A millimeter-wave metal dangerous goods detection method based on the polarization wavelet domain

Abstract

The present invention discloses a millimeter-wave metal dangerous goods detection method based on the polarization wavelet domain, which specifically relates to the technical field of radar signal processing, and includes the following steps: Step 10: Obtain the echo signal; Step 20: Construct a polarization information-time multi-dimensional data volume; Step 30: Perform wavelet transform; Step 40: Extract features; Step 50: Build a model; Step 60: Identify and classify. By constructing a polarization information-time multi-dimensional data volume, the present invention makes full use of the echo information under four polarization configurations, enriches the polarization information of the target, comprehensively analyzes the target features, and adopts a high-dimensional continuous wavelet transform in time-frequency analysis, which has good spatial and wavenumber domain localization properties and can better analyze the instantaneous polarization characteristics of the target. Finally, a convolutional neural network is used to identify and classify the extracted target polarization features, realizing high-accuracy identification and classification of the target.

Description

Technical Field

[0001] The present invention relates to the technical field of radar signal processing. More specifically, the present invention relates to a millimeter-wave metal dangerous goods detection method based on the polarization wavelet domain. Background Art

[0002] The polarization of electromagnetic waves indicates the property that the orientation and amplitude of its electric field strength change with time. It reflects the vector characteristics of electromagnetic waves. The polarization domain where it is located is another important information domain that can reflect the characteristics of electromagnetic waves in addition to the time domain, frequency domain, and spatial domain. Through the polarization characteristics of electromagnetic waves, the polarization characteristics of the target can be obtained, thus enriching the comprehensive information of the target.

[0003] Traditional analysis and processing of radar echo signals often use the Fourier transform. However, the Fourier transform can only analyze the spectrum of the signal in the entire time domain and is difficult to perform local time-frequency analysis. As a result, when using the Fourier transform to process signals, it is difficult to separate useful signals at different frequencies. In 1946, Gabor introduced a window function to perform localized analysis on signals, which developed into the window Fourier transform (also known as the short-time Fourier transform). However, it has the disadvantage of a fixed localization format and is difficult to adaptively process singular signals and non-stationary signals. The wavelet transform inherits and develops the idea of localization of the short-time Fourier transform, and at the same time overcomes the disadvantages such as the window size not changing with frequency. It can provide a "time-frequency" window that changes with frequency, overcomes the above disadvantages and has the characteristics of translation and scaling, and has good localization characteristics in both the time-frequency domain. At the same time, using the information in the polarization domain and the wavelet domain can comprehensively analyze the target characteristics.

[0004] Therefore, a millimeter-wave metal dangerous goods detection method based on the polarization wavelet domain is proposed.

[0005] Summary of the Invention

[0006] In order to overcome the above defects of the prior art, an embodiment of the present invention provides a millimeter-wave metal dangerous goods detection method based on the polarization wavelet domain. By means of the wavelet transform, it overcomes the disadvantages such as the window size not changing with frequency and has the characteristics of translation and scaling. At the same time, using the information in the polarization domain and the wavelet domain can comprehensively analyze the target characteristics to solve the problems raised in the above background art.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A millimeter-wave metal dangerous goods detection method based on the polarization wavelet domain specifically includes the following steps:

[0009] Step 10: Obtain echo signals, and use professional receiving tools to receive information under four polarization configurations respectively, receiving four different echo signals; the four signals are HH, HV, VH, and VV respectively.

[0010] Step 20: Construct a polarization information-time multi-dimensional data volume. Use HH, VV, HV, VH generated in Step 10 and time as five variables to generate a five-dimensional space of the polarization information-time multi-dimensional data volume; for the target echo under different moments and different polarization configurations, use the polarization information-time multi-dimensional data volume to generate different polarization characteristics.

[0011] Step 30: Wavelet transform. Perform a high-dimensional wavelet transform on the polarization information-time multi-dimensional data volume constructed in Step 20. For the mother wavelet in the wavelet transform, perform operations of translation, scaling, and rotation to generate continuous wavelets.

[0012] Step 40: Feature extraction. Extract the polarization distribution characteristics of the high-dimensional wavelet domain space obtained in Step 3. Use the modulus maxima of the 5DCWT coefficients to process the echo signals, generate and fix the main energy frequency bands; scan the four polarization configurations to generate the polarization distribution characteristics of the target under different polarization configurations.

[0013] Step 50: Build a model. Build a high-dimensional convolutional neural network (CNN) model composed of multiple layers of alternating convolution + pooling structures.

[0014] Step 60: Identification and classification. Input the polarization distribution characteristics under different polarization configurations generated in Step 40 into the training model in Step 50 to generate a specific convolutional neural network; perform identification and classification on the trained convolutional neural network to achieve the detection and identification of metal dangerous goods.

[0015] In a preferred embodiment, the operations on the mother wavelet in Step 30 are specifically defined as follows:

[0016] The five-dimensional continuous wavelet transform (5DWCT) is defined as:

[0017]

[0018] where f(x) is a five-dimensional signal, is the five-dimensional mother wavelet, X = (x, y, z, p, q) T is a five-dimensional vector, b is the translation factor of the wavelet, b = (b x , b y , b z , b p , b q ), T a is the scale factor of the wavelet, is the rotation factor of the wavelet transform.

