WO2021203364A1

WO2021203364A1 - Radar signal processing method for identity recognition

Info

Publication number: WO2021203364A1
Application number: PCT/CN2020/083985
Authority: WO
Inventors: 王勇; 曹佳禾; 陈君毅
Original assignee: 浙江大学
Priority date: 2020-04-09
Filing date: 2020-04-09
Publication date: 2021-10-14

Abstract

A radar signal processing method for identity recognition. The method comprises: first reading a radar echo signal which is received by a radar sensor and is reflected by a recognition target, performing frequency mixing on the reflected signal and a signal emitted by a radar, performing fast Fourier transform on the signal subjected to frequency mixing, and filtering out a wave band having abnormal energy; obtaining signals of different frequency bands by means of N band-pass filters; then normalizing amplitudes of the signals of different frequency bands; and finally, performing weighted averaging on the normalized signal according to a preset weight. According to the method, low-frequency noise in the signals can be reduced without increasing redundant data, the signals of different frequency bands are reserved, the identity recognition accuracy is greatly improved, and the convergence speed of a classification model is increased.

Description

A radar signal processing method for identity recognition

Technical field

The present invention relates to the field of radar signal processing, in particular to a radar signal processing method for identity recognition.

Background technique

Identity recognition is ubiquitous in our daily lives. When you pick up the phone, the phone recognizes you. When you turn on the computer, the computer verifies your identity. When you check in at work, the check-in machine confirms your identity. At present, the mainstream identification methods include contact identification and non-contact identification. Contact recognition mainly includes fingerprint recognition, palmprint recognition, etc. Non-contact recognition is mainly image-based recognition methods, such as face recognition and iris recognition. These two types of identification methods have their own shortcomings: contact identification methods have hidden health hazards and non-contact identification methods will inevitably bring about privacy leakage problems, especially in public places.

A radar system usually has multiple antennas. Some antennas are used to transmit signals, and the other antennas are used to receive signals reflected by objects. Through these signals, we can obtain information about the target object. But the original radar signal will contain a lot of noise, and sometimes these noises will seriously affect the acquisition of information. For example, the signal received by the receiving antenna is not completely reflected by the detected object, but the signal transmitted by the transmitting antenna is directly transmitted to the receiving antenna, which will cause the received signal to have strong low-frequency noise.

If the radar signal cannot be properly processed, these powerful noises will make your signal useless numbers, and the really useful signals will be buried in these noises. If only a high-pass filter with a single cut-off frequency is used, a lot of information will be lost in the signal and the recognition accuracy will be affected. Therefore, how to design a data processing technology that can not only reduce the influence of noise in the signal, but also retain most of the frequency band information of the signal, is an urgent problem in radar signal identification.

Summary of the invention

In view of this, the purpose of the present invention is to provide a radar signal processing method for identity recognition, which can solve the problem that the noise energy in the original radar signal is large, and it is difficult to extract useful identity feature information from the signal, thereby affecting the accuracy of identity recognition. technical problem.

In order to solve the above technical problems, the present invention provides a radar signal processing method for identity recognition, which includes the following steps:

Step 1. Read the radar echo signal received by the radar sensor and reflected by the identified target.

Step 2. Mix the reflected signal with the signal emitted by the radar.

Step 3. Perform fast Fourier transform on the mixed signal to detect whether there is a frequency band with abnormal energy caused by noise. If it exists, go to step 4, otherwise go to step 5.

Step 4. Use the Butterworth band stop filter to filter out the abnormal frequency bands detected in step 3.

Step 5. Divide the entire frequency band into N parts, and establish N Butterworth bandpass filters corresponding to the cut-off frequency; use N filters to filter the data after step 4 to obtain N groups of filtered data .

Step 6. Calculate the absolute value of each group of filtered data first, and then calculate the average value; according to the average value obtained from each group, multiply the N groups of filtered data by different coefficients to normalize the amplitude, so that each The average value of a group of data after calculating the absolute value becomes the same.

Step 7. Perform a weighted average of the N groups of amplitude normalized data obtained in Step 6 according to the preset weight ratios of different frequency bands, combine the N groups of data into one group, and use the combined radar data for identification .

Further, in the step 3, the way to detect the frequency band with abnormal energy is: first perform constant false alarm detection on the frequency domain information after the fast Fourier transform, and determine the position and number of the target detected by the radar; if the radar detects too much For each object, only the information of the object closest to the radar is kept, and the frequency part of other detected objects is regarded as noise interference.

Further, in the step 5, the division of N frequency bands is based on the actual collected data, weighing speed and accuracy settings, the larger N is, the higher the recognition accuracy is, and the smaller the N is, the faster the data processing speed is.

