A package anomaly detection system and method based on millimeter wave imaging

Abstract

The present invention discloses a package anomaly detection system and method based on millimeter wave imaging, including: a millimeter wave signal acquisition module and a data processing platform; the millimeter wave signal acquisition module includes a millimeter wave sensor and a data acquisition board; the millimeter wave sensor transmits a wireless radio frequency signal, receives the wireless radio frequency signal reflected by the package and mixes the transmitted signal and the received signal; the data acquisition board obtains the mixed signal data and transmits it to the data processing platform in real time; the data processing platform includes: a signal imaging module, which is used to process the mixed signal to obtain a perspective imaging map of the package; an anomaly detection module, which is used to extract features in the perspective imaging map of the package to form a feature map, and train a package anomaly classifier based on the feature map, thereby realizing package anomaly detection. The present invention uses millimeter wave signals to realize perspective imaging of packages on a conveyor belt, and realizes contactless and non-invasive detection of abnormal postures of items in the package based on intuitive imaging.

Description

Technical Field

[0001] The present invention belongs to the field of millimeter wave wireless sensing and industrial Internet technology, and specifically relates to a package anomaly detection system and method based on millimeter wave imaging. Background Art

[0002] With the development of e-commerce and logistics networks, thousands of goods are bought and sold online, and the corresponding packages are delivered to customers by express delivery. Among these packages, there are special categories of items that should be properly packaged and shipped to prevent damage. For example, bottles and cans containing liquid preparations or chemical reagents should be placed correctly in the package. If the bottle is tilted or inverted, excessive pressure at the bottle mouth may cause it to rupture, causing internal liquid leakage, resulting in economic losses or even life damage. Therefore, it is of great significance to effectively detect abnormal packages, correct the position of items in the package in time, and avoid losses in the logistics system. In order to solve this problem, a package anomaly detection system is needed to efficiently obtain the status of the goods inside the package without opening the package, so as to report package anomalies immediately.

[0003] In order to detect package anomalies, the current main solutions include the following:

[0004] 1) Computer vision-based detection method: It is widely used to detect the position and shape of objects, but it can only detect objects within the line of sight and cannot obtain the internal state of the package.

[0005] 2) Detection method based on X-ray imaging: widely used in security inspections, it can penetrate the package packaging to obtain the status of the internal items. However, for each express delivery station in the logistics scenario, the detection equipment is expensive and cannot be popularized.

[0006] 3) Detection method based on wireless signals: Use wireless signals such as RFID (Radio Frequency Identification) tag backscatter signals, acoustic signals, etc. to sense the internal status of the package by penetration and diffraction. Due to the inherent characteristics of the signal, it is easily disturbed by the surrounding environment.

[0007] 4) Detection method based on millimeter wave imaging: Customized equipment generates package perspective images similar to X-rays, but the hardware cost is high and the equipment is bulky.

[0008] Based on this, the present invention designs a package anomaly detection system, which uses relatively low-cost commercial millimeter wave equipment to sense the status of items inside the package by processing reflected millimeter wave signals. Summary of the invention

[0009] In view of the above-mentioned deficiencies in the prior art, the purpose of the present invention is to provide a parcel anomaly detection system and method based on millimeter wave imaging, so as to solve the susceptibility to interference of non-radio frequency signal detection in the prior art, and the high equipment and deployment costs caused by relying on customized hardware detection. The present invention uses millimeter wave signals to realize perspective imaging of parcels on the conveyor belt, and realizes non-contact and non-invasive detection of abnormal posture of items in the parcel based on intuitive imaging.

[0010] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0011] A package anomaly detection system based on millimeter wave imaging of the present invention comprises: a millimeter wave signal acquisition module and a data processing platform;

[0012] The millimeter wave signal acquisition module includes a millimeter wave sensor and a data acquisition board; the millimeter wave sensor transmits a wireless radio frequency signal, receives the wireless radio frequency signal reflected by the package, and mixes the transmitted signal and the received signal; the data acquisition board obtains the mixed signal data and transmits it to the data processing platform in real time;

[0013] The data processing platform includes: signal imaging module and anomaly detection module;

[0014] A signal imaging module, used for processing the mixed signal to obtain a perspective imaging image of the package;

[0015] The anomaly detection module is used to extract features from the perspective imaging image of the package to form a feature map, and train a package anomaly classifier based on the feature map to achieve package anomaly detection.

