Radar one-dimensional range profile multi-target detection method based on CFAR and isolated forest fusion

Abstract

The invention relates to a radar one-dimensional range profile multi-target detection method based on CFAR and isolated forest fusion. The radar one-dimensional range profile multi-target detection method comprises the steps of obtaining radar original data and performing preprocessing to obtain radar one-dimensional range profile data; detecting a target in the radar one-dimensional range profile data by using a CA-CFAR detection adaptive threshold technology and normalizing an energy value; calculating a target score through an isolated forest method; performing weighted fusion on results of the CA-CFAR detection and the isolated forest to obtain a fusion result, delimiting a new detection threshold to confirm a target, setting a new detection threshold # imgabs0 #, and judging whether the target exists or not; according to the method, CFAR detection and an isolated forest method are combined, false alarms caused by radar clutter interference are effectively suppressed, the risks of false detection and missing detection are reduced, the target detection accuracy is improved, the method is suitable for various clutter environments, the operation speed is high, the calculation complexity is low, and the radar multi-target detection requirement in a complex scene is met.

Description

Technical Field

[0001] The present invention relates to the technical field of radar signal multi-target detection, and particularly to a method for multi-target detection of radar one-dimensional range profiles based on the fusion of CFAR and isolation forest. Background Art

[0002] In a radar system, target detection is one of the key tasks; the traditional constant false alarm rate (CFAR) detection is a commonly used radar target detection method and has been widely applied in the field of radar signal processing. The traditional CFAR is mainly based on statistical theory, assuming that clutter follows a specific probability distribution (such as Gaussian distribution, Weibull distribution, etc.), and adaptively adjusts the detection threshold by estimating clutter parameters to maximize the target detection probability while keeping the false alarm probability unchanged. However, with the increasing complexity of radar application scenarios, some problems and deficiencies of the traditional CFAR detection algorithm have gradually emerged in practical applications: it is sensitive to the assumption of clutter distribution, and when the actual clutter distribution does not match the assumption, the detection performance will decrease significantly; moreover, in complex scenarios such as sidelobes, azimuth ambiguity, ghosts, strong coherent speckle noise, and multi-target environments, there are obvious performance bottlenecks in the detection of weak targets by the traditional CFAR detection algorithm, and false alarms are likely to occur, which limits the effectiveness of the CFAR detection algorithm in practical applications; in addition, current radar technology is developing towards high resolution, multi-target detection, and adaptability to complex environments. In some emerging radar application fields such as autonomous driving, unmanned aerial vehicle monitoring, and security monitoring, higher requirements are put forward for the accuracy and reliability of radar target detection. The traditional CFAR algorithm and single machine learning algorithms are difficult to meet the requirements for weak target detection in these applications. With the continuous development of machine learning technology, some anomaly detection algorithms based on machine learning have also been applied to the field of radar target detection; among them, the isolation forest algorithm, as an unsupervised learning anomaly detection algorithm, constructs multiple isolation trees by randomly splitting the data space and uses the path length to measure whether a data point is an anomaly point. Due to its advantages such as low computational complexity and suitability for high-dimensional data, it has received wide attention; however, when the isolation forest algorithm is applied to radar target detection, there are problems with insufficient performance and it cannot adapt to the characteristics of radar signals, and often the algorithm needs to be improved and optimized. Summary of the Invention

[0003] In view of this, the present invention provides a method for multi-target detection of radar one-dimensional range profiles based on the fusion of CFAR and isolation forest, which effectively solves the problem that radar target detection in complex scenarios is prone to false alarms in the prior art, and improves the accuracy of target detection and the self-adaptability to the environment.

[0004] To achieve the above object, the present invention provides a multi-target detection method for radar one-dimensional range profiles based on the fusion of CFAR and isolation forest, comprising the following steps: S1. Obtain the radar one-dimensional range profile data by acquiring the radar raw data and performing preprocessing; S2. Perform CFAR detection on the radar one-dimensional range profile data to obtain range cells, and perform normalization processing on the energy values of the range cells obtained by CFAR detection to obtain ; S3. Calculate the target scores corresponding to the indices of each range cell in the radar one-dimensional range profile data by using the isolation forest method ; S4. Perform weighting processing on the energy values of the normalized range cells of the CFAR detection results and the target scores corresponding to the indices of each range cell obtained by the isolation forest to obtain a new fusion result , and the expression is: ; wherein, represents the confidence weight factor of CFAR detection, represents the confidence weight factor of the isolation forest; Set a new detection threshold , filter out the targets below the detection threshold. When the new fusion result exceeds the new detection threshold , it is confirmed that there is a target; when the new fusion result does not exceed the new detection threshold , it is determined that there is no target, and the expression is: ; wherein, represents a decision function.

