Radar recognition algorithm based on improved recurrent neural network

By improving recurrent neural network and radar technology, extracting and classifying human posture characteristics, the problem of low recognition accuracy of optical systems under environmental interference is solved, and high accuracy and robust human posture recognition is achieved.

CN118938153APending Publication Date: 2024-11-12WUXI XINCHENGHUI INFORMATION TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410894441.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-04
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

When identifying human postures, existing optical systems are susceptible to environmental factors such as lighting conditions, weather and smoke, resulting in low recognition accuracy and poor stability, and feature selection may ignore important information, resulting in unreliable identification results.

Method used

Using radar recognition algorithm based on improved recurrent neural network, the radar transmits and receives time domain echo data, performs three-dimensional Fourier transform and phase offset, and combines supervised learning and backpropagation algorithm to extract and classify human posture features to achieve Storage and re-identification of radar imaging results.

Benefits of technology

It realizes high accuracy recognition of human posture, avoids feature omissions, can work around the clock in various environments, has strong system robustness, high recognition accuracy, and reduces dependence on environmental factors.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure BDA0004928508300000031
    Figure BDA0004928508300000031
  • Figure BDA0004928508300000032
    Figure BDA0004928508300000032
  • Figure FDA0004928508280000021
    Figure FDA0004928508280000021
Patent Text Reader

Abstract

The invention discloses a radar recognition algorithm based on an improved recurrent neural network. The radar recognition algorithm comprises the following steps that A, a radar recognizes the posture of a human body and carries out data calculation and marking; the method comprises the following steps: A1, starting a radar, transmitting radar waves by the radar, and receiving time domain echo data d '2 (x, y, z = 0, t) of the radar waves at the position of z = 0; according to the method, radar echoes are analyzed, human postures are recognized in real time, echo signal time-frequency diagrams are recognized and classified through recurrent neural network clamping and adopting a phase shift formula, three-dimensional Fourier transform and a back propagation algorithm, so that the situation of feature omission is avoided, and the accuracy of human posture recognition is improved. The problems that due to the fact that optical systems such as a camera are interfered by environmental factors such as illumination conditions, weather and smoke, especially at night, the recognition effect is greatly reduced, important clues can be ignored and supplementary information can be deleted in the feature selection process, time and energy are consumed, and the recognition accuracy is not high are solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of human body posture recognition computing, and in particular to a radar recognition algorithm based on an improved recurrent neural network. Background Art

[0002] Human posture recognition methods aim to identify different types of human target postures such as standing, sitting, squatting, etc., and are an important part of human perception tasks. Most human posture recognition methods are designed based on optical systems and can directly identify the posture of human targets from photos or videos. However, optical systems such as cameras first acquire images and then perform image recognition processing on the acquired images. In this way, due to interference from environmental factors such as lighting conditions, weather, and smoke, especially at night, the recognition effect is greatly reduced, the recognition accuracy is not high, and even unrecognizable situations occur. The stability is very poor, and important clues may be ignored and supplementary information may be deleted in the process of feature selection. If there is an error in the statistics of the features, the results will not be very reliable. It is not only time-consuming and labor-intensive, but also has a low recognition accuracy. For this reason, we propose a xx to solve this problem. Summary of the invention

[0003] The purpose of the present invention is to provide a radar recognition algorithm based on an improved recurrent neural network, which has the advantage of being easy to recognize and calculate human body postures, and solves the problem that an optical system such as a camera first acquires an image and then performs image recognition processing on the acquired image. In this way, due to interference from environmental factors such as lighting conditions, weather, and smoke, especially at night, the recognition effect is greatly reduced, the recognition accuracy is not high, and even recognition cannot be achieved. The stability is very poor, and important clues may be ignored and supplementary information may be deleted during the feature selection process. If errors occur in the statistics of the features, the results will not be very reliable, which not only wastes time and energy but also has a low recognition accuracy.

[0004] To achieve the above object, the present invention provides the following technical solution: a radar recognition algorithm based on an improved recurrent neural network, wherein the radar recognition algorithm comprises the following steps:

[0005] A. Radar recognizes human body posture and performs data calculation and annotation;

[0006] A1. First, the radar is turned on to transmit radar waves and receive the time domain echo data d′2 (x, y, z=0, t) of the radar waves at z=0.

