampling frequency is 20 times / second and the disturbance period is 5s, the size K of the sliding average window should be no less than 100 to ensure that short-term fluctuations in the target point information are eliminated. In this embodiment, after experimental verification, the size K of the sliding average window is preferably 120.

[0070] Step 4: Extract the target point echo signal based on the target point information processed in step 3, the clutter-suppressed radar echo signal obtained in step 2, and the guidance vector.

[0071] Since multi-pulse accumulation is used in the MTD (Moving Target Detection) process, the echo signal of the target point cannot be extracted directly by beamforming. Therefore, the present invention extracts the echo signal from the radar echo signal after clutter suppression and the guidance vector based on the target point information after the processing in step 3. In a specific embodiment of the present invention, the target point echo signal is extracted according to the target point information, the radar echo signal after clutter suppression and the guidance vector, which specifically includes:

[0072] Step 4.1: Extract the Nth signal from the radar echo signal after clutter suppression A The echo signal of each distance unit;

[0073] Step 4.2: Extract the Pth A The steering vector of the azimuth unit;

[0074] Step 4.3: According to the Nth A The echo signal of the first distance unit and the A The guidance vector of each azimuth unit is used to calculate the target point echo signal. The specific calculation formula is:

[0075] I(N A ,P A )=X N θ p (2)

[0076] Among them, X N Indicates the Nth A The echo signal of the range unit, θ p Indicates P A The steering vector of the azimuth unit, I(N A ,P A ) indicates the Nth A The distance unit is in P A The signal strength of each azimuth unit, N A and P A They represent the distance unit and azimuth unit of the target. Based on the radar echo signals of multiple pulses, the target point echo signals of a certain length can be obtained.

[0077] Step 5: Correct the target point echo signal extracted in step 4.

[0078] Since the position information of a static target may change within a small range, this change will cause the echo signal between two consecutive frames to be discontinuous (such as Figure 2 The obvious discontinuity of the target point echo signal will cause the micro-Doppler image to have spectrum diffusion phenomenon at the discontinuous moment (as shown in Figure 3As shown), the target point echo signal extracted in step 4 may not be used directly and needs to be corrected. In a specific embodiment of the present invention, the target point echo signal extracted in step 4 is corrected, specifically including:

[0079] Step 5.1: Perform first-order difference processing on the target point echo signal to obtain the difference between the data before and after each moment;

[0080] Step 5.2: Use the one-dimensional constant false alarm rate (CFAR) method to detect the difference between the data before and after each moment to obtain the discontinuity point; when the difference is greater than the threshold value, the moment is the discontinuity point, such as Figure 2 As shown;

[0081] Step 5.3: Calculate the amplitude and phase difference of the data before and after each discontinuity point, and use the amplitude and phase difference as the correction weight of the discontinuity point;

[0082] Step 5.4: According to the correction weight, the data after the discontinuous point is weighted point by point to make the amplitude and phase continuous, and obtain the corrected target point echo signal.

[0083] Figure 4 The time-frequency diagram of the target point echo signal after correction is shown in Figure 4 It can be seen that after correction, the micro-Doppler spectrum diffusion phenomenon of the target point echo signal is well suppressed.

[0084] Step 6: Determine whether the length of the target point echo signal corrected in step 5 reaches the set threshold.

[0085] By setting the threshold reasonably, the target point echo signal with too short length can be filtered out to prevent the micro-Doppler features extracted in the subsequent steps from being insufficient due to the short signal, thus affecting the subsequent recognition effect.

[0086] When the length of the corrected target point echo signal reaches the set threshold, go to step 7, that is, extract the micro-Doppler features of the target point echo signal; when the length of the corrected target point echo signal does not reach the set threshold, the target point echo signal is discarded.

[0087] Step 7: Extract the micro-Doppler features of the target point echo signal to obtain a micro-Doppler map.