[0016] Furthermore, the millimeter wave sensor transmits a wireless radio frequency signal and collects the wireless radio frequency signal reflected by a package placed on a conveyor belt and moving at a constant speed; the millimeter wave sensor is placed on one side of the conveyor belt, and the center of the wireless radio frequency signal beam it transmits is perpendicular to the moving direction of the package.

[0017] Furthermore, the millimeter wave sensor transmits an FMCW radio frequency signal chirp with a frequency of 76-81 GHz, and a horizontal field of view of a 6 db beam of the wireless radio frequency signal is ±50°; the data acquisition board obtains the mixed intermediate frequency signal and transmits it to the data processing platform via Ethernet.

[0018] Furthermore, the signal imaging module obtains an optimized range migration correction algorithm based on the constructed target reflection model; and designs an imaging algorithm based on the optimized range migration correction algorithm to convert the collected signal data S IF (t,t d ) is processed into a wrapped perspective image S img (R ran ,R azi), where t is the timestamp within a chirp signal, t d is the timestamp of different chirps, R ran is the distance axis coordinate of the reflection point in the imaging scene, R azi is the azimuth axis coordinate of the reflection point in the imaging scene; the range direction is the direction of target package movement, and the azimuth direction is the direction of the signal beam center, which is perpendicular to the range direction.

[0019] Furthermore, the target reflection model construction method is as follows:

[0020] A certain reflection point in the imaging area moves on the transmission, and the distance between it and the millimeter wave sensor is R, that is, the moving trajectory is expressed as follows:

[0021]

[0022] Where v is the conveyor belt moving speed, T sa It is the synthetic aperture time, and the target moving trajectory is in an arc shape.

[0023] Furthermore, the imaging algorithm ISAR in the signal imaging module is defined as follows:

[0024] S img (R ran ,R azi )=ISAR(S IF (t,t d ))

[0025] The intermediate frequency signal S IF Processed into a wrapped perspective image S img , showing the status of the items inside the package, the imaging algorithm includes three steps:

[0026] (1) Perform fast Fourier transform on the intermediate frequency signal to obtain the target movement trajectory;

[0027] (2) Optimize the range motion correction algorithm to obtain a segmented range motion correction algorithm for small area imaging to correct the target movement trajectory;

[0028] (3) Compressing the target movement trajectory obtains the relative position of the target in the moving area to be imaged, forming a perspective imaging map of the package.

[0029] Furthermore, the step (1) specifically includes:

[0030] In the chirp signal frequency domain, we get S IF There is a frequency domain expression S ISAR as follows:

[0031] S ISAR (f,t d )=FFT1D [S IF (t,t d )]

[0032] Among them, f is the frequency corresponding to the slow time dimension of the intermediate frequency signal, FFT 1D It is a one-dimensional FFT calculation, that is, S IF The slow time dimension is used to perform FFT; the FFT 1D The number of points is twice the sampling points of an intermediate frequency signal; the intermediate frequency signal frequency f is proportional to the distance R from the reflection point target to the radar K is the modulation frequency of the original transmitted signal, c is the speed of light, and the signal matrix S ISAR Rewrite as S′ ISAR :

[0033]

[0034] Thus, the target moving trajectory R(t d ).

[0035] Furthermore, the step (2) specifically includes:

[0036] Correct the target moving trajectory R(t d ), thus obtaining the data matrix S″ ISAR as follows:

[0037] S″ ISAR (R ran ,t d ) = PRCMC(S′ ISAR (R,t d ))

[0038] Among them, PRCMC is a segmented range migration correction algorithm, that is, an optimized range migration correction algorithm; the optimized range migration correction algorithm PRCMC includes two parts of optimization:

[0039] (21) Propose segmented correction: The reason for the migration is that the change range of the target trajectory ΔR is greater than the signal distance resolution R res : So we can calculate the target distance R ran Less than No correction is required, θ bm is half of the horizontal viewing angle of the signal beam; if the target distance R ran Greater than or equal to Correction is needed;

[0040] (22) Correction value calculation: distance correction value R mig =R(t d )-R ran .