[0007] A2, taking a three-dimensional Fourier transform of the time domain echo data to obtain frequency domain echo data D2 (kx,ky,z=0,ω);

[0008] A3. Use the phase shift formula to perform phase shift on the frequency domain echo data D2(kx,ky,z=0,ω) to obtain the frequency domain echo data D2(kx,ky,z,ω) of the entire imaging area.

[0009] A4. Then, the radar is used to collect and receive signals, collect these signals and make a source data set; then, k types of action categories of the target are pre-specified, and these data are labeled, that is, the human posture category corresponding to each radar signal is specified;

[0010] B. Send the data in the source data set into the recurrent neural network model, and extract the features containing human posture information through the N-layer feature extraction module; send these features into the classifier and classify them into different posture categories;

[0011] C. Since the source data set has been labeled, the supervised learning method is used to construct a loss function between the labeled category and the predicted category of the recurrent neural network model, that is, the posture category loss function. The labeled category can be set to o1, and the predicted category of the recurrent neural network model is o2. By using the back propagation algorithm on the loss function, the parameters of the feature extraction module and the classifier are updated;

[0012] D. Then, NUFFT is used to perform inverse transformation to the space-time domain d′2(x, y, z, t=0)), thereby obtaining the radar imaging result after eliminating the influence of the wall;

[0013] E. Finally, the radar imaging results are transferred to the memory for storage. When recognition and calculation are needed again, the recurrent neural network technology is used to select the value with the highest degree of conformity from the source data in the memory, and then the human body posture can be recognized and determined.

[0014] Preferably, in step A, it is taken into account that the radar pulse signal may experience pulse loss during transmission and may be interfered with by other surrounding electromagnetic radiation sources and environmental factors such as ground clutter and sea clutter. The various parameters of the actually received radar pulse signal may be slightly different from the original data set of the radiation source description word. The radar pulse signals obtained may be collected and recorded based on the actual working performance indicators of the radar and a comprehensive evaluation of the external electromagnetic environment, and the working modes of the actually measured radar signals may be identified.

[0015] Preferably, in step A, the radar direct echo, environmental clutter, and target echo; the direct echo of the wall can be eliminated by using an average cancellation method.

[0016] Preferably, the calculation formula of o1 in step C is

[0017]

[0018] Preferably, the calculation formula of O2 in step C is

[0019]

[0020] Preferably, the memory mentioned in step E adopts W25X10CLSNIG25X10CLNIG SMD SOP-8 memory.

[0021] Preferably, the radar collection in step A adopts dual antenna transmission and dual antenna reception.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] The present invention recognizes human posture in real time by analyzing radar echoes, and recognizes and classifies echo signal time-frequency diagrams by using a recurrent neural network and a phase shift formula, a three-dimensional Fourier transform, and a back propagation algorithm, thereby avoiding feature omissions. The method is implemented based on radar. Compared with optical cameras, the invention is not affected by factors such as lighting conditions, weather, and smoke, and can work all day and all weather. It has strong feasibility and can effectively perform high-accuracy recognition and calculation. At the same time, it is not easily affected by various environmental factors, and the system has good robustness. It solves the problem that optical systems such as cameras first acquire images and then perform image recognition processing on the acquired images. In this way, due to the interference of environmental factors such as lighting conditions, weather, and smoke, especially at night, the recognition effect is greatly reduced, the recognition accuracy is not high, and even the recognition cannot be recognized. The stability is very poor, and important clues may be ignored and supplementary information may be deleted in the process of feature selection. If errors occur in the statistics of features, the results will not be very reliable, which not only wastes time and energy but also has low recognition accuracy. DETAILED DESCRIPTION

[0024] A radar recognition algorithm based on an improved recurrent neural network, the radar recognition algorithm includes the following steps: A, radar recognition of human body posture and data calculation and labeling;

[0025] A1. First, the radar is turned on to transmit radar waves and receive the time domain echo data d′2 (x, y, z=0, t) of the radar waves at z=0.

[0026] A2, taking a three-dimensional Fourier transform of the time domain echo data to obtain frequency domain echo data D2 (kx,ky,z=0,ω);

[0027] A3. Use the phase shift formula to perform phase shift on the frequency domain echo data D2(kx,ky,z=0,ω) to obtain the frequency domain echo data D2(kx,ky,z,ω) of the entire imaging area.