[0019] In a preferred embodiment, the four different echo signals in step 10 are specifically as follows: HH is the horizontal transmit - horizontal receive signal, HV is the horizontal transmit - vertical receive signal, VH is the vertical transmit - horizontal receive signal, and VV is the vertical transmit - vertical receive signal.

[0020] In a preferred embodiment, the specific operation steps for building the high - dimensional convolutional neural network model in step 50 are as follows:

[0021] Step 501: Pre - process the polarization distribution features generated in step 40 to generate a data set;

[0022] Step 502: Input the data set generated in step 501 into the convolutional neural network model for training to generate a training model;

[0023] Step 503: Adjust the network structure and parameters of the training model generated in step 502 according to the recognition accuracy until a relatively high recognition accuracy is achieved, and generate the current training model.

[0024] In a preferred embodiment, in step 20, eigenvalue extraction is performed on the target echo under different polarization configurations to generate different polarization features.

[0025] In a preferred embodiment, the translation factor for the operation on the mother wavelet in step 30 is a five - dimensional vector, and the rotation operation contains four variables.

[0026] In a preferred embodiment, the storage medium stores computer - executable instructions for performing the millimeter - wave metal dangerous goods detection method based on the polarization wavelet domain described in any one of the above.

[0027] In a preferred embodiment, it includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the millimeter - wave metal dangerous goods detection method based on the polarization wavelet domain described in any one of the above.

[0028] The technical effects and advantages of the present invention:

[0029] 1. The present invention proposes a millimeter-wave metal dangerous goods detection method based on the polarization wavelet domain. By collecting the echo signals of a full-polarization millimeter-wave radar and constructing a polarization information-time multi-dimensional data volume, time-frequency analysis is performed on it in the high-dimensional wavelet domain to obtain the polarization distribution characteristics of the target, and then it is fed into a convolutional neural network for classification and recognition, so as to detect and identify the metal dangerous goods carried by pedestrians, and it can be well applied in the millimeter-wave security inspection and detection system;

[0030] 2. The present invention has the following advantages compared with the prior art: First, a polarization information-time multi-dimensional data volume is constructed, which makes full use of the echo information under four polarization configurations, enriches the polarization information of the target, and comprehensively analyzes the target characteristics. Second, when performing time-frequency analysis, a high-dimensional continuous wavelet transform is used, which has multi-scale and inclination and azimuth angle selection characteristics, and has good spatial and wavenumber domain localization properties, and can better analyze the instantaneous polarization characteristics of the target. Finally, the convolutional neural network is used to identify and classify the extracted target polarization characteristics, realizing high-accuracy identification and classification of the target. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a flowchart of a millimeter-wave metal dangerous goods detection method based on the polarization wavelet domain proposed by the present invention.

[0032] Figure 2 It is a schematic diagram of the principle of wavelet decomposition proposed by the present invention.

[0033] Figure 3 It is a schematic diagram of the basic structure of the convolutional neural network proposed by the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0034] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0035] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0036] As shown in the attached Figures 1-3 A millimeter-wave metal dangerous goods detection method based on the polarization wavelet domain specifically includes the following steps:

[0037] Step 10: Obtain echo signals. Use professional receiving tools to receive information under four polarization configurations respectively, and receive four different echo signals; the four signals are HH, HV, VH, and VV, where HH is the horizontally transmitted - horizontally received signal, HV is the horizontally transmitted - vertically received signal, VH is the vertically transmitted - horizontally received signal, and VV is the vertically transmitted - vertically received signal.

[0038] Step 20: Construct a polarization information - time multi - dimensional data volume. Use HH, VV, HV, VH generated in Step 10 and time as 5 variables to generate a five - dimensional space of the polarization information - time multi - dimensional data volume; for the target echo at different times and different polarization configurations, use the polarization information - time multi - dimensional data volume to generate different polarization characteristics.

[0039] Step 30: Wavelet transform. Perform a high - dimensional wavelet transform on the polarization information - time multi - dimensional data volume constructed in Step 20. For the mother wavelet in the wavelet transform, perform operations of translation, dilation, and rotation to generate a continuous wavelet.

[0040] Step 40: Feature extraction. Extract the polarization distribution characteristics from the high - dimensional wavelet domain space obtained in Step 3. Use the modulus maxima of the 5DCWT coefficients to process the echo signals, generate and fix the main energy frequency bands; scan the four polarization configurations to generate the polarization distribution characteristics of the target under different polarization configurations.

[0041] Step 50: Build a model. Build a high - dimensional convolutional neural network (CNN) model composed of multiple layers of alternating convolution + pooling structures.

[0042] Step 60: Recognition and classification. Input the polarization distribution characteristics under different polarization configurations generated in Step 40 into the training model in Step 50 to generate a specific convolutional neural network; perform recognition and classification on the trained convolutional neural network to achieve the detection and recognition of metal dangerous goods.