[0041] Furthermore, the step (3) specifically includes:

[0042] According to the Doppler effect, the zero-frequency point reflected by different point targets corresponds to its coordinate R_azi on the azimuth axis. Thus, the coordinates of all point targets in the enclosed imaging area (R ran ,R azi ), forming a perspective image S of the package img ; Calculate the coordinates of the point target azimuth axis R_azi to obtain the perspective imaging image S img The process is as follows:

[0043] S img (R ran ,R azi )=PC(S″ ISAR (R ran ,t d ))

[0044] Among them, PC is the pulse compression function, which compresses the FM signal with time length to the zero frequency point to obtain the azimuth axis coordinate R azi .

[0045] Furthermore, the anomaly detection module uses an image feature operator to extract features from the perspective imaging image of the package to form a feature map, trains a lightweight classifier SVM to classify the feature map, and implements package anomaly detection, as follows:

[0046] The perspective image of the package is preliminarily processed and blurred using a Gaussian low-pass filter to suppress the meaningless image ripples in the image caused by FFT calculation and highlight the contour features in the image; the Gaussian cutoff frequency is set to 33 Hz;

[0047] The image feature HOG operator is used to extract the image contour features in the wrapped perspective imaging image: the HOG operator constructs the contour feature map by calculating the gradient direction histogram of the local area of ​​the image. In the calculation process of the feature map, the relevant parameter settings include: the block normalization mode is set to L1, a block is set to 4×4 units, and a unit is set to 8×8 image pixels;

[0048] Package anomaly detection is achieved by training a classifier SVM based on feature maps: the feature map is enhanced by rotating it at four angles (-30°, -15°, 15°, and 30°), that is, the classifier SVM training samples are enhanced to resist the detection interference caused by the rotation of the package on the conveyor belt; the enhanced feature map samples are flattened into feature vectors as the training input of the classifier SVM, and the output is the posture of the items in the package corresponding to the feature map. The posture is divided into three categories: normal standing, abnormal tilt, and abnormal dumping; the kernel function of the classifier SVM is set to a polynomial kernel. The trained classifier SVM divides the postures of the items in the package into three categories based on the perspective imaging of the package to achieve package anomaly detection.

[0049] A method for detecting package anomalies based on millimeter wave imaging of the present invention, based on the above system, comprises the following steps:

[0050] The millimeter wave sensor sends a wireless radio frequency signal and receives the reflected signal generated by the package surface and the items inside;

[0051] Processing the reflected signal to obtain an intermediate frequency signal, the data acquisition board collects the intermediate frequency signal data obtained by the millimeter wave sensor in real time, and sends it to the data processing platform;

[0052] The imaging algorithm is designed based on the optimized range motion correction algorithm, and the intermediate frequency signal is processed into a wrapped perspective imaging image through the imaging algorithm;

[0053] The contour features of the package imaging are extracted from the perspective imaging image of the package to form a feature map, which is classified by a classifier to obtain the posture classification of the items in the package, thereby realizing package anomaly detection.

[0054] Beneficial effects of the present invention:

[0055] 1. The present invention can realize high-precision package anomaly detection: the state and posture of special items in the package are detected, and the detection accuracy reaches above 90.48.

[0056] 2. The present invention improves the safety and reliability of the logistics system: by utilizing wireless sensing technology and intelligent algorithm design, it can accurately detect items in a package, such as abnormal postures of bottles and jars containing special medicines.

[0057] 3. The present invention reduces economic losses in logistics: by using low-cost commercial equipment to non-invasively detect the status of the package, the abnormal position of the items in the package can be corrected immediately before an accident occurs, thereby reducing logistics costs and increasing economic benefits.