[0028] A4. Then, the radar is used to collect and receive signals, collect these signals and make a source data set; then, k types of action categories of the target are pre-specified, and these data are labeled, that is, the human posture category corresponding to each radar signal is specified;

[0029] B. Send the data in the source data set into the recurrent neural network model, and extract the features containing human posture information through the N-layer feature extraction module; send these features into the classifier and classify them into different posture categories;

[0030] C. Since the source data set has been labeled, the supervised learning method is used to construct a loss function between the labeled category and the predicted category of the recurrent neural network model, that is, the posture category loss function. The labeled category can be set to o1, and the predicted category of the recurrent neural network model is o2. By using the back propagation algorithm on the loss function, the parameters of the feature extraction module and the classifier are updated;

[0031] D. Then, NUFFT is used to perform inverse transformation to the space-time domain d′2(x, y, z, t=0)), thereby obtaining the radar imaging result after eliminating the influence of the wall;

[0032] E. Finally, the radar imaging results are transferred to the memory for storage. When recognition and calculation are needed again, the recurrent neural network technology is used to select the value with the highest degree of conformity from the source data in the memory, and then the human body posture can be recognized and determined.

[0033] Embodiment 2:

[0034] In the first embodiment, the following steps are added:

[0035] In step A, it is considered that the radar pulse signal may experience pulse loss during transmission and may also be interfered by other surrounding electromagnetic radiation sources and environmental factors such as ground clutter and sea clutter. The various parameters of the actually received radar pulse signal may be slightly different from the original data set of the radiation source description word. Based on the actual working performance indicators of the radar and the comprehensive evaluation of the external electromagnetic environment, the obtained radar pulse signals can be collected and recorded, and the working modes of each actually measured radar signal can be identified.

[0036] Its radar recognition algorithm includes the following steps:

[0037] A. Radar recognizes human body posture and performs data calculation and annotation;

[0038] A1. First, the radar is turned on to transmit radar waves and receive the time domain echo data d′2 (x, y, z=0, t) of the radar waves at z=0.

[0039] A2, taking a three-dimensional Fourier transform of the time domain echo data to obtain frequency domain echo data D2 (kx,ky,z=0,ω);

[0040] A3. Use the phase shift formula to perform phase shift on the frequency domain echo data D2(kx,ky,z=0,ω) to obtain the frequency domain echo data D2(kx,ky,z,ω) of the entire imaging area.

[0041] A4. Then, the radar is used to collect and receive signals, collect these signals and make a source data set; then, k types of action categories of the target are pre-specified, and these data are labeled, that is, the human posture category corresponding to each radar signal is specified;

[0042] B. Send the data in the source data set into the recurrent neural network model, and extract the features containing human posture information through the N-layer feature extraction module; send these features into the classifier and classify them into different posture categories;

[0043] C. Since the source data set has been labeled, the supervised learning method is used to construct a loss function between the labeled category and the predicted category of the recurrent neural network model, that is, the posture category loss function. The labeled category can be set to o1, and the predicted category of the recurrent neural network model is o2. By using the back propagation algorithm on the loss function, the parameters of the feature extraction module and the classifier are updated;

[0044] D. Then, NUFFT is used to perform inverse transformation to the space-time domain d′2(x, y, z, t=0)), thereby obtaining the radar imaging result after eliminating the influence of the wall;

[0045] E. Finally, the radar imaging results are transferred to the memory for storage. When recognition and calculation are needed again, the recurrent neural network technology is used to select the value with the highest degree of conformity from the source data in the memory, and then the human body posture can be recognized and determined.

[0046] Embodiment three:

[0047] In the second embodiment, the following steps are added:

[0048] In step A, the radar's direct echo, environmental clutter, and target echo; the average cancellation method can eliminate the direct echo of the wall.

[0049] Its radar recognition algorithm includes the following steps:

[0050] A. Radar recognizes human body posture and performs data calculation and annotation;

[0051] A1. First, the radar is turned on to transmit radar waves and receive the time domain echo data d′2 (x, y, z=0, t) of the radar waves at z=0.

