A method and system for personnel classification and recognition based on neural network

Through the neural network and the improved CNN-LSTM model combined with the intelligent safety helmet, the problem of the existing technology being unable to accurately judge the position and identity of personnel is solved, and the accurate identification and safety management of personnel in the construction area is achieved, and construction safety is improved.

CN118349885BActive Publication Date: 2024-11-26HANGZHOU MAIJIAN INTELLIGENT TECH CO LTD
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202410454016.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-16
Publication Date
2024-11-26
Estimated Expiration
2044-04-16

AI Technical Summary

Technical Problem

The prior art cannot accurately determine the location and identity of personnel, which leads to construction personnel entering live areas by mistake, posing safety hazards.

Method used

Using a neural network-based personnel classification recognition method, by obtaining perceptual radar detection signals, constructing personnel and non-personnel signal label data sets, using improved feature fusion CNN-LSTM neural network model for classification recognition, and combining the positioning information of intelligent safety helmets , determine whether the target person is within the preset range.

Benefits of technology

Accurate perception, positioning and identification of personnel in the construction area is achieved, the risk of accidentally entering live areas is avoided, and the efficiency of construction safety management is improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118349885B_ABST
    Figure CN118349885B_ABST
Patent Text Reader

Abstract

The present invention provides a personnel classification and identification method and system based on a neural network, belonging to the field of radar identification technology, and specifically comprising: acquiring detection signals of two or more perception radars for detecting personnel, performing data processing on the detection signals of the perception radars, constructing a personnel and non-personnel signal label data set, constructing a neural network model, and performing supervised training on the label data set to realize the construction of a classification and identification model for personnel and other targets, outputting the coordinates of a target person according to a classification result of the classification and identification model, continuously sending a representative signal to a cloud platform at a set time interval, judging whether there is an operating person near a physical channel according to the coordinates of the target person, and identifying the target person as an operating person or an illegal intruder based on a position matching situation between the two within an error range, thereby ensuring the accuracy of classification and identification.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of radar recognition technology, and in particular relates to a method and system for classifying and identifying personnel based on a neural network. Background Art

[0002] During the maintenance and overhaul of power facilities, there are often energized areas and non-energized areas. Therefore, during the maintenance process, it is inevitable that construction workers will mistakenly enter the energized area. Therefore, how to prevent construction and maintenance personnel from mistakenly entering the energized area has become a technical problem that needs to be solved urgently.

[0003] In order to solve the above technical problems, the existing technical solution connects two adjacent unit guardrails through laser signals, and no intermediate pull rope or fence frame is required. The structure is simple, and the laying is fast and reliable. The controller is connected with an alarm device and an intelligent searchlight device, which can realize all-round prompting, warning and protection of the fence range. In addition, the classification and recognition of radar waveforms are realized in the invention patent CN112487933A "A radar waveform recognition method and system based on automated deep learning", but it is not difficult to find the following technical problems through analysis:

[0004] The existing technical solutions cannot accurately determine the exact location of a person, and cannot identify the identity of the person, resulting in.

[0005] In response to the above technical problems, the present invention provides a method and system for personnel classification and recognition based on neural network. Summary of the invention

[0006] To achieve the purpose of the present invention, the present invention adopts the following technical solutions:

[0007] According to one aspect of the present invention, a method for classifying and identifying people based on a neural network is provided.

[0008] A method for classifying and identifying people based on a neural network, characterized by comprising:

[0009] Acquire detection signals of two or more perception radars for detecting personnel, perform data processing on the detection signals of the perception radars, construct a label data set of personnel and non-personnel signals, construct a neural network model, and perform supervised training on the label data set to realize the construction of a classification and recognition model for personnel and other targets;

[0010] Output the coordinates of the target person according to the classification result of the classification recognition model, and continuously send representative signals to the cloud platform at set time intervals;

[0011] According to the coordinates of the target person, it is determined whether there is an operator near the physical channel, and the target person is identified as an operator or an illegal intruder based on the position matching between the two within the error range.

[0012] A further technical solution is that the sensing radar is located in a horizontal plane and in a straight line, arranged at a preset distance, and the detection area is a sector with a central angle of 120°.