[0058] 4. The present invention has low environmental requirements and is easy to deploy: it has low environmental requirements, is not easily disturbed, has high robustness, and can work normally in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a diagram of the package anomaly detection system architecture of the present invention;

[0060] Figure 2 It is a schematic diagram of the target reflection model of the present invention;

[0061] Figure 3 is a flow chart of the imaging algorithm of the present invention;

[0062] Figure 4 This is a schematic diagram of the signal imaging principle of the present invention;

[0063] Figure 5 Schematic diagram of package anomaly detection according to the present invention. DETAILED DESCRIPTION

[0064] In order to facilitate the understanding of those skilled in the art, the present invention is further described below in conjunction with embodiments and drawings. The contents mentioned in the implementation modes are not intended to limit the present invention.

[0065] Reference Figure 1 As shown, a package anomaly detection system based on millimeter wave imaging of the present invention includes: a millimeter wave signal acquisition module and a data processing platform; wherein,

[0066] The millimeter wave signal acquisition module includes a millimeter wave sensor and a data acquisition board; the millimeter wave sensor transmits a wireless radio frequency signal, receives the wireless radio frequency signal reflected by the package, and mixes the transmitted signal and the received signal; the data acquisition board obtains the mixed signal data and transmits it to the data processing platform in real time;

[0067] The millimeter wave sensor transmits a wireless radio frequency signal and collects the wireless radio frequency signal reflected by a package placed on a conveyor belt and moving at a constant speed; the millimeter wave sensor is placed on one side of the conveyor belt, and the center of the wireless radio frequency signal beam it transmits is perpendicular to the moving direction of the package.

[0068] The millimeter wave sensor transmits an FMCW radio frequency signal chirp with a frequency of 76-81 GHz, and the horizontal field of view of the wireless radio frequency signal 6db beam is ±50°; the data acquisition board obtains the mixed intermediate frequency signal and transmits it to the data processing platform via Ethernet.

[0069] The data processing platform includes: signal imaging module and anomaly detection module;

[0070] A signal imaging module, used for processing the mixed signal to obtain a perspective imaging image of the package;

[0071] The anomaly detection module is used to extract features from the perspective imaging image of the package to form a feature map, and train a package anomaly classifier based on the feature map to achieve package anomaly detection.

[0072] Specifically, the signal imaging module obtains an optimized range migration correction algorithm based on the constructed target reflection model; and designs an imaging algorithm based on the optimized range migration correction algorithm to convert the collected signal data S IF (t,t d ) is processed into a wrapped perspective image S img (R ran ,R azi ), where t is the timestamp within a chirp signal, t d is the timestamp of different chirps, R ran is the distance axis coordinate of the reflection point in the imaging scene, R azi is the azimuth axis coordinate of the reflection point in the imaging scene; the range direction is the direction of target package movement, and the azimuth direction is the direction of the signal beam center, which is perpendicular to the range direction.

[0073] Reference Figure 2 As shown in the figure, the target reflection model is constructed as follows:

[0074] A certain reflection point in the imaging area moves on the transmission, and the distance between it and the millimeter wave sensor is R, that is, the moving trajectory is expressed as follows:

[0075]

[0076] Where v is the conveyor belt moving speed, T sa It is the synthetic aperture time, and the target moving trajectory is in an arc shape.

[0077] Reference Figure 3 As shown, the imaging algorithm ISAR is defined as follows:

[0078] S img (R ran ,R azi )=ISAR(S IF (t,t d ))

[0079] The intermediate frequency signal S IF Processed into a wrapped perspective image S img , showing the status of the items inside the package, the imaging algorithm includes three steps:

[0080] (1) Performing Fast Fourier Transform (FFT) on the intermediate frequency signal to obtain the target movement trajectory;

[0081] (2) Optimize the range cell migration correction (RCMC) algorithm to obtain the piecewise range cell migration correction (PRCMC) algorithm for small area imaging to correct the target movement trajectory;

[0082] (3) Compressing the target movement trajectory obtains the relative position of the target in the moving area to be imaged, forming a perspective imaging map of the package.