[0052] A2, taking a three-dimensional Fourier transform of the time domain echo data to obtain frequency domain echo data D2 (kx,ky,z=0,ω);

[0053] A3. Use the phase shift formula to perform phase shift on the frequency domain echo data D2(kx,ky,z=0,ω) to obtain the frequency domain echo data D2(kx,ky,z,ω) of the entire imaging area.

[0054] A4. Then, the radar is used to collect and receive signals, collect these signals and make a source data set; then, k types of action categories of the target are pre-specified, and these data are labeled, that is, the human posture category corresponding to each radar signal is specified;

[0055] B. Send the data in the source data set into the recurrent neural network model, and extract the features containing human posture information through the N-layer feature extraction module; send these features into the classifier and classify them into different posture categories;

[0056] C. Since the source data set has been labeled, the supervised learning method is used to construct a loss function between the labeled category and the predicted category of the recurrent neural network model, that is, the posture category loss function. The labeled category can be set to o1, and the predicted category of the recurrent neural network model is o2. By using the back propagation algorithm on the loss function, the parameters of the feature extraction module and the classifier are updated;

[0057] D. Then, NUFFT is used to perform inverse transformation to the space-time domain d′2(x, y, z, t=0)), thereby obtaining the radar imaging result after eliminating the influence of the wall;

[0058] E. Finally, the radar imaging results are transferred to the memory for storage. When recognition and calculation are needed again, the recurrent neural network technology is used to select the value with the highest degree of conformity from the source data in the memory, and then the human body posture can be recognized and determined.

[0059] Embodiment 4:

[0060] In the third embodiment, the following steps are added:

[0061] The calculation formula for o1 in step C is

[0062] Its radar recognition algorithm includes the following steps:

[0063] A. Radar recognizes human body posture and performs data calculation and annotation;

[0064] A1. First, the radar is turned on to transmit radar waves and receive the time domain echo data d′2 (x, y, z=0, t) of the radar waves at z=0.

[0065] A2, taking a three-dimensional Fourier transform of the time domain echo data to obtain frequency domain echo data D2 (kx,ky,z=0,ω);

[0066] A3. Use the phase shift formula to perform phase shift on the frequency domain echo data D2(kx,ky,z=0,ω) to obtain the frequency domain echo data D2(kx,ky,z,ω) of the entire imaging area.

[0067] A4. Then, the radar is used to collect and receive signals, collect these signals and make a source data set; then, k types of action categories of the target are pre-specified, and these data are labeled, that is, the human posture category corresponding to each radar signal is specified;

[0068] B. Send the data in the source data set into the recurrent neural network model, and extract the features containing human posture information through the N-layer feature extraction module; send these features into the classifier and classify them into different posture categories;

[0069] C. Since the source data set has been labeled, the supervised learning method is used to construct a loss function between the labeled category and the predicted category of the recurrent neural network model, that is, the posture category loss function. The labeled category can be set to o1, and the predicted category of the recurrent neural network model is o2. By using the back propagation algorithm on the loss function, the parameters of the feature extraction module and the classifier are updated;

[0070] D. Then, NUFFT is used to perform inverse transformation to the space-time domain d′2(x, y, z, t=0)), thereby obtaining the radar imaging result after eliminating the influence of the wall;

[0071] E. Finally, the radar imaging results are transferred to the memory for storage. When recognition and calculation are needed again, the recurrent neural network technology is used to select the value with the highest degree of conformity from the source data in the memory, and then the human body posture can be recognized and determined.

[0072] Embodiment five:

[0073] In the fourth embodiment, the following steps are added:

[0074] The calculation formula for 02 in step C is

[0075] Its radar recognition algorithm includes the following steps:

[0076] A. Radar recognizes human body posture and performs data calculation and annotation;

[0077] A1. First, the radar is turned on to transmit radar waves and receive the time domain echo data d′2 (x, y, z=0, t) of the radar waves at z=0.

[0078] A2, taking a three-dimensional Fourier transform of the time domain echo data to obtain frequency domain echo data D2 (kx,ky,z=0,ω);

[0079] A3. Use the phase shift formula to perform phase shift on the frequency domain echo data D2(kx,ky,z=0,ω) to obtain the frequency domain echo data D2(kx,ky,z,ω) of the entire imaging area.