[0013] A further technical solution is to construct a dataset of human and non-human signal labels, including:

[0014] Performing distance-FFT processing on the detection signal of the perception radar, converting the detection signal into a distance-slow time graph that can reflect the distance of the person, and locating the high-reflection area of ​​the chest and breathing on the distance-slow time graph;

[0015] The high-reflection area is subjected to filtering and denoising processing to obtain a person's chest waveform and a breathing waveform, and feature labels are produced for the person's chest waveform and the breathing waveform to construct a person and non-person signal label data set.

[0016] A further technical solution is that the specific steps of constructing the distance-slow time graph are:

[0017] Extracting a detection signal from a sensing radar, applying a range-FFT to each sampling point of a frequency modulated continuous wave in the detection signal, and placing the sampling points after the range-FFT into a matrix by column to form a range-slow time matrix;

[0018] The matrix data of the distance-slow time matrix is ​​visualized to form a distance-slow time graph.

[0019] A further technical solution is that before locating the high-reflection areas of the chest and breathing, the DC wave of the distance-slow time diagram is first eliminated by using an average removal method, and then the DC wave is eliminated of clutter by using a moving average method.

[0020] A further technical solution is to locate the high-reflective areas of the chest and breathing, specifically including:

[0021] The target phase information of the distance-slow time diagram is extracted by using an inverse tangent function, the phase drift of the target phase information is eliminated by using a differential method, and the sudden change signal of the target phase information is solved by using a sliding median to obtain the processed target phase information;

[0022] The processed target phase information is subjected to denoising by using wavelet transform, and a high-reflection region waveform is obtained by performing inverse wavelet transform on the processed target phase information, thereby realizing the positioning of the high-reflection region of the chest cavity and the breathing.

[0023] A further technical solution is that the neural network model is constructed based on an improved feature fusion CNN-LSTM neural network model.

[0024] A further technical solution is that the improved feature fusion CNN-LSTM neural network model consists of 4 parts, the first part consists of a one-dimensional convolution module, a DWConv_BN_RELU convolution block, a maximum pooling layer and a Flatten layer, with a total of 12 stages; the second part consists of an LSTM layer and a Flatten layer, with a total of 2 stages; the third part consists of a one-dimensional convolution module, a DWConv_BN_RELU convolution block, a maximum pooling layer, a LSTM layer and a Flatten layer, with a total of 8 stages; the fourth part consists of a fully connected layer, with a total of 1 stage.

[0025] A further technical solution is to identify the target person as an operator or an illegal intruder based on the position matching between the two within the error range, specifically including:

[0026] When the position deviation between the two is within a preset range, the target person is determined to be an operator;

[0027] When the position deviation between the two is not within the preset range, it is determined that the target person is not an operating person.

[0028] A further technical solution is to determine whether there are any operators near the physical channel, including:

[0029] Based on the positioning information of the smart helmet worn by the target person, it is determined whether there are workers near the physical passage.

[0030] In a second aspect, the present invention provides a computer system comprising: a memory and a processor that are communicatively connected, and a computer program stored in the memory and capable of running on the processor, characterized in that: when the processor runs the computer program, it executes the above-mentioned neural network-based personnel classification and recognition method.

[0031] The beneficial effects of the present invention are:

[0032] The present invention combines the human presence sensing radar with the intelligent helmet positioning to achieve the sensing, positioning and identification of approaching suspicious persons, thus facilitating the personnel management in the construction area.

[0033] Other features and advantages will be described in the following description. The objects and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description and drawings.

[0034] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The above and other features and advantages of the present invention will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings.

[0036] Figure 1 It is a flowchart of a method for personnel classification and recognition based on a neural network;

[0037] Figure 2 It is a diagram of the radar arrangement method;

[0038] Figure 3 This is the structural diagram of the improved feature fusion CNN-LSTM model;

[0039] Figure 4 is a schematic diagram of the layout of an embodiment of the present application;

[0040] Figure 5 This is a schematic diagram of the electronic fence division of an embodiment of the present application;

[0041] Figure 6 It is a schematic diagram of radar perception positioning calculation in an embodiment of the present application. DETAILED DESCRIPTION

[0042] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0043] The following will further elaborate from two perspectives: method-based embodiments and system-based embodiments.