[0005] Preferably, the millimeter-wave radar reflected signal received by the receiving antenna is amplified by a low-noise amplifier (LNA) and then mixed with the millimeter-wave radar transmitted signal to obtain a complex baseband signal, and then the discrete radar raw data is obtained through low-pass filtering and analog-to-digital converter (ADC) sampling.

[0006] Preferably, the preprocessing includes performing one-dimensional Fourier transform processing on the radar raw data to obtain the radar one-dimensional range profile data.

[0007] Preferably, the CFAR detection adopts cell-averaging constant false alarm rate detection (CA-CFAR). The CA-CFAR detector includes a detection unit D and 2h reference units, and the sampling values of the reference units are averaged as the background power level estimate , the expression is: ; The detection threshold of the detection unit D The expression is: ; Wherein, represents the threshold coefficient.

[0008] Preferably, the one-dimensional range profile data of the radar is input into the detection unit D of the CA-CFAR detector, and the obtained range cell and the corresponding energy value , perform normalization processing on the energy value of the range cell to obtain the normalized energy value , the expression is: ; Wherein, represents the number of detection targets, represents the th range cell corresponding to the detection target, represents the th energy value corresponding to the detection target, represents the th energy value after normalization processing of the detection target, represents the maximum energy value in, represents the minimum energy value in; Compared with the adaptive threshold obtained by the CA-CFAR detector, if the input signal exceeds the adaptive threshold, it is determined that there is a target, and if the input signal does not exceed the adaptive threshold, it is determined that there is no target.

[0009] Preferably, the expression for the isolation forest to calculate the target score corresponding to the range cell index is: ; Wherein, represents the average path length of the isolation forest, represents the expected path length of the isolation forest, and the target score has a value range of ; represents the th target score of the range cell corresponding to the detection target. The closer the target score is to 1, the more likely the th detection target is a target point. The closer the target score is to 0, the less likely the th detection target is a target point.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention combines two methods, namely CFAR detection and Isolation Forest, for radar target detection, giving full play to the CFAR detection in false alarm probability control and the sensitivity of the Isolation Forest to abnormal data distribution, improving the accuracy and stability of radar target detection, reducing the risk of false detection or missed detection of a single algorithm, effectively solving the problems of poor detection performance of traditional CFAR detection methods for weak targets in complex scenarios and easy false alarms, maintaining good detection performance in different clutter environments and multi-target scenarios, effectively coping with various complex situations such as sidelobes, azimuth ambiguity, phantoms, and strong coherent speckle noise, enhancing the target detection ability of the radar system in complex environments, and expanding the applicability of the radar system in various application scenarios. Description of the Drawings

[0011] Fig. 1 is a flow chart of the integration of CFAR and Isolation Forest of the present invention; Figure 2 Fig. is a schematic diagram of the CA-CFAR detector of the present invention; Figure 3 Fig. is a flow chart of the Isolation Forest method of the present invention. Detailed Embodiments

[0012] To further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following will, in conjunction with the accompanying drawings and preferred embodiments, describe in detail the specific embodiments, structures, features, and effects of the present invention as follows.

[0013] A multi-target detection method for radar one-dimensional range profiles based on the integration of CFAR and Isolation Forest includes the following steps: S1. Configure the millimeter-wave radar waveform. Mix the millimeter-wave radar reflected signal received by the receiving antenna, which is amplified by a low-noise amplifier (LNA), with the millimeter-wave radar transmitted signal to obtain a baseband signal in complex form. Then, after low-pass filtering and sampling by an analog-to-digital converter (ADC), discrete radar raw data is obtained. Perform one-dimensional Fourier transform processing on the radar raw data to obtain the one-dimensional range profile data of the radar, which can be used to represent the variation of the intensity value (power) of the radar echo signal with power.