[0080] A4. Then, the radar is used to collect and receive signals, collect these signals and make a source data set; then, k types of action categories of the target are pre-specified, and these data are labeled, that is, the human posture category corresponding to each radar signal is specified;

[0081] B. Send the data in the source data set into the recurrent neural network model, and extract the features containing human posture information through the N-layer feature extraction module; send these features into the classifier and classify them into different posture categories;

[0082] C. Since the source data set has been labeled, the supervised learning method is used to construct a loss function between the labeled category and the predicted category of the recurrent neural network model, that is, the posture category loss function. The labeled category can be set to o1, and the predicted category of the recurrent neural network model is o2. By using the back propagation algorithm on the loss function, the parameters of the feature extraction module and the classifier are updated;

[0083] D. Then, NUFFT is used to perform inverse transformation to the space-time domain d′2(x, y, z, t=0)), thereby obtaining the radar imaging result after eliminating the influence of the wall;

[0084] E. Finally, the radar imaging results are transferred to the memory for storage. When recognition and calculation are needed again, the recurrent neural network technology is used to select the value with the highest degree of conformity from the source data in the memory, and then the human body posture can be recognized and determined.

[0085] Embodiment six:

[0086] In the fifth embodiment, the following steps are added:

[0087] : The memory mentioned in step E uses W25X10CLSNIG25X10CLNIG SMD SOP-8 memory.

[0088] Its radar recognition algorithm includes the following steps:

[0089] A. Radar recognizes human body posture and performs data calculation and annotation;

[0090] A1. First, the radar is turned on to transmit radar waves and receive the time domain echo data d′2 (x, y, z=0, t) of the radar waves at z=0.

[0091] A2, taking a three-dimensional Fourier transform of the time domain echo data to obtain frequency domain echo data D2 (kx,ky,z=0,ω);

[0092] A3. Use the phase shift formula to perform phase shift on the frequency domain echo data D2(kx,ky,z=0,ω) to obtain the frequency domain echo data D2(kx,ky,z,ω) of the entire imaging area.

[0093] A4. Then, the radar is used to collect and receive signals, collect these signals and make a source data set; then, k types of action categories of the target are pre-specified, and these data are labeled, that is, the human posture category corresponding to each radar signal is specified;

[0094] B. Send the data in the source data set into the recurrent neural network model, and extract the features containing human posture information through the N-layer feature extraction module; send these features into the classifier and classify them into different posture categories;

[0095] C. Since the source data set has been labeled, the supervised learning method is used to construct a loss function between the labeled category and the predicted category of the recurrent neural network model, that is, the posture category loss function. The labeled category can be set to o1, and the predicted category of the recurrent neural network model is o2. By using the back propagation algorithm on the loss function, the parameters of the feature extraction module and the classifier are updated;

[0096] D. Then, NUFFT is used to perform inverse transformation to the space-time domain d′2(x, y, z, t=0)), thereby obtaining the radar imaging result after eliminating the influence of the wall;

[0097] E. Finally, the radar imaging results are transferred to the memory for storage. When recognition and calculation are needed again, the recurrent neural network technology is used to select the value with the highest degree of conformity from the source data in the memory, and then the human body posture can be recognized and determined.

[0098] Embodiment seven:

[0099] In the sixth embodiment, the following steps are added:

[0100] The radar acquisition in step A adopts dual antenna transmission and dual antenna reception.

[0101] Its radar recognition algorithm includes the following steps:

[0102] A. Radar recognizes human body posture and performs data calculation and annotation;

[0103] A1. First, the radar is turned on to transmit radar waves and receive the time domain echo data d′2 (x, y, z=0, t) of the radar waves at z=0.

[0104] A2, taking a three-dimensional Fourier transform of the time domain echo data to obtain frequency domain echo data D2 (kx,ky,z=0,ω);

[0105] A3. Use the phase shift formula to perform phase shift on the frequency domain echo data D2(kx,ky,z=0,ω) to obtain the frequency domain echo data D2(kx,ky,z,ω) of the entire imaging area.