[0088] In a specific embodiment of the present invention, STFT (short-time Fourier transform) is used to extract micro-Doppler features from the echo signal of the target point or the echo signal of the target in the abnormal target library. For a continuous echo signal x(t), its short-time Fourier transform X(τ,ω) can be expressed as:

[0089]

[0090] Among them, w(t) represents the window function (such as Hamming window, Hamming window, etc.), τ represents the time offset (that is, the position of the window), t represents time, and ω represents the angular frequency. When implementing STFT on a computer device, a discrete form needs to be used. For a discrete echo signal x[n], its discrete form of short-time Fourier transform X[m,k] can be expressed as:

[0091]

[0092] Where m represents the window index, H represents the sliding step size, w[n] represents the discrete window function, n represents the time index or frame index, k represents the frequency index, and N represents the length of the window function (i.e., the number of points of the short-time Fourier transform). The implementation process of STFT is as follows Figure 5 As shown, the window function is applied to the echo signal at different time periods, and then the signal in each window is FFTed to obtain the frequency characteristics of the signal over time, that is, the micro-Doppler map. In this embodiment, the length N of the window function is preferably 256, the sliding step H is preferably 16, and the number of sliding times M is preferably 255, then the length of the required echo signal is M×H+N=4336, that is, the threshold value can be set to 4336.

[0093] In another embodiment of the present invention, other time-frequency analysis methods may be used to extract micro-Doppler features.

[0094] Step 8: Annotate the micro-Doppler map obtained in step 7, and construct a sample data set based on the micro-Doppler map and its annotations.

[0095] For through-wall radar, the target types only include human targets (indicated by "1") and false targets (indicated by "0"). The target type is marked on each micro-Doppler map. For the micro-Doppler map of false targets (i.e. non-human targets), "0" is marked, such as Figure 6 As shown; for the micro-Doppler image of the human target, mark "1", such as Figure 7 shown.

[0096] Before constructing the sample dataset, the micro-Doppler images are preprocessed, including:

[0097] Data enhancement processing: Use rotation, flipping, etc. to transform the micro-Doppler map to increase the diversity of training samples and enhance the generalization ability of the model;

[0098] Resizing: resizing the micro-Doppler image to meet the model input requirements;

[0099] Normalization: subtract the mean and divide by the standard deviation so that the micro-Doppler map has zero mean and unit variance;

[0100] Normalization: Scale the pixel values ​​of the micro-Doppler image to a fixed range, usually [0, 1];

[0101] Quantization: Convert floating point pixels to integer values ​​between 0 and 255.

[0102] A sample data set is constructed based on the preprocessed micro-Doppler map and its annotations; the sample data set is divided into a training set and a test set, the training set is used to train the deep learning network model, and the test set is used to test the deep learning network model.

[0103] Step 9: Build a deep learning network model, use the sample data set to train and test the deep learning network model, and obtain a false alarm recognition model.

[0104] According to the balance between the requirements for accuracy and computing resources, this embodiment uses the MobileNetV3 model as the classifier, and builds the MobileNetV3 model based on the pytorch deep learning framework. According to the characteristics of the micro-Doppler map, the number of channels of the input layer of the MobileNetV3 model is changed to 1, that is, the input is changed from the original 224×224×3 to 224×224×1. Converting the input image from three channels (RGB) to a single channel (grayscale image) can significantly reduce memory usage, which is especially important for embedded systems or devices with limited computing resources. Compared with three-channel input, single-channel input reduces the amount of calculation by two-thirds, especially in convolution operations. The reduction in calculation can significantly improve the operational efficiency of the model.

[0105] The change in the model input layer will lead to a reduction in the number of parameters in subsequent layers. For example, the number of parameters in the first layer of convolution kernels will be reduced from 3×c×c to 1×c×c, where c represents the size of the convolution kernel. Due to the reduction in the amount of computation, the training time of the model will also be shortened accordingly, which helps to complete the development and tuning of the model faster.

[0106] In some application scenarios, the original data itself is single-channel (such as grayscale images, infrared images, etc.). By converting the input data to a single channel, the model can be more easily applied to these application scenarios. In some cases, removing color information may cause the detection accuracy of the model to decrease, especially in scenarios where color information is very important for target detection. However, the micro-Doppler image can be regarded as a single-channel grayscale image, so the change in the number of channels does not affect the recognition accuracy of the model itself, ensuring the recognition accuracy of the model.

[0107] During training, the input of the deep learning network model is the micro-Doppler map, and the output is the target category of the micro-Doppler map. All training and testing are performed on the Windows platform, and the CPU used on the Windows platform is Intel i7-8700 and the GPU is NVIDIARTX 1050ti. During training, the hyperparameters are set as follows: epochs is set to 200, batch size is set to 64, and the initial learning rate is set to 1e -2 , the final learning rate is 1e -7 The learning rate adjustment strategy is the cosine annealing algorithm, and Adam is used to optimize the model parameters.