[0043] Refer to Figure 2 , in the operation on the mother wavelet in Step 30, the translation factor is a five - dimensional vector, and the rotation operation contains four variables. The specific definition of the operation on the mother wavelet is as follows:

[0044] The five - dimensional continuous wavelet transform (5DWCT) is defined as:

[0045]

[0046] where f(x) is a five - dimensional signal, is the five - dimensional mother wavelet, x = (x, y, z, p, q) T is a five - dimensional vector, b(b = (b x , b y , bz , b p , b q ) T ) is the translation factor of the wavelet, a is the scale factor of the wavelet, is the rotation factor of the wavelet transform.

[0047] Referring to Figure 3 , the specific operation steps for building the high-dimensional convolutional neural network model in step 50 are as follows:

[0048] Step 501: Preprocess the polarization distribution features generated in step 40 to generate a data set;

[0049] Step 502: Input the data set generated in step 501 into the convolutional neural network model for training to generate a training model;

[0050] Step 503: Adjust the network structure and parameters of the training model generated in step 502 until a relatively high recognition accuracy is achieved to generate the current training model.

[0051] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A millimeter-wave metal dangerous goods detection method based on the polarization wavelet domain, characterized in that: Specifically, it includes the following steps: Step 10: Obtain echo signals, and use professional receiving tools to receive information under four polarization configurations respectively, and receive four different echo signals; the four signals are HH, HV, VH, and VV; Step 20: Construct a polarization information-time multi-dimensional data body, use HH, VV, HV, VH generated in Step 10 and time as 5 variables to generate a five-dimensional space of the polarization information-time multi-dimensional data body; for the target echo at different times and different polarization configurations, use the polarization information-time multi-dimensional data body to generate different polarization characteristics; Step 30: Wavelet transform, perform a high-dimensional wavelet transform on the polarization information-time multi-dimensional data body constructed in Step 20, and perform operations such as translation, scaling, and rotation on the mother wavelet in the wavelet transform to generate a continuous wavelet; Step 40: Feature extraction, extract the polarization distribution characteristics of the high-dimensional wavelet domain space obtained in Step 3, use the modulus maximum value of the 5DCWT coefficient to process the echo signal, generate the main energy frequency band and fix it; scan the four polarization configurations to generate the polarization distribution characteristics of the target under different polarization configurations; Step 50: Build a model, build a high-dimensional convolutional neural network (CNN) model composed of a multi-layer alternating convolution + pooling structure; Step 60: Recognition and classification, input the polarization distribution characteristics under different polarization configurations generated in Step 40 into the training model in Step 50 to generate a specific convolutional neural network; perform recognition and classification on the trained convolutional neural network to achieve the detection and recognition of metal dangerous goods.

2. The millimeter-wave metal dangerous goods detection method based on the polarized wavelet domain according to claim 1, wherein: The operations on the mother wavelet in Step 30 are specifically defined as follows: The five-dimensional continuous wavelet transform (5DWCT) is defined as: Among them, f(x) is a five-dimensional signal, is a five-dimensional mother wavelet, X = (x, y, z, p, q) T is a five-dimensional vector, b is the translation factor of the wavelet, b = (b x , b y , b z , b p , b q ), T a is the scale factor of the wavelet, is the rotation factor of the wavelet transform.

3. A millimeter-wave metal dangerous goods detection method based on the polarization wavelet domain according to claim 1, characterized in that: The four different echo signals in Step 10 are specifically as follows: HH is a horizontal transmission-horizontal reception signal, HV is a horizontal transmission-vertical reception signal, VH is a vertical transmission-horizontal reception signal, and VV is a vertical transmission-vertical reception signal.

4. A millimeter-wave metal dangerous goods detection method based on the polarization wavelet domain according to claim 1, characterized in that: The specific operation steps for building the high-dimensional convolutional neural network model in Step 50 are as follows: Step 501: Preprocess the polarization distribution characteristics generated in Step 40 to generate a data set; Step 502: Input the data set generated in Step 501 into the convolutional neural network model for training to generate a training model; Step 503: Adjust the network structure and parameters of the training model generated in Step 502 according to the recognition accuracy until a higher recognition accuracy is obtained to generate the current training model.

5. A millimeter-wave metal dangerous goods detection method based on the polarization wavelet domain according to claim 1, characterized in that: In Step 20, feature values are extracted from the target echo under different polarization configurations to generate different polarization characteristics.

6. A millimeter-wave metal dangerous goods detection method based on the polarization wavelet domain according to claim 2, characterized in that: The translation factor for the operation on the mother wavelet in Step 30 is a five-dimensional vector, and the rotation operation contains four variables.

7. A storage medium, characterized in that: The storage medium stores computer-executable instructions for executing the millimeter-wave metal dangerous goods detection method based on the polarization wavelet domain according to any one of claims 1-6.

8. An electronic device, characterized in that: It includes: At least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the millimeter-wave metal dangerous goods detection method based on the polarization wavelet domain according to any one of claims 1-6.

Images