[0044] Method Example

[0045] To solve the above problems, according to one aspect of the present invention, Figure 1 As shown, a method for personnel classification and recognition based on a neural network is provided, which is characterized by specifically comprising:

[0046] Acquire detection signals of two or more perception radars for detecting personnel, perform data processing on the detection signals of the perception radars, construct a label data set of personnel and non-personnel signals, construct a neural network model, and perform supervised training on the label data set to realize the construction of a classification and recognition model for personnel and other targets;

[0047] The characteristic labels of the chest cavity and respiratory waveforms of the personnel are produced to construct a personnel and non-personnel signal label data set, and the characteristic labels are all [φ1, φ2, ..., φ N ,1],[φ1,φ2,…,φ N , 0]. 1 represents a human target, and 0 represents a non-human target. The label dataset is randomly divided into a training set and a validation set in a ratio of 7:3 for model training.

[0048] Output the coordinates of the target person according to the classification result of the classification recognition model, and continuously send representative signals to the cloud platform at set time intervals;

[0049] According to the coordinates of the target person, it is determined whether there is an operator near the physical channel, and the target person is identified as an operator or an illegal intruder based on the position matching between the two within the error range.

[0050] Specifically, the perception radar is located in a horizontal plane and in a straight line, arranged at a preset distance, and the detection area is a sector with a central angle of 120°.

[0051] Radar layout reference Figure 2 As shown, the present invention provides a personnel classification and identification method integrating human presence sensing radar and intelligent helmet positioning technology, wherein a physical passage 5 facing outward is arranged at a certain position of the physical fence 1, which can limit outsiders within the detection range of the human presence sensing radars 3 and 4. The human presence sensing radars 3 and 4 are installed on fixed poles at the vertices on both sides of the overlapping edge of the physical passage 5 and the physical fence 1, and the positive directions of the installations are biased toward one side of the physical passage 5, ensuring full coverage of the range of the physical passage 5.

[0052] Specifically, we build a dataset of human and non-human signal labels, including:

[0053] Performing distance-FFT processing on the detection signal of the perception radar, converting the detection signal into a distance-slow time graph that can reflect the distance of the person, and locating the high-reflection area of ​​the chest and breathing on the distance-slow time graph;

[0054] It should be noted that, according to extracting distance information from a chirp, distance-FFT is applied to each sampling point of a set of N frequency modulated continuous waves Z(1), Z(2), ..., Z(N), and they are put into a matrix by column to form a distance-slow time matrix. The distance units all reflect the target reflection intensity to form a distance-slow time map;

[0055] The high-reflection area is subjected to filtering and denoising processing to obtain a person's chest waveform and a breathing waveform, and feature labels are produced for the person's chest waveform and the breathing waveform to construct a person and non-person signal label data set.

[0056] It should be noted that the specific steps of constructing the distance-slow time graph are:

[0057] Extracting a detection signal from a sensing radar, applying a range-FFT to each sampling point of a frequency modulated continuous wave in the detection signal, and placing the sampling points after the range-FFT into a matrix by column to form a range-slow time matrix;

[0058] The matrix data of the distance-slow time matrix is ​​visualized to form a distance-slow time graph.

[0059] Furthermore, before locating the high-reflection area of ​​the chest and breathing, the average removal method is first used to eliminate the DC wave of the distance-slow time diagram, and then the moving average method is used to eliminate clutter from the DC wave.

[0060] The mean removal method and moving average method can be expressed as D ij =D ij -avg(i);D ij =D ij , D ij >avg_f i (2f c ), D ij =avg_f i (2f c ), D ij ≤avg_f i (2f c );

[0061] Where D ij represents the distance-slow time matrix of row i and column j, 1≤i≤N, avg(i) is the average value of row i; avg_f i (2f c ) means taking the mean of data with twice the sampling frequency.

[0062] Specifically, the high-reflective areas of the chest and breathing are located, including:

[0063] The target phase information of the distance-slow time diagram is extracted by using an inverse tangent function, the phase drift of the target phase information is eliminated by using a differential method, and the sudden change signal of the target phase information is solved by using a sliding median to obtain the processed target phase information;

[0064] Wherein, the inverse tangent function can be expressed as

[0065] Where I(n) is the imaginary part of the complex signal of the nth frame in the distance unit where the chest cavity is located, and R(n) is the real part.