Further, in the step 5, the frequency ranges included in the N different frequency bands may be the same, or the frequencies included in each frequency band may be set according to actual conditions.

Further, in the step 7, the weight coefficients of the data of different frequency bands may be the same, or weights of different proportions may be selected according to the actual situation, and a greater weight may be assigned to the frequency band corresponding to the distance at which the identification target may appear.

Further, the recognition target may be the palm, sole or face, and the recognition target is directly facing and hanging above the radar sensor.

The invention has the beneficial effects: the radar signal processing method of the invention greatly alleviates the noise interference in the original signal, and retains the information of each frequency band, and improves the accuracy of recognition. In addition, the data processed by the method of the present invention can make the classification model converge faster than other methods, and the fluctuation of the recognition accuracy rate is also smaller.

Description of the drawings

Figure 1 is a flow chart of the original radar signal processing;

Figure 2 is a schematic diagram of the frequency domain of the original radar signal;

Fig. 3 is a schematic diagram of the frequency domain of data processed by the method of the present invention.

Detailed ways

The specific implementation manners of the embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific implementations described here are only used to illustrate and explain the embodiments of the present invention, and are not used to limit the embodiments of the present invention.

Without loss of generality, this implementation example is based on the palm radar echo signal of 70G millimeter wave FMCW radar for radar signal processing, and 21 individuals are identified in total, including the following steps:

Step 1. Place the radar sensor on the desktop, place the hand in the air about 15cm above the radar sensor, and read the radar echo signal received by the radar sensor and reflected by the palm, which is recorded as S1.

In this embodiment, the palm is taken as an example. In addition, the face, the sole of the foot, etc. can also be used as the identification object, and the implementation principle is the same.

Step 2. Mix the reflected signal S1 with the signal S2 emitted by the radar, and the signal obtained after mixing

Wherein ₁ and ₂ respectively represent the frequency of the transmitted signal and the signal reflected by the palm,

with

Denote the phases of the two signals respectively, and mark the mixed signal as D.

Step 3. Perform a fast Fourier transform on D to detect whether there is a frequency band with abnormally strong energy that is affected by noise, that is, an abnormal frequency band. If there is, continue to step 4; otherwise, go directly to step 5 and mark D as F . The frequency domain information of the original signal after the fast Fourier transform is shown in Figure 2. It can be seen that there are obvious signals with unusually strong energy in the center of the frequency domain, that is, in the low frequency band. This is because the mixed signal contains signals that are directly injected into the receiving antenna by the transmitting antenna without being reflected by the palm. It can be regarded as an abnormal signal and needs to be filtered out by a band-stop filter.

One way to detect abnormal energy frequency bands is as follows, but not limited to this: First, perform constant false alarm detection on the frequency domain information after the fast Fourier transform, and determine the position and number of targets detected by the radar; if the radar detects more than one For objects, only the information of the objects closest to the radar is kept, and the frequency parts of other detected objects are regarded as noise interference.

Step 4. Use the Butterworth band stop filter to filter out the abnormal frequency band detected in step 3, and mark the filtered data as F.

Step 5. Divide the entire frequency band into N parts, and establish N Butterworth bandpass filters corresponding to the cut-off frequency. For example: the frequency range is 0-100Hz, we set the frequency band to be divided into 2 parts, the first part is 0-60Hz, and the second part is 60-100Hz, then the cutoff frequencies of the two corresponding Butterworth bandpass filters are Set to 0-60Hz and 60-100Hz respectively. The N filters are used to filter F respectively, and N sets of filtered data are obtained.

In step 5, the division method of N frequency bands can be adjusted according to the actual situation, weighing the speed and accuracy setting. Generally speaking, the larger the N, the higher the accuracy of the final recognition, but the speed will be correspondingly slower. In this example In, the entire frequency band is divided into 5 parts. In the frequency domain of the FMCW radar mixing signal, the higher the frequency, the farther the object is from the radar. We filter out the frequency band corresponding to the possible distance of the hand (approximately 5-25cm) as the first group, and then use the frequency band corresponding to (0-5cm) as the second group, and divide the remaining frequency bands into 3 equally. Use 3 corresponding band-pass filters for filtering.

The frequency ranges included in the N different frequency bands can be the same, or the frequencies included in each frequency band can be set according to the actual situation.

Step 6. First find the absolute value of each group of filtered data, and then find the average value. According to the average value obtained in each group, the N groups of filtered data are multiplied by different coefficients to normalize the amplitude, so that the average value of each group of data after the absolute value becomes the same, and the N groups of amplitude The normalized data is denoted as A.