[0014] S2. Use the cell-averaging constant false alarm rate detection (CA-CFAR) to perform CFAR detection on the radar one-dimensional range profile data, obtaining the distance cell , and perform normalization processing on the energy value of the distance cell obtained by CFAR detection to obtain ; Traditional CFAR detection performs CFAR detection processing on the one-dimensional range profile data of the amplitude matrix obtained by accumulation after radar preprocessing accumulation, and realizes the detection of radar signals through an adaptive threshold technique. In this embodiment, the cell-averaging constant false alarm rate detection (CA-CFAR) is adopted; As shown Figure 2 in the figure, the CA-CFAR detector includes a detection unit D and 2h reference units. The reference units are located on both sides of the detection unit, with h units in the front and h units in the back. The main function of the protection unit P is to prevent the target energy from leaking into the reference units and affecting the detection effect in the case of a single target. The CA-CFAR detector obtains an adaptive threshold through a sliding window, and the sampled values of the reference units are averaged as an estimate of the background power level , and the expression is: ; The detection threshold of the detection unit D The expression is: ; Among them, represents the threshold coefficient; Input the one-dimensional range profile data of the radar into the detection unit D of the CA-CFAR detector, and obtain the range cell and the corresponding energy value . Normalize the energy value of the range cell to obtain the normalized energy value , and the expression is: ; Among them, represents the number of detected targets, represents the th range cell corresponding to the detected target, represents the th energy value corresponding to the detected target, represents the th normalized energy value of the detected target, represents the maximum energy value in, represents the minimum energy value in; Compared with the adaptive threshold obtained by the CA-CFAR detector, if the input signal exceeds the adaptive threshold, it is determined that there is a target. If the input signal does not exceed the adaptive threshold, it is determined that there is no target

[0015] S3. Use the Isolation Forest method to calculate the target score corresponding to each range cell index in the radar one-dimensional range profile data ; Isolation Forest belongs to the boosting tree model of unsupervised learning. For each subtree, a random value within the range of the feature values of the dataset is randomly selected for different feature values. According to the principle of whether the feature value is greater than or less than the random value, all data is divided into two parts according to the random value of this feature, gradually separating all the data; since abnormal data usually occupies a small proportion, they can be distinguished from normal data using fewer attributes, so the average path length can be used as a measurement index for the target; the isolation forest mainly includes the following steps: Initialize parameters, including the number of isolation trees to be constructed, the maximum tree depth of each isolation tree, the size of the samples used for each isolation tree, and the seed controlling the random process; The more the number of isolation trees, the higher the stability and accuracy of the model, but the computational cost will also increase accordingly; The greater the depth of the isolation tree, the better it can capture the details of the data, but it may lead to overfitting; The smaller the number of samples used for the isolation tree, the higher the efficiency of the model can be improved, but the accuracy may be reduced; Controlling the seed of the random process can ensure the repeatability of the results; Build the isolation tree model. The isolation forest reduces the bias caused by the randomness of a single tree through the integration of multiple trees, improving the stability and accuracy of the model, including: Randomly select a feature at each splitting node; Randomly select a splitting value within the range of the selected feature; Separate the dataset into the left subtree or the right subtree according to the splitting value until each data point is isolated or the maximum tree depth is reached; Form an isolation forest by constructing multiple isolation trees; Calculate the path length, calculate the path length of each data point in the isolation tree where it is located, and find the average value. The shorter the path length, the easier it is for the data point to be isolated, and thus the more likely it is to be the target point; specifically, it includes the following steps: Randomly select from the training data sample points to form the subset of ; For the current node (data subset ), and the randomly selected feature dimension , the cutting point expression is: ; Among them, represents all the value sets of the data at the current node in dimension , represents dimension Lower minimum value of the data, indicating the dimension lower maximum value of the data, indicating a value randomly sampled from a uniform distribution in; Select a cutting point Generate a hyperplane to divide the data space of the current node into 2 subspaces, and place the points less than the cutting point under the current selected dimension on the left branch of the current node, and place the points greater than or equal to the cutting point on the right branch of the current node; Recursively perform steps (2) and (3) until all leaf nodes have only one sample point or the isolation tree has reached the specified height; Loop and repeat steps (1)-(4) until the isolation tree is generated; For each distance unit, traverse each isolation tree and record the path length from the root node to the leaf node ; In the isolation forest, the construction process of each isolation tree can be regarded as gradually separating data points through random partitioning in a multi-dimensional space, and the path length refers to the number of partitions required to isolate the distance unit from the root node; Calculate the average of the path lengths in all isolation trees to obtain the average path length of the data points ; Normalize the path length and calculate the expected path length through the harmonic number , and the expression is: ; ; Among them, represents the total number of distance units. In this embodiment , represents the th harmonic number; Calculate the target score based on the average path length and the expected path length to calculate the target score of each data point , and the expression is: ; Among them, represents the average path length of the isolation forest, represents the expected path length of the isolation forest, and the target score ranges from ; Indicates the target score of the distance unit corresponding to the th detection target. The closer the target score is to 1, the path length is very small, and then the data points are easily isolated. Then the th detection target is more likely to be the target point; the closer the target score is to 0, the path length will increase, and then the th detection target is less likely to be the target point.