[0083] In the example, step (1) specifically includes:

[0084] In the chirp signal frequency domain, we get S IF There is a frequency domain expression S ISAR as follows:

[0085] S ISAR (f,t d )=FFT 1D [S IF (t,t d )]

[0086] Among them, f is the frequency corresponding to the slow time dimension of the intermediate frequency signal, FFT 1D It is a one-dimensional FFT calculation, that is, S IF The slow time dimension is used to perform FFT; the FFT 1D The number of points is twice the sampling points of an intermediate frequency signal; the intermediate frequency signal frequency f is proportional to the distance R from the reflection point target to the radar K is the modulation frequency of the original transmitted signal, c is the speed of light, and the signal matrix S ISAR Rewrite as S′ ISAR :

[0087]

[0088] Thus, the target moving trajectory R(t d ).

[0089] In the example, step (2) specifically includes:

[0090] Correct the target moving trajectory R(t d ), thus obtaining the data matrix S″ ISAR as follows:

[0091] S″ ISAR (R ran ,t d ) = PRCMC(S′ ISAR (R,t d ))

[0092] Among them, PRCMC is a segmented range migration correction algorithm, that is, an optimized range migration correction algorithm; the optimized range migration correction algorithm PRCMC includes two parts of optimization:

[0093] (21) Propose segmented correction: The reason for the migration is that the change range of the target trajectory ΔR is greater than the signal distance resolution R res : So we can calculate the target distance R ran Less than No correction is required, θ bm is half of the horizontal viewing angle of the signal beam; if the target distance R ran Greater than or equal to Correction is needed;

[0094] (22) Correction value calculation: distance correction value R mig =R(t d )-R ran .

[0095] In the example, step (3) specifically includes:

[0096] According to the Doppler effect, Figure 4 As shown in the figure, when the package moves at a constant speed on the conveyor belt, the phase of the wireless radio frequency signal collected by the millimeter wave sensor will change with the change of the position of the package, that is, the zero frequency point reflected by different point targets corresponds to its coordinate R_azi on the azimuth axis, so that the coordinates of all point targets in the imaging area of ​​the package are obtained (R ran ,R azi ), forming a perspective image S of the package img ; Calculate the coordinates of the point target azimuth axis R_azi to obtain the perspective imaging image S img The process is as follows:

[0097] S img (R ran ,R azi )=PC(S″ ISAR (R ran ,t d ))

[0098] Among them, PC is the pulse compression function, which compresses the FM signal with time length to the zero frequency point to obtain the azimuth axis coordinate R azi .

[0099] And, the anomaly detection module uses the image feature operator to extract the features in the perspective imaging image of the package to form a feature map, trains a lightweight classifier SVM to classify the feature map, and realizes package anomaly detection, referring to Figure 5 As shown, the details are as follows:

[0100] The perspective image of the package is preliminarily processed and blurred using a Gaussian low-pass filter to suppress the meaningless image ripples in the image caused by FFT calculation and highlight the contour features in the image; the Gaussian cutoff frequency is set to 33 Hz;

[0101] The image feature HOG operator is used to extract the image contour features in the wrapped perspective imaging image: the HOG operator constructs a contour feature map by calculating the gradient direction histogram of the local area of ​​the image. In the calculation process of the feature map, the relevant parameter settings include: the block normalization method is set to L1, a block is set to 4×4 cells, and a cell is set to 8×8 image pixels.

[0102] Package anomaly detection is achieved by training a classifier SVM based on feature maps: the feature map is enhanced by rotating it at four angles (-30°, -15°, 15°, and 30°), that is, the classifier SVM training samples are enhanced to resist the detection interference caused by the rotation of the package on the conveyor belt; the enhanced feature map samples are flattened into feature vectors as the training input of the classifier SVM, and the output is the posture of the items in the package corresponding to the feature map. The posture is divided into three categories: normal standing, abnormal tilt, and abnormal dumping; the kernel function of the classifier SVM is set to a polynomial kernel. The trained classifier SVM divides the postures of the items in the package into three categories based on the perspective imaging of the package to achieve package anomaly detection.

