An object recognition method based on the combination of density clustering and amplitude-distance features

Abstract

The present invention relates to a target recognition method based on the combination of density clustering and amplitude-distance features, belonging to the technical fields of radar interference recognition and target recognition. The target recognition method includes: S1: performing non-coherent accumulation within each frame of data to obtain the data after non-coherent accumulation of this frame; S2: performing CFAR detection on the data after non-coherent accumulation obtained in S1 to obtain the points that cross the threshold after CFAR detection, and saving them; S3: performing density clustering frame by frame on the points that cross the threshold after CFAR detection output in S2 to obtain a clustering result; S4: performing recognition of targets and interferences frame by frame on the clustering result output in S3. The method can cluster data of any shape and has a better clustering effect for one-dimensional data; it does not need to preset initial values, so there is no influence of initial values on the clustering result; it makes full use of the intra-frame and inter-frame information of the data, and the target recognition accuracy is higher.

Description

Technical Field

[0001] The present invention relates to a target recognition method based on the combination of density clustering and amplitude-distance features, belonging to the technical fields of radar interference recognition and target recognition. Background Art

[0002] In the sea detection of anti-ship missiles, corner reflector interference is a typical passive deception interference, which is widely used in radar countermeasure scenarios. In order to effectively deceive anti-ship missiles, the enemy fleet usually arranges effective corner reflectors or corner reflector arrays around the ship, generating dilution and centroid interference during the search and tracking phases of the radar seeker, so as to deceive the radar seeker into detecting the ship. In order to achieve the precise strike of anti-ship missiles on high-value targets on the sea surface, it is very necessary to study the technology to counter corner reflector interference. At present, the methods for distinguishing corner reflector interference and ship targets mainly include: the method based on polarization decomposition and the method based on micro-Doppler features.

[0003] Although these methods have a certain anti-interference effect on corner reflector interference, it is difficult for the radar seeker to extract the polarization information of the target or interference. At the same time, due to the strong clutter on the ground / sea surface generating clutter spectra in the echo signal spectrum, the anti-interference method based on micro-Doppler features will also have very limited effects. Aiming at the above problems, the present invention is committed to exploring the combination of multi-dimensional features within and between frames of radar one-dimensional range profile data to realize the recognition of ship targets and corner reflector interference. Summary of the Invention

[0004] The purpose of the present invention is to propose a target recognition method based on the combination of density clustering and amplitude-distance features for the problems that the practicability of anti-interference by polarization processing is poor in the presence of corner reflector interference, and the anti-interference method based on micro-Doppler features is affected by clutter spectra, resulting in low target recognition efficiency.

[0005] In order to achieve the above purpose, the present invention adopts the following technical solutions.

[0006] The target recognition method based on the combination of density clustering and amplitude-distance features includes:

[0007] S1: Perform non-coherent integration within each frame of data to obtain the data after non-coherent integration of this frame;

[0008] S2: Perform CFAR detection on the data after non-coherent integration obtained in S1, obtain the points that cross the threshold after CFAR detection, and save them;

[0009] S3: Perform density clustering on the points that cross the threshold after CFAR detection output by S2 frame by frame to obtain the clustering result, which specifically includes the following sub-steps:

[0010] S31. Cluster the points that exceed the threshold after CFAR detection and output by S2 according to their density distribution on the one-dimensional range profile to obtain the density clustering result. Specifically: Calculate the Gaussian density around each point, select the points with higher density and larger intervals as the clustering centers, and divide other points into the category of the clustering center with the shortest distance to this point; Traverse all points to obtain the density clustering result;

[0011] S32. Perform secondary clustering on the density clustering result output by S31 to obtain the clustering result. Specifically: Calculate the distance between each category in the density clustering result output by S31, and merge the two categories with closer distance to obtain the clustering result;

[0012] S4: Identify targets and interferences frame by frame for the clustering result output by S3, which specifically includes the following sub-steps:

[0013] S41. Perform in-frame judgment based on the clustering of each frame of data output by S3 to obtain the type recognition result. Specifically: Calculate the amplitude entropy of each category in a frame of data. The category with the largest amplitude entropy is considered the ship target category, and the category with the smallest amplitude entropy is considered the interference category;

[0014] S42. Determine whether the type recognition result of S41 is correct. If the type recognition is correct, save the classification recognition result and target information. Otherwise, if the type recognition is incorrect, jump to S43. Specifically: Calculate the absolute value of the difference between the current distance of the ship category in S41 and the average distance of the ship category in the previous five frames; If the absolute value of the difference is too large, it is considered that the type recognition is incorrect, and jump to S43 for secondary judgment. Otherwise, it is considered that the type recognition is correct, and save the classification recognition result and target information;

[0015] Among them, the current distance of the ship category and the distance of each category in the ship category distance of the previous five frames refer to the distance corresponding to the highest amplitude point in this category;

[0016] S43. Use the inter-frame information of the data for secondary judgment to obtain the classification recognition result. Specifically: Calculate the variance between the distance of each category in the current frame and the ship category distance in the previous five frames respectively, and consider the category with the smallest variance as the ship;

[0017] So far, from S1 to S4, a target recognition method based on the combination of density clustering and amplitude-distance features has been completed.