[0066] The difference method can be expressed as φ d =φ(n+1)-φ(n);

[0067] In the formula, φ d represents the phase after differentiation;

[0068] The sliding median method can be expressed as

[0069] Where φ dm represents the phase after sliding median, φ(mid) represents the median phase;

[0070] Even if the above-mentioned difference method and sliding median method are used for filtering, there are still high-frequency clutter in the phase. The present invention uses discrete wavelet transform for decomposition, formula:

[0071] Wavelet transform has two variables: scale a and translation τ. Scale a controls the expansion and contraction of the wavelet function, and translation τ controls the translation of the wavelet function. The scale corresponds to the frequency (inversely proportional), and the translation τ corresponds to the time. Decomposing φ will obtain the wavelet transform coefficient W, and the soft threshold denoising is performed on it to obtain W λ The wavelet coefficients estimated by the soft threshold have good overall continuity, and are subjected to inverse wavelet transform to obtain the reconstructed signal φ.

[0072] The processed target phase information is subjected to denoising by using wavelet transform, and a high-reflection region waveform is obtained by performing inverse wavelet transform on the processed target phase information, thereby realizing the positioning of the high-reflection region of the chest cavity and the breathing.

[0073] It should be noted that the neural network model is constructed based on an improved feature fusion CNN-LSTM neural network model.

[0074] Specifically, Figure 3 As shown in the figure, the improved feature fusion CNN-LSTM neural network model consists of 4 parts. The first part consists of a one-dimensional convolution module, a DWConv_BN_RELU convolution block, a maximum pooling layer and a Flatten layer, with a total of 12 stages; the second part consists of an LSTM layer and a Flatten layer, with a total of 2 stages; the third part consists of a one-dimensional convolution module, a DWConv_BN_RELU convolution block, a maximum pooling layer, a LSTM layer and a Flatten layer, with a total of 8 stages; the fourth part consists of a fully connected layer, with a total of 1 stage.

[0075] A further technical solution is to identify the target person as an operator or an illegal intruder based on the position matching between the two within the error range, specifically including:

[0076] When the position deviation between the two is within a preset range, the target person is determined to be an operator;

[0077] When the position deviation between the two is not within the preset range, it is determined that the target person is not an operating person.

[0078] The present invention provides a method for classifying and identifying people by integrating human presence sensing radar and intelligent helmet positioning technology, wherein the positioning algorithms of the two radars are as follows:

[0079] (1) Set the coordinate position of perception radar 3 to (0, 0), and the distance between perception radar 3 and perception radar 4 to L, then the coordinate position of perception radar 4 is (L, 0), and the coordinates of the human body are (xp, yp); perception radar 3 detects the human body, and the distance between the radar and the human body is R1; perception radar 4 detects the human body, and the distance between the radar and the human body is R2.

[0080] (2) The target person coordinates (xp, yp) can be calculated as:

[0081]

[0082]

[0083] The radar distance extraction formula is as follows:

[0084]

[0085] Where R is the straight-line distance from the target point to the radar, c is the speed of light, and f IF is the frequency of the intermediate frequency signal, T c is the duration of the chirp signal, and B is the bandwidth.

[0086] If the classification result is non-personnel, a signal representing "no one" will be continuously sent to the cloud platform at the set time interval, otherwise a signal representing "someone" will be sent.

[0087] Specifically, determine whether there are any operators near the physical channel, including:

[0088] Based on the positioning information of the smart helmet worn by the target person, it is determined whether there are workers near the physical passage.

