Laser push-broom imaging fuze point cloud rapid identification method based on two-dimensional GRA

Abstract

The invention discloses a laser push-broom imaging fuze point cloud rapid identification method based on a two-dimensional GRA, and belongs to the technical field of laser fuze target identification. The method is applied to the non-rotating laser linear array push-broom imaging fuze, and fast identification of the target is completed in a mode of processing while scanning. Generating labeled target linear array distance information and target intensity information by an actual measurement and simulation method to form a point cloud two-dimensional sequence priori knowledge base; during fuze horizontal-pushing scanning detection, linear array distance information and intensity information of a target scene are continuously obtained; and constructing an identification model by adopting two-dimensional grey correlation analysis, loading the model in the laser fuze, and performing feature matching on a two-dimensional feature sequence which is acquired and processed in real time to complete rapid identification of a battlefield target.

Description

Technical Field

[0001] The present invention belongs to the technical field of laser fuze target recognition, and particularly relates to a method for quickly recognizing point clouds of a laser push-broom imaging fuze based on two-dimensional GRA (Grey Relational Analysis). Background Art

[0002] The essence of a two-dimensional image is obtained by projecting a three-dimensional image, which causes a two-dimensional object to lose many characteristic attributes of a three-dimensional object. At the same time, the future battlefield environment will become more complex, and simple target recognition algorithms based on one-dimensional images and two-dimensional images often fail due to occlusion or interference from other complex battlefield environments. The three-dimensional point cloud obtained by a laser push-broom imaging fuze is the three-dimensional scale information of a target in the real world, which contains more information, such as distance, intensity, coordinates, etc. It is not easily affected by external factors such as the environment. When detecting a target, target information can be directly obtained without adding other detectors, and the warhead can be detonated at the best time to improve the target hit rate and damage effectiveness.

[0003] The laser push-broom imaging system is an imaging method based on a linear array optoelectronic detector. In essence, it uses a linear array laser beam to scan the object to be recorded, and according to the missile-target encounter motion, cooperates with steps such as hardware circuits and software processing to complete the acquisition, storage, and processing of target scene information. However, the current mainstream point cloud target recognition methods are based on the complete three-dimensional point cloud of the target, such as PointNet, OctNet, PV-RCNN++, etc., which have extremely high requirements for the computing power and storage capacity of the carrier system. Given that there is a time interval between each frame of data obtained by laser line scanning, this time difference can be fully utilized for data preprocessing to reduce the algorithm processing time.

[0004] The grey system theory takes an uncertain system with "partially