[0106] A4. Then, the radar is used to collect and receive signals, collect these signals and make a source data set; then, k types of action categories of the target are pre-specified, and these data are labeled, that is, the human posture category corresponding to each radar signal is specified;

[0107] B. Send the data in the source data set into the recurrent neural network model, and extract the features containing human posture information through the N-layer feature extraction module; send these features into the classifier and classify them into different posture categories;

[0108] C. Since the source data set has been labeled, the supervised learning method is used to construct a loss function between the labeled category and the predicted category of the recurrent neural network model, that is, the posture category loss function. The labeled category can be set to o1, and the predicted category of the recurrent neural network model is o2. By using the back propagation algorithm on the loss function, the parameters of the feature extraction module and the classifier are updated;

[0109] D. Then, NUFFT is used to perform inverse transformation to the space-time domain d′2(x, y, z, t=0)), thereby obtaining the radar imaging result after eliminating the influence of the wall;

[0110] E. Finally, the radar imaging results are transferred to the memory for storage. When recognition and calculation are needed again, the recurrent neural network technology is used to select the value with the highest degree of conformity from the source data in the memory, and then the human body posture can be recognized and determined.

[0111] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0112] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A radar recognition algorithm based on improved recurrent neural network, Features: Its radar recognition algorithm includes the following steps: A. Radar recognizes human body posture and performs data calculation and annotation; A1. First, the radar is turned on to transmit radar waves and receive the time domain echo data d′2 (x, y, z=0, t) of the radar waves at z=0. A2, taking a three-dimensional Fourier transform of the time domain echo data to obtain frequency domain echo data D2 (kx,ky,z=0,ω); A3. Use the phase shift formula to perform phase shift on the frequency domain echo data D2(kx,ky,z=0,ω) to obtain the frequency domain echo data D2(kx,ky,z,ω) of the entire imaging area. A4. Then, the radar is used to collect and receive signals, collect these signals and make a source data set; then, k types of action categories of the target are pre-specified, and these data are labeled, that is, the human posture category corresponding to each radar signal is specified; B. Send the data in the source data set into the recurrent neural network model, and extract the features containing human posture information through the N-layer feature extraction module; send these features into the classifier and classify them into different posture categories; C. Since the source data set has been labeled, the supervised learning method is used to construct a loss function between the labeled category and the predicted category of the recurrent neural network model, that is, the posture category loss function. The labeled category can be set to o1, and the predicted category of the recurrent neural network model is o2. By using the back propagation algorithm on the loss function, the parameters of the feature extraction module and the classifier are updated; D. Then, NUFFT is used to perform inverse transformation to the space-time domain d′2(x, y, z, t=0)), thereby obtaining the radar imaging result after eliminating the influence of the wall; E. Finally, the radar imaging results are transferred to the memory for storage. When recognition and calculation are needed again, the recurrent neural network technology is used to select the value with the highest degree of conformity from the source data in the memory, and then the human body posture can be recognized and determined.

2. According to claim 1, a radar recognition algorithm based on an improved recurrent neural network, Features: In step A, it is considered that the radar pulse signal may experience pulse loss during transmission and may be interfered by other surrounding electromagnetic radiation sources and environmental factors such as ground clutter and sea clutter. The various parameters of the actually received radar pulse signal may be slightly different from the original data set of the radiation source description word. Based on the actual working performance indicators of the radar and the comprehensive evaluation of the external electromagnetic environment, the obtained radar pulse signals can be collected and recorded, and the working modes of each actually measured radar signal can be identified.

3. According to the radar recognition algorithm based on improved recurrent neural network described in claim 1, Features: In step A, the radar's direct echo, environmental clutter, and target echo can all be eliminated by using an average cancellation method to eliminate the direct echo from the wall.

4. According to claim 1, a radar recognition algorithm based on an improved recurrent neural network, Features: The calculation formula of o1 in step C is:

5. The radar recognition algorithm based on improved recurrent neural network according to claim 1, Features: The calculation formula of 02 in step C is:

6. The radar recognition algorithm based on improved recurrent neural network according to claim 1, Features: The memory mentioned in the step E adopts W25X10CLSNIG25X10CLNIG SMD SOP-8 memory.

7. The radar recognition algorithm based on improved recurrent neural network according to claim 1, Features: The radar acquisition in step A adopts dual antenna transmission and dual antenna reception.