[0089] Reference Figure 4As shown, the present invention provides a method for classifying and identifying personnel that integrates human presence sensing radar and intelligent helmet positioning technology, including a physical fence 1, an electronic fence 2, a human presence sensing radar 3, a human presence sensing radar 4, and a physical channel 5. The human presence sensing radars 3 and 4 are millimeter-wave radars equipped with relevant intelligent algorithms, and their detection areas are sectors with a central angle of 120°; specifically, when there is no human body in the detection area of ​​the sensing radar (edge ​​computing is performed by the radar's host computer), it will continuously send a signal representing "no one" to the cloud platform at a set time interval; when the sensing radar detects a human body, it will continuously send a signal representing "someone" and the distance of the human body relative to the radar to the cloud platform at a set time interval. The intelligent helmet 6 is equipped with a positioning module and a 4G communication module, and can send the positioning data of the helmet to the cloud platform according to relevant instructions. During construction, the staff deploys the physical fence 1 on site, the management personnel set up the electronic fence 2 on the remote cloud platform, and group the construction workers, and bind the corresponding intelligent helmets 6 on the cloud platform. The classification and identification method based on electronic fence technology and radar sensing is:

[0090] (1) When a person 6 enters the physical channel 5, the person presence sensing radars 3 and 4 sense the person, and the radar host computer analyzes and calculates the corresponding distance. Based on the predetermined positions of the two radars and the distance obtained by the test, the position of the person is determined, and an early warning message is sent to the cloud platform.

[0091] (2) After receiving the warning information, the cloud platform obtains the positioning information of all the construction workers in the group. When a worker (wearing a smart safety helmet 6) matches the two radar positioning positions within the error range, the human body 6 entering the physical channel 5 is identified as a construction worker; when no construction worker matches the two radar positioning positions, the human body 6 entering the physical channel 5 is identified as an illegal intruder.

[0092] Reference Figure 6 As shown, the present invention provides a personnel classification and identification method integrating human presence sensing radar and smart helmet positioning technology, wherein the construction area is surrounded by a physical fence 1, and the electronic fence 2 is divided according to the specific shape and position of the actual construction area, including an electronic fence 2-1 of the construction area divided for positioning, trajectory tracking and early warning management of construction personnel wearing smart helmets and an electronic fence 2-2 of the radar detection range. Assuming that the positioning error of the smart helmet is, the electronic fence 2 area is set to be larger than the physical fence.

[0093] The present invention provides a personnel classification and identification method that integrates human presence sensing radar and intelligent helmet positioning technology. Considering that the sensing radar has an algorithm error R and the intelligent helmet has a positioning error, the matching deviation between the two radar positioning positions and the intelligent helmet positioning is , that is, the construction personnel are identified within this error range.

[0094] System Example

[0095] On the other hand, the present invention provides a computer system, comprising: a memory and a processor connected in communication, and a computer program stored in the memory and capable of running on the processor, characterized in that: when the processor runs the computer program, it executes the above-mentioned neural network-based personnel classification and recognition method.

[0096] Through the above embodiments, the present invention achieves the following beneficial effects:

[0097] The present invention combines the human presence sensing radar with the intelligent helmet positioning to achieve the sensing, positioning and identification of approaching suspicious persons, thus facilitating the personnel management in the construction area.

[0098] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0099] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0100] The above description is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, one or more embodiments of this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included in the scope of the claims of this specification.

Claims

1. A method for personnel classification and recognition based on neural network, characterized in that: Specifically include: Acquire detection signals of two or more perception radars for detecting personnel, perform data processing on the detection signals of the perception radars, construct a label data set of personnel and non-personnel signals, construct a neural network model, and perform supervised training on the label data set to realize the construction of a classification and recognition model for personnel and other targets; Output the coordinates of the target person according to the classification result of the classification recognition model, and continuously send representative signals to the cloud platform at set time intervals; According to the coordinates of the target person, determine whether there is an operator near the physical channel, and identify the target person as an operator or an illegal intruder based on the position matching between the two within the error range; Constructing a dataset of personnel and non-personnel signal labels, specifically including: performing distance-FFT processing on the detection signal of the perception radar, converting the detection signal into a distance-slow time diagram that can reflect the distance of the personnel, and locating the high-reflection area of ​​the chest and breathing on the distance-slow time diagram, specifically including: using the inverse tangent function to extract the target phase information of the distance-slow time diagram, and using the difference method to eliminate the phase drift of the target phase information, using the sliding median to solve the sudden change signal of the target phase information to obtain the processed target phase information; using wavelet transform to denoise the processed target phase information, and performing inverse wavelet transform on it to obtain the high-reflection area waveform, so as to locate the high-reflection area of ​​the chest and breathing; performing filtering and denoising on the high-reflection area to obtain the chest waveform and breathing waveform of the personnel, and performing feature labeling on the chest waveform and breathing waveform of the personnel to construct a dataset of personnel and non-personnel signal labels; the feature label is a phase sequence of [φ1, φ2, ..., φ N, 1], [φ1, φ2, ..., φN, 0], 1 represents a personnel target, and 0 represents a non-personnel target; Based on the position matching between the two within the error range, the target person is identified as an operating person or an illegal intruder, specifically including: When the position deviation between the two is within a preset range, the target person is determined to be an operator; When the position deviation between the two is not within the preset range, it is determined that the target person is not an operating person; When the operating personnel matches the two radar positioning positions within the error range, the human body entering the physical channel is identified as a construction worker; when there is no construction worker and the two radar positioning positions match, the human body entering the physical channel is identified as an illegal intruder.