[0018] Beneficial Effects

[0019] A target recognition method based on the combination of density clustering and amplitude-distance features proposed by the present invention has the following beneficial effects compared with the existing target recognition methods:

[0020] 1. The method can cluster data of any shape, while other clustering algorithms such as K-Means are generally only applicable to convex data sets, and the method has better clustering effect for the one-dimensional data adopted;

[0021] 2. The method does not need to preset initial values, so there is no influence of initial values on the clustering results;

[0022] 3. The method makes full use of the intra-frame and inter-frame information of the data, and the target recognition accuracy is higher. Description of the Drawings

[0023] Figure 1 is a flowchart of the target recognition method based on density clustering and amplitude-distance feature combination of the present invention;

[0024] Figure 2 is a clustering result diagram of the example data using the target recognition method based on density clustering and amplitude-distance feature combination of the present invention in the corner reflector interference scenario and using the K-means clustering algorithm;

[0025] Figure 3 is a target and interference recognition result diagram of the target recognition method based on density clustering and amplitude-distance feature combination of the present invention in the corner reflector interference scenario. Detailed Embodiments

[0026] The following combines the drawings and specific embodiments to elaborate in detail on a target recognition method based on density clustering and amplitude-distance feature combination of the present invention.

[0027] Embodiment 1

[0028] The method proposed by the present invention can theoretically be applied to the target and interference recognition of any radar one-dimensional range profile data, such as target recognition under corner reflector interference. Usually, for target recognition under corner reflector interference, micro-Doppler and polarization information are used. On the one hand, it is difficult to extract polarization information; on the other hand, due to the influence of clutter, the anti-interference effect of micro-Doppler information is very limited. The present invention is committed to exploring the combination of multi-dimensional features within and between frames of radar one-dimensional range profile data to realize the recognition of ship targets and corner reflector interference. The specific implementation of the target recognition method based on density clustering and amplitude-distance feature combination is as Figure 1 shown. The method only needs to change the parameter settings of the application scenario.

[0029] This embodiment describes the specific implementation of applying the target recognition method based on density clustering and amplitude-distance feature combination of the present invention to recognize ship targets and corner reflector interference respectively in the corner reflector interference scenario.

[0030] First, perform pulse compression, non-coherent integration, and CFAR detection on the echo signals received by the radar. There are a total of 500 frames of data, and each frame of data includes 64 pulses.

[0031] Next, cluster the points that pass the threshold after CFAR detection for each frame.

[0032] Here, the clustering of the method is divided into two steps. In the first step, density clustering is performed on all data, and the data is divided into many small classes according to Gaussian density. Here, the number of classes does not need to be set in advance. According to the respective characteristics of each frame of data, different classes are divided. Taking the points that pass the threshold after CFAR detection of the tenth frame as an example, after the first density clustering, a total of 3 classes are divided; in the second step, the class spacing is calculated, and the classes with a class spacing less than 50 are merged. Calculate the class spacing of these 3 classes in the tenth frame. The class spacing of all of them is greater than 50, and the final clustering result of this frame is still 3 classes, denoted as cluster1, cluster2, and cluster3. Traverse all frames according to this method to obtain the clustering results of all frames as Figure 2 shown in a;

[0033] Here, the K-Means algorithm is used to cluster 500 frames of data, and the number of clusters is fixed at 4. The clustering result is as Figure 2 shown in b. It can be seen from the figure that many points that should be in the second class are assigned to the third class, and its classification effect and accuracy are not as good as those of the classification method.

[0034] Then, identify the clustering results. First, use the amplitude entropy within the frame for identification. Calculate the amplitude entropy of each class. The class with the largest amplitude entropy is considered to be a ship, and the class with the smallest amplitude entropy is considered to be a corner reflector. Here, still taking the tenth frame of data as an example, calculate the amplitude entropy of cluster1, cluster2, and cluster3 as: 1.8512, 1.7913, 3.1542. The cluster3 with the largest amplitude entropy is considered to be the ship class, denoted as ship_clust, and the cluster2 with the smallest amplitude entropy is considered to be the corner reflector class, denoted as corner_clust.

[0035] Among them, the definition of amplitude entropy is shown in Equation (2), where z(i), i = 1, 2,..., n is the amplitude of the scatterer, n is the total number of data, and p i represents the probability of z(i) appearing. Entropy is used to characterize the spread degree of energy. The more concentrated the energy, the smaller the entropy value.