[0016] S4. Weight the energy value of the normalized distance unit of the CFAR detection result and the target score corresponding to each distance unit index obtained by the isolation forest to obtain a new fusion result , and the expression is: ; wherein, represents the confidence weight factor of the CFAR detection, represents the confidence weight factor of the isolation forest; Set a new detection threshold , filter the targets below the detection threshold. When the new fusion result exceeds the new detection threshold , it is confirmed that there is a target; when the new fusion result does not exceed the new detection threshold , it is determined that there is no target, and the expression is: ; wherein, represents a decision function, the input is the fusion result and the detection threshold, and the output is a binary decision (presence / absence of a target).

[0017] As described above, it is only a preferred embodiment of the present invention, and does not impose any form of limitation on the present invention. Although the present invention has been disclosed above with preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to the above-disclosed technical content within the scope of the technical solution of the present invention to obtain equivalent embodiments with equivalent changes. However, as long as it does not depart from the content of the technical solution of the present invention, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A multi-target detection method for radar one-dimensional range profiles based on the fusion of CFAR and isolation forest, characterized in that, Including the following steps: S1. Obtain the radar one-dimensional range profile data obtained by preprocessing the original radar data; S2. Perform CFAR detection on the radar one-dimensional range profile data to obtain range cells , perform normalization processing on the energy values of the range cells obtained by CFAR detection to obtain ; S3. Use the Isolation Forest method to calculate the target scores corresponding to the indices of each range cell in the radar one-dimensional range profile data ; S4. Normalize the energy value of the distance unit of the CFAR detection result The target score corresponding to each distance unit index obtained by the Isolation Forest Perform weighted processing to obtain a new fusion result , and the expression is: ； Among them, represents the confidence weight factor of CFAR detection, represents the confidence weight factor of the isolation forest; Set a new detection threshold , filter the targets below the detection threshold. When the new fusion result exceeds the new detection threshold , it is confirmed that there is a target; when the new fusion result does not exceed the new detection threshold , it is determined that there is no target. The expression is: ； Among them, represents a decision function.

2. The method for multi-target detection of radar one-dimensional range profiles based on the fusion of CFAR and isolation forest according to claim 1, wherein, The millimeter-wave radar reflected signal received by the receiving antenna is amplified by a low-noise amplifier (LNA) and then mixed with the millimeter-wave radar transmitted signal to obtain a baseband signal in complex form. Then, after low-pass filtering and sampling by an analog-to-digital converter (ADC), discrete original radar data is obtained.

3. The method for multi-target detection of radar one-dimensional range profile based on the fusion of CFAR and isolation forest according to claim 2, characterized in that The preprocessing includes performing one-dimensional Fourier transform processing on the original radar data to obtain the radar one-dimensional range profile data.

4. The method for multi-target detection of radar one-dimensional range profile based on the fusion of CFAR and isolation forest according to claim 1, characterized in that The CFAR detection adopts cell-averaging constant false alarm rate detection CA-CFAR. The CA-CFAR detector includes a detection unit D and 2h reference units, and the sampling values of the reference units are averaged as the background power level estimate , and the expression is: ； The detection threshold of the detection unit D The expression is: ; Among them, represents the threshold coefficient.

5. The method for multi-target detection of radar one-dimensional range profiles based on the fusion of CFAR and isolation forest according to claim 4, wherein, Input the one-dimensional range profile data of the radar into the detection unit D of the CA-CFAR detector to obtain the range cell and the corresponding energy value . Normalize the energy value of the range cell to obtain the normalized energy value . The expression is: ； Among them, represents the number of detection targets, represents the th distance unit corresponding to the detection target, represents the th energy value corresponding to the detection target, represents the th energy value after normalization processing of the detection target, represents the maximum energy value in represents the minimum energy value in Compared with the adaptive threshold obtained by the CA-CFAR detector, if the input signal exceeds the adaptive threshold, it is determined that there is a target; if the input signal does not exceed the adaptive threshold, it is determined that there is no target.

6. The method for multi-target detection of radar one-dimensional range profile based on the fusion of CFAR and isolation forest according to claim 1, characterized in that The expression for the Isolation Forest to calculate the target score corresponding to the distance cell index is: ； Among them, represents the average path length of the isolation forest, represents the expected path length of the isolation forest, and the target score has a value range of ; represents the target score of the distance unit corresponding to the th detection target.

Images