[0103] A method for detecting package anomalies based on millimeter wave imaging of the present invention, based on the above system, comprises the following steps:

[0104] The millimeter wave sensor sends a wireless radio frequency signal and receives the reflected signal generated by the package surface and the items inside;

[0105] Processing the reflected signal to obtain an intermediate frequency signal, the data acquisition board collects the intermediate frequency signal data obtained by the millimeter wave sensor in real time, and sends it to the data processing platform;

[0106] The imaging algorithm is designed based on the optimized range motion correction algorithm, and the intermediate frequency signal is processed into a wrapped perspective imaging image through the imaging algorithm;

[0107] The contour features of the package imaging are extracted from the perspective imaging image of the package to form a feature map, which is classified by a classifier to obtain the posture classification of the items in the package, thereby realizing package anomaly detection.

[0108] The present invention has many specific application paths. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements can be made without departing from the principle of the present invention. These improvements should also be regarded as the protection scope of the present invention.

Claims

1. A package anomaly detection system based on millimeter wave imaging, characterized in that: include: Millimeter wave signal acquisition module and data processing platform; Millimeter wave signal acquisition module, including millimeter wave sensor and data acquisition board; The millimeter wave sensor transmits a wireless radio frequency signal, receives the wireless radio frequency signal reflected by the package, and mixes the transmitted signal and the received signal; the data acquisition board obtains the mixed signal data and transmits it to the data processing platform in real time; The data processing platform includes: signal imaging module and anomaly detection module; A signal imaging module, used for processing the mixed signal to obtain a perspective imaging image of the package; An anomaly detection module is used to extract features from the perspective imaging image of the package to form a feature map, and train a package anomaly classifier based on the feature map to achieve package anomaly detection; The millimeter wave sensor transmits a wireless radio frequency signal and collects the wireless radio frequency signal reflected by a package placed on a conveyor belt and moving at a constant speed; the millimeter wave sensor is placed on one side of the conveyor belt, and the center of the wireless radio frequency signal beam emitted by the millimeter wave sensor is perpendicular to the moving direction of the package; The signal imaging module obtains an optimized range motion correction algorithm based on the constructed target reflection model; and designs an imaging algorithm based on the optimized range motion correction algorithm to collect the signal data S IF (t,t d ) is processed into a wrapped perspective image S img (R ran ,R azi ), t is the timestamp in a chirp signal, t d is the timestamp of different chirps, R ran is the distance axis coordinate of the reflection point in the imaging scene, R azi is the azimuth axis coordinate of the reflection point in the imaging scene; the range direction is the direction of target package movement, and the azimuth direction is the direction of the signal beam center, which is perpendicular to the range direction; The target reflection model construction method is as follows: A certain reflection point in the imaging area moves on the transmission, and the distance between it and the millimeter wave sensor is R, that is, the moving trajectory is expressed as follows: Where v is the conveyor belt moving speed, T sa It is the synthetic aperture time, and the target moving trajectory is in an arc shape; The imaging algorithm ISAR in the signal imaging module is defined as follows: S img (R ran ,R azi )=ISAR(S IF (t,t d )) The intermediate frequency signal S IF Processed into a wrapped perspective image S img , showing the status of the items inside the package, the imaging algorithm includes the following steps: (1) Perform fast Fourier transform on the intermediate frequency signal to obtain the target movement trajectory; (2) Optimize the range motion correction algorithm to obtain a segmented range motion correction algorithm for small area imaging to correct the target movement trajectory; (3) compressing the target movement trajectory to obtain the relative position of the target in the moving area to be imaged, and forming a perspective imaging map of the package; The anomaly detection module uses an image feature operator to extract features from the perspective imaging image of the package to form a feature map, trains a lightweight classifier SVM to classify the feature map, and implements package anomaly detection, as follows: Preliminarily process the perspective imaging image of the package, and use a Gaussian low-pass filter to blur the perspective imaging image of the package to suppress the meaningless image ripples in the imaging image caused by FFT calculation and highlight the contour features in the imaging image; The image feature HOG operator is used to extract the image contour features in the wrapped perspective imaging image: the HOG operator constructs the contour feature map by calculating the gradient direction histogram of the local area of ​​the image. In the calculation process of the feature map, the relevant parameter settings include: the block normalization mode is set to L1, a block is set to 4×4 units, and a unit is set to 8×8 image pixels; Package anomaly detection is achieved by training a classifier SVM based on feature maps: the feature map is enhanced by rotating it at four angles (-30°, -15°, 15°, and 30°), that is, the classifier SVM training samples are enhanced to resist the detection interference caused by the rotation of the package on the conveyor belt; the enhanced feature map samples are flattened into feature vectors as the training input of the classifier SVM, and the output is the posture of the items in the package corresponding to the feature map. The posture is divided into three categories: normal standing, abnormal tilt, and abnormal dumping; the kernel function of the classifier SVM is set to a polynomial kernel. The trained classifier SVM divides the postures of the items in the package into three categories based on the perspective imaging of the package to achieve package anomaly detection.