[0108] Example 2

[0109] like Figure 8 As shown, a false alarm identification method provided by an embodiment of the present invention comprises the following steps:

[0110] Step 1: calling a false alarm recognition model; wherein the false alarm recognition model is constructed according to the false alarm recognition modeling method in Example 1 of the present application;

[0111] Step 2: Obtain the micro-Doppler map of the unknown target, and use the false alarm identification model to perform false alarm identification on the micro-Doppler map of the unknown target.

[0112] In a specific embodiment of the present invention, obtaining a micro-Doppler map of an unknown target includes:

[0113] Step 2.1: Obtain unknown target trajectory data; wherein the unknown target trajectory data includes unknown target point information at different times, and the unknown target point information includes the distance unit and the azimuth unit where the unknown target is located;

[0114] Step 2.2: Obtain radar echo signal and guidance vector after clutter suppression;

[0115] Step 2.3: Extract the unknown target point echo signal according to the unknown target point information, the radar echo signal after clutter suppression and the guidance vector, see formula (2) in Example 1 of the present application;

[0116] Step 2.4: When the length of the echo signal of the unknown target point reaches the set threshold, the micro-Doppler feature of the echo signal of the unknown target point is extracted to obtain the micro-Doppler map of the unknown target.

[0117] In a specific embodiment of the present invention, when the length of the corrected target point echo signal does not reach the set threshold, it is determined whether the target (i.e., the current target) corresponding to the unknown target point echo signal matches the target in the abnormal target library; if so, the process proceeds to step 2.4, i.e., performing micro-Doppler feature extraction on the echo signal of the target in the abnormal target library to obtain a micro-Doppler map of the unknown target; if not, the target point echo signal is discarded, i.e., the target point echo signal is not processed.

[0118] When the false alarm recognition model identifies an unknown target as a false target, it determines whether the false target matches the target in the abnormal target library. If so, the information of the target in the abnormal target library is updated; if not, the false target is entered into the abnormal target library.

[0119] The nearest neighbor algorithm is used to match the current target with the targets in the abnormal target library, that is, to find the abnormal target that is closest to the current target among the targets in the abnormal target library. If the current target matches the target in the abnormal target library, it indicates a match; if the current target does not match the target in the abnormal target library, it indicates a mismatch.

[0120] In the application stage, the micro-Doppler map of the unknown target is first obtained, and then it is input into the false alarm recognition model for classification to obtain the category of the unknown target. When the category of the unknown target is a false target, the target is filtered out to prevent the false target from affecting subsequent decisions, ensuring that only the real target is displayed by the terminal or the real target and the false target are clearly identified.

[0121] During a certain classification, if the target is determined to be a false target, the area where the target is located is determined to be an abnormal target area. Since abnormal targets may not persist in the tracking process, in order to ensure that such targets can quickly respond to the false alarm identification and filtering process, it is necessary to perform special management on targets that behave abnormally in the entire false alarm identification and filtering process. Abnormal target management mainly includes abnormal target information creation, abnormal target information association and update, etc. Among them, abnormal target information association and update are: when a false target matches a target in the abnormal target library, the target in the abnormal target library is updated according to the information of the false target; abnormal target information creation is: when a false target does not match any target in the abnormal target library, a new target is created in the abnormal target library according to the information of the false target. Abnormal target management will persist the location of the false target, target echo data and other information.

[0122] When extracting micro-Doppler features, if the length of the echo signal of the unknown target point does not reach the set threshold, and the current target matches a target in the abnormal target library, the echo signal of the target in the abnormal target library is directly used to extract the micro-Doppler features.

[0123] Targets judged as abnormal are not necessarily false alarms, because there are at least two situations: the first is that the model may misjudge targets that are difficult to distinguish; the second is that false alarms are dominant in the abnormal target area at the previous moment, but the real target becomes dominant due to the appearance of the target at the next moment. In both cases, the results of one or a few judgments cannot assume that the target appearing in the abnormal target area is a false alarm, so repeated judgments are required, and finally the category that appears most is considered to be the real category of the target. Through abnormal target management, it is ensured that false targets can still respond quickly to the false alarm identification process after tracking is interrupted.