2. The method for classifying and identifying people based on a neural network as claimed in claim 1, characterized in that: The sensing radar is located in a horizontal plane and in a straight line, arranged at a preset distance, and the detection area is a sector with a central angle of 120°.

3. The method for classifying and identifying people based on a neural network as claimed in claim 1, characterized in that: Construct a dataset of human and non-human signal labels, including: Performing distance-FFT processing on the detection signal of the perception radar, converting the detection signal into a distance-slow time graph that can reflect the distance of the person, and locating the high-reflection area of ​​the chest and breathing on the distance-slow time graph; The high-reflection area is subjected to filtering and denoising processing to obtain a person's chest waveform and a breathing waveform, and feature labels are produced for the person's chest waveform and the breathing waveform to construct a person and non-person signal label data set.

4. The method for classifying and identifying people based on a neural network as claimed in claim 3, characterized in that: The specific steps of constructing the distance-slow time graph are: Extracting a detection signal from a sensing radar, applying a range-FFT to each sampling point of a frequency modulated continuous wave in the detection signal, and placing the sampling points after the range-FFT into a matrix by column to form a range-slow time matrix; The matrix data of the distance-slow time matrix is ​​visualized to form a distance-slow time graph.

5. The method for classifying and identifying people based on a neural network as claimed in claim 3, characterized in that: Before locating the high-reflection area of ​​the chest and breathing, the average removal method is first used to eliminate the DC wave of the distance-slow time diagram, and then the moving average method is used to eliminate clutter from the DC wave.

6. The method for classifying and identifying people based on a neural network as claimed in claim 1, characterized in that: Locate the highly reflective areas of the chest and respiratory cavity, including: The target phase information of the distance-slow time diagram is extracted by using an inverse tangent function, the phase drift of the target phase information is eliminated by using a differential method, and the sudden change signal of the target phase information is solved by using a sliding median to obtain the processed target phase information; The processed target phase information is subjected to denoising by using wavelet transform, and a high-reflection region waveform is obtained by performing inverse wavelet transform on the processed target phase information, thereby realizing the positioning of the high-reflection region of the chest cavity and the breathing.

7. The method for classifying and identifying people based on a neural network as claimed in claim 1, characterized in that: The neural network model is constructed based on an improved feature fusion CNN-LSTM neural network model.

8. The method for classifying and identifying people based on a neural network as claimed in claim 1, characterized in that: Determine whether there are any workers near the physical passage, including: Based on the positioning information of the smart helmet worn by the target person, it is determined whether there are workers near the physical passage.

9. A computer system comprising: A memory and a processor connected in communication, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes a neural network-based personnel classification and recognition method as described in any one of claims 1-8 when running the computer program.

Citation Information

Patent Citations

  • Radar waveform recognition method and system based on automatic deep learning

    CN112487933A

  • Human body target identity recognition method and system based on vital sign SAR (Synthetic Aperture Radar) imaging

    CN110596705A

  • Power plant safety behavior information automatic detection method based on deep learning

    CN112364778A

  • Safety helmet detection and identity recognition method and device for electric power operation site

    CN114067268A

  • Method for detecting human body in non-line-of-sight area by millimeter wave frequency modulation continuous wave radar

    CN115372959A