2. The package anomaly detection system based on millimeter wave imaging according to claim 1 is characterized in that: The step (1) specifically comprises: In the chirp signal frequency domain, we get S IF There is a frequency domain expression S ISAR as follows: S ISAR (f,t d )=FFT 1D [S IF (t,t d )], Among them, f is the frequency corresponding to the slow time dimension of the intermediate frequency signal, FFT 1D It is a one-dimensional FFT calculation, that is, S IF The slow time dimension is used to perform FFT; the FFT 1D The number of points is twice the sampling points of an intermediate frequency signal; the intermediate frequency signal frequency f is proportional to the distance R from the reflection point target to the radar K is the modulation frequency of the original transmitted signal, c is the speed of light, and the signal matrix S ISAR Rewrite as S′ ISAR : Thus, the target moving trajectory R(t d ).

3. The package anomaly detection system based on millimeter wave imaging according to claim 2 is characterized in that: The step (2) specifically comprises: Correct the target moving trajectory R(t d ), thus obtaining the data matrix S″ ISAR as follows: S″ ISAR (R ran ,t d )=PRCMC(S′ ISAR (R,t d )) Among them, PRCMC is a segmented range migration correction algorithm, that is, an optimized range migration correction algorithm; the optimized range migration correction algorithm PRCMC includes two parts of optimization: (21) Propose segmented correction: The reason for the migration is that the change range of the target trajectory ΔR is greater than the signal distance resolution R res : So we can calculate the target distance R ran Less than No correction is required, θ bm is half of the horizontal viewing angle of the signal beam; if the target distance R ran Greater than or equal to Correction is needed; (22) Correction value calculation: distance correction value R img =R(t d )-R ran .

4. The package anomaly detection system based on millimeter wave imaging according to claim 3 is characterized in that: The step (3) specifically comprises: According to the Doppler effect, the zero-frequency point reflected by different point targets corresponds to its coordinate R_azi on the azimuth axis. Thus, the coordinates of all point targets in the enclosed imaging area (R ran ,R azi ), forming a perspective image S of the package img ; Calculate the coordinates of the point target azimuth axis R_azi to obtain the perspective imaging image S img The process is as follows: S img (R ran ,R azi )=PC(S″ ISAR (R ran ,t d )) Among them, PC is the pulse compression function, which compresses the FM signal with time length to the zero frequency point to obtain the azimuth axis coordinate R azi .

5. A method for detecting package anomalies based on millimeter wave imaging, based on the system described in any one of claims 1 to 4, characterized in that: The steps include: The millimeter wave sensor sends a wireless radio frequency signal and receives the reflected signal generated by the package surface and the items inside; Processing the reflected signal to obtain an intermediate frequency signal, the data acquisition board collects the intermediate frequency signal data obtained by the millimeter wave sensor in real time, and sends it to the data processing platform; The imaging algorithm is designed based on the optimized range motion correction algorithm, and the intermediate frequency signal is processed into a wrapped perspective imaging image through the imaging algorithm; The contour features of the package imaging are extracted from the perspective imaging image of the package to form a feature map, which is classified by a classifier to obtain the posture classification of the items in the package, thereby realizing package anomaly detection.

Images