For example, N takes 5 groups, and the average results of each group are 1, 2, 3, 4, and 5 respectively. Using 1 as the normalization standard, the corresponding normalization coefficients for the 5 groups are 1, 1/2, 1 3. 1/4, 1/5, multiply each group of filtered data by its corresponding normalization coefficient, and mark the data after N groups of amplitude normalization as A.

Step 7. Perform a weighted average of A according to the preset weight ratios of different frequency bands, and merge the N groups of data into one group.

The different weight coefficients in step 7 can be adjusted according to specific conditions. In this example, the amplitude ratio of the 5 groups of waveforms in different frequency bands is 1.2:1:1:1:1, which means that the frequency band corresponding to the distance (approximately 5-25cm) that the hand may appear is given greater weight.

The data obtained by fast Fourier transform of the data obtained in step 7 is shown in Figure 3. It can be seen that the low-frequency noise of the data is no longer prominent after processing, and the signals of all frequencies are taken into account.

The classifier used in this example is a 3-layer perceptron model, and the parameters of each layer are 768-256-128 respectively. We use the cross-validation method to divide all training samples into 5 equally, numbered 1, 2, 3, 4, 5, the first set of experiments use numbers 1, 2, 3, 4 for training, 5 for testing, the second set of experiments use numbers 1, 2, 3, 5 for training, 4 for testing, and so on for 5 sets of experiments , And finally average the accuracy of the 5 groups of experiments. Experiments show that the accuracy of classification using raw data is about 4.5%; the accuracy of data after only using a high-pass filter to filter out low-frequency noise is about 78.4%; and the radar signal processing method proposed by the present invention can reach 97.6%. Accuracy.

In summary, the present invention adopts the method of segmented filtering on the original signal, normalized amplitude and re-weighted average, which well alleviates the influence of low-frequency noise on the signal, and takes into account the signals of different frequency bands to ensure the signal High-frequency signals with weaker strength will not be buried in low-frequency signals with stronger signal strength. This processing method can not only greatly improve the accuracy of identification, but also greatly accelerate the convergence speed of the model.

The above content is a further detailed description of the present invention in combination with specific preferred embodiments, and it cannot be considered that the specific implementation of the present invention is limited to these descriptions. For those of ordinary skill in the technical field to which the present invention belongs, a number of simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as belonging to the protection scope of the invention.

Claims

A radar signal processing method for identity recognition, characterized in that the method includes the following steps:

Step 1. Read the radar echo signal received by the radar sensor and reflected by the identified target.

Step 2. Mix the reflected signal with the signal emitted by the radar.

Step 3. Perform fast Fourier transform on the mixed signal to detect whether there is a frequency band with abnormal energy caused by noise. If it exists, go to step 4, otherwise go to step 5.

Step 4. Use the Butterworth band stop filter to filter out the abnormal frequency bands detected in step 3.

Step 5. Divide the entire frequency band into N parts, and establish N Butterworth bandpass filters corresponding to the cut-off frequency; use N filters to filter the data after step 4 to obtain N groups of filtered data .

Step 6. Calculate the absolute value of each group of filtered data first, and then calculate the average value; according to the average value obtained from each group, multiply the N groups of filtered data by different coefficients to normalize the amplitude, so that each The average value of a group of data after calculating the absolute value becomes the same.

Step 7. Perform a weighted average of the N groups of amplitude normalized data obtained in Step 6 according to the preset weight ratios of different frequency bands, combine the N groups of data into one group, and use the combined radar data for identification .
The radar signal processing method for identity recognition according to claim 1, characterized in that, in the step 3, the method of detecting abnormal energy frequency bands is: firstly, the frequency domain information after the fast Fourier transform is constant False alarm detection, to determine the location and number of targets detected by the radar; if the radar detects multiple objects, only the information of the object closest to the radar is retained, and the frequency part of other detected objects is regarded as noise interference.
The radar signal processing method for identity recognition according to claim 1, characterized in that, in said step 5, the division of N frequency bands is based on the actual collected data, weighing the speed and accuracy settings, and N The larger the value, the higher the recognition accuracy, and the smaller the value of N, the faster the data processing speed.
A radar signal processing method for identity recognition according to claim 1, characterized in that, in the step 5, the frequency ranges contained in the N different frequency bands can be the same, or each can be set according to actual conditions. The frequencies contained in a frequency band.
A radar signal processing method for identity recognition according to claim 1, characterized in that, in said step 7, the weight coefficients of data in different frequency bands can be the same, or different proportions of weights can be selected according to actual conditions. The frequency band corresponding to the distance at which the identification target may appear is given greater weight.
The radar signal processing method for identity recognition according to claim 1, wherein the recognition target can be a palm, a sole or a face, and the recognition target is directly facing and floating on the radar sensor.