[0036]

[0037]

[0038] Then, the data is secondarily judged using the inter-frame distance information. Calculate the difference between the ship distance in this frame and the average value of the ship distances in the previous five frames. If the absolute value is greater than 100, it is considered that the recognition result may be incorrect. Then calculate the variance of the distances of all classes and the ship distances in the previous five frames, and take the class with the smallest variance as the new ship class, and update ship_clust; if the absolute value is less than 100, it is considered that the recognition is correct and the next step can be directly carried out;

[0039] For the data of the tenth frame calculated by the method, the difference between the ship distance and the average value of the ship distances in the previous five frames is 2.2, and its absolute value is less than 100, so it is considered that the recognition is correct.

[0040] Similarly, calculate the difference between the corner reflector anti-jamming distance in this frame and the average value of the corner reflector anti-jamming distances in the previous five frames. If the absolute value is greater than 90, it is considered that the recognition may be incorrect. Then calculate the variance of the distances of all classes except the ship class and the corner reflector distances in the previous five frames, and take the class with the smallest variance as the new corner reflector class, and update corner_clust; if the absolute value is less than 90, it is considered that the recognition is correct and the next step can be directly carried out;

[0041] For the data of the tenth frame calculated by the method, the difference between the corner reflector anti-jamming distance and the average value of the corner reflector anti-jamming distances in the previous five frames is 44.4, and its absolute value is less than 100, so it is considered that the recognition is correct.

[0042] In addition, in the secondary judgment of the method, the information of the previous five frames of data needs to be used. Therefore, the first five frames of data in the method are not secondarily judged, and all the distances mentioned above are distance units.

[0043] Then, select the one with the largest amplitude in the ship class and the corner reflector class as the representative point for result output. After performing the above operations on all 500 frames of data, output the point trace diagrams of the ship target and the corner reflector interference, as Figure 3 shown. Only using the intra-frame amplitude entropy information for recognition, the target recognition accuracy rate is 90.45%. After using the inter-frame distance information for secondary judgment, the target recognition accuracy rate is 95.25%. Compared with only using a single target feature for recognition, the target recognition accuracy rate of the amplitude-distance multi-dimensional feature joint recognition method has increased by 4.8%.

[0044] So far, the specific implementation of the target recognition method based on density clustering and amplitude-distance feature joint processing described in the present invention for target and interference recognition in the corner reflector interference scenario has been completed.

[0045] The above is the preferred embodiment of the present invention. The present invention should not be limited to the content disclosed in this embodiment and the drawings. Any equivalent or modification completed without departing from the spirit disclosed by the present invention falls within the protection scope of the present invention.

Claims

1. A target recognition method based on the combination of density clustering and amplitude-distance features, characterized in that: Including: S1: Perform non-coherent integration within each frame of data to obtain the data after non-coherent integration for that frame; S2: Perform CFAR detection on the data after non-coherent integration obtained in S1, obtain the points that cross the threshold after CFAR detection, and save them; S3: Perform density clustering frame by frame on the points that cross the threshold after CFAR detection output by S2 to obtain a clustering result; S4: Perform target and interference recognition frame by frame on the clustering result output by S3, specifically including the following sub-steps: S41: Perform in-frame judgment based on the clustering of each frame of data output by S3 to obtain a type recognition result. Specifically: Calculate the amplitude entropy of each class in a frame of data. The class with the largest amplitude entropy is considered the ship target class, and the class with the smallest amplitude entropy is considered the interference class; S42: Judge whether the type recognition result in S41 is correct. If the type recognition is correct, save the classification recognition result and target information. Otherwise, if the type recognition is incorrect, jump to S43; S43: Use the inter-frame information of the data for secondary judgment to obtain a classification recognition result. Specifically: Calculate the variance between the distances of each class in the current frame and the distances of the ship classes in the previous five frames. The class with the smallest variance is considered the ship; 2. The target recognition method according to claim 1, characterized in that: S3 specifically includes the following sub-steps: S31: Cluster the points that cross the threshold after CFAR detection output by S2 according to the density distribution on the one-dimensional range profile to obtain a density clustering result; S32: Perform secondary clustering on the density clustering result output by S31 to obtain a clustering result. Specifically: Calculate the distances between each class in the density clustering result output by S31, and merge the two classes with closer distances to obtain a clustering result; 3. The target recognition method according to claim 2, characterized in that: S31 specifically is: Calculate the Gaussian density around each point, select the points with higher density and larger intervals as clustering centers, and divide other points into the class of the clustering center with the shortest distance to that point; Traverse all points to obtain a density clustering result; 4. The target recognition method according to claim 1, characterized in that: S42 specifically is: Calculate the absolute value of the difference between the current distance of the ship class in S41 and the average distance of the ship classes in the previous five frames; If the absolute value of the difference is too large, it is considered that the type recognition is incorrect, and jump to S43 for secondary judgment. Otherwise, it is considered that the type recognition is correct, and save the classification recognition result and target information; 5. The target recognition method according to claim 4, characterized in that: In S42, the current distance of the ship class and the distance of each class in the distances of the ship classes in the previous five frames refer to the distance corresponding to the highest amplitude point in that class.

Images