[0124] The micro-Doppler effect refers to the frequency change in radar echoes caused by the different speeds of different parts of the target. This change can provide information about the physical characteristics of the target. Deep learning algorithms can automatically extract features from a large number of samples. Through the learning ability of multi-layer neural networks, it can more accurately identify the difference between human targets and false targets. Combining micro-Doppler features and deep learning algorithms can more accurately distinguish between real targets and false targets in complex backgrounds, thereby reducing the false alarm rate.

[0125] In today's increasingly complex electromagnetic environment, radar systems face a variety of man-made or natural interference. Using micro-Doppler features, even in the presence of interference, the true attributes of the target can be identified by analyzing specific frequency changes. After deep learning network model training and learning, these interference signals can be effectively identified and filtered out, further enhancing the robustness of the system.

[0126] The working environment of radars in different application scenarios varies greatly. Deep learning methods have strong generalization capabilities, that is, they can adapt to new environmental changes by adjusting model parameters. Micro-Doppler features show different characteristics for different types of targets (such as pedestrians, vehicles, drones, etc.), which makes cross-scenario applications possible. The deep learning framework optimized by the input layer can achieve rapid response while ensuring accuracy; micro-Doppler processing usually involves efficient algorithms such as fast Fourier changes, which are suitable for real-time processing of large amounts of data. Therefore, combining these two technologies can achieve real-time monitoring of dynamic environments while maintaining high accuracy. In summary, the present invention significantly improves the working efficiency and reliability of the radar system by improving detection accuracy, enhancing anti-interference ability, improving adaptability, and maintaining good real-time performance.

[0127] Example 3

[0128] An embodiment of the present application also provides an electronic device, which includes: a memory, a processor, and a computer program / instructions stored in the memory, and the processor executes the computer program / instructions to implement the false alarm identification modeling method in embodiment 1 of the present application or the false alarm identification method in embodiment 2 of the present application.

[0129] Although not shown, the electronic device includes a processor, which can perform various appropriate operations and processes according to the program and / or data stored in the read-only memory (ROM) or the program and / or data loaded from the storage portion into the random access memory (RAM). The processor can be a multi-core processor, or it can include multiple processors. In some embodiments, the processor can include a general main processor and one or more special coprocessors, such as a central processing unit, a graphics processing unit (GPU), a neural network processor (NPU), a digital signal processor (DSP), etc. In RAM, various programs and data required for device operation are also stored. The processor, ROM and RAM are connected to each other via a bus. The input / output (I / O) interface is also connected to the bus.

[0130] The processor and memory are used together to execute the program / instructions stored in the memory. When the program / instructions are executed by the computer, the methods, steps or functions described in the above embodiments can be implemented.

[0131] Although not shown, the embodiments of the present application also provide a computer-readable storage medium on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the false alarm identification modeling method in Embodiment 1 of the present application or the false alarm identification method in Embodiment 2 of the present application is implemented.

[0132] Readable storage media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0133] Although not shown, the embodiment of the present application also provides a computer program product, including: a computer program / instruction, which, when executed by a processor, implements the false alarm identification modeling method in Embodiment 1 of the present application or the false alarm identification method in Embodiment 2 of the present application.

[0134] What is disclosed above is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or modifications within the technical scope disclosed in the present invention, which should be covered within the protection scope of the present invention.

Claims

1. A false alarm identification modeling method, characterized in that: The modeling method comprises: Acquire target trajectory data; wherein the target trajectory data includes target point information at different times, and the target point information includes a distance unit and an azimuth unit where the target is located; Obtain radar echo signals and guidance vectors after clutter suppression; Extracting the target point echo signal according to the target point information, the radar echo signal after clutter suppression and the guidance vector; When the length of the target point echo signal reaches a set threshold, micro-Doppler feature extraction is performed on the target point echo signal to obtain a micro-Doppler map; Annotating the micro-Doppler image, and constructing a sample data set according to the micro-Doppler image and its annotations; A deep learning network model is constructed, and the deep learning network model is trained and tested using the sample data set to obtain a false alarm recognition model.

2. The false alarm identification modeling method according to claim 1, characterized in that: Before extracting the target point echo signal, the modeling method further includes performing sliding average processing on the target point information. The specific processing formula is: Among them, N A and P A They represent the distance unit and azimuth unit of the target after sliding average processing, round() represents the rounding function, K represents the size of the sliding average window, and N i and P i They respectively represent the distance unit and azimuth unit where the target is located at the i-th moment before sliding average processing.

3. The false alarm identification modeling method according to claim 1, characterized in that: Extracting the target point echo signal according to the target point information, the radar echo signal after clutter suppression and the guidance vector, specifically includes: Extract the Nth A Extract the echo signal of the Pth distance unit from the guidance vector A The steering vector of the azimuth unit; According to the Nth A The echo signal of the first distance unit and the A The guidance vector of each azimuth unit is used to calculate the target point echo signal. The specific calculation formula is: I(N A ,P A )=X N θ p ; Among them, X N Indicates the Nth A The echo signal of the range unit, θ p Indicates P A The steering vector of the azimuth unit, I(N A ,P A ) indicates the Nth A The distance unit is in P A The signal strength of each azimuth unit, N A and P A They respectively represent the distance unit and azimuth unit where the target is located.

4. The false alarm identification modeling method according to claim 1, characterized in that: When the length of the target point echo signal reaches a set threshold, before extracting the micro-Doppler feature of the target point echo signal, the modeling method further includes correcting the target point echo signal, specifically including: Performing first-order difference processing on the target point echo signal to obtain the difference between the data before and after each moment; The one-dimensional constant false alarm rate method is used to detect the difference of the data before and after each moment to obtain the discontinuity point; Calculate the amplitude and phase difference of the data before and after each discontinuous point, and use the amplitude and phase difference as the correction weight of the discontinuous point; According to the correction weight, the data after the discontinuous point is weighted point by point to obtain a corrected target point echo signal.

5. The false alarm identification modeling method according to claim 1, characterized in that: Short-time Fourier transform is used to extract micro-Doppler features of the target point echo signal.

6. The false alarm identification modeling method according to any one of claims 1 to 5, characterized in that: The deep learning network model selects the MobileNetV3 model, and builds the MobileNetV3 model based on the pytorch deep learning framework, and changes the number of channels of the input layer of the MobileNetV3 model to 1.

7. A false alarm identification method, characterized in that: The identification method comprises: Calling a false alarm recognition model; wherein the false alarm recognition model is constructed according to the false alarm recognition modeling method according to any one of claims 1 to 6; A micro-Doppler map of an unknown target is obtained, and false alarm identification is performed on the micro-Doppler map of the unknown target using the false alarm identification model.

8. The false alarm identification method according to claim 7, characterized in that: Get micro-Doppler images of unknown targets, including: Acquire unknown target trajectory data; wherein the unknown target trajectory data includes unknown target point information at different times, and the unknown target point information includes the distance unit and the azimuth unit where the unknown target is located; Obtain radar echo signals and guidance vectors after clutter suppression; Extracting the unknown target point echo signal according to the unknown target point information, the radar echo signal after clutter suppression and the guidance vector; When the length of the unknown target point echo signal reaches a set threshold, micro-Doppler feature extraction is performed on the unknown target point echo signal to obtain a micro-Doppler map of the unknown target.

9. The false alarm identification method according to claim 8, characterized in that: The identification method further includes abnormal target management, and the abnormal target management specifically includes: When the length of the unknown target point echo signal does not reach the set threshold, it is determined whether the target corresponding to the unknown target point echo signal matches the target in the abnormal target library. If so, the micro-Doppler feature extraction is performed on the echo signal of the target in the abnormal target library to obtain the micro-Doppler map of the unknown target; if not, the unknown target point echo signal is discarded; When an unknown target is identified as a false target, it is determined whether the false target matches the target in the abnormal target library. If so, the information of the target in the abnormal target library is updated; if not, the false target is entered into the abnormal target library.

10. An electronic device comprising a memory, a processor, and a computer program / instruction stored in the memory, characterized in that: The processor executes the computer program / instructions to implement the false alarm identification modeling method according to any one of claims 1 to 6 or the false alarm identification method according to claims 7 to 9.

11. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instruction is executed by a processor, the false alarm identification modeling method according to any one of claims 1 to 6 or the false alarm identification method according to claims 7 to 9 is implemented.

12. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the false alarm identification modeling method according to any one of claims 1 to 6 or the false alarm identification method according to claims 7 to 9 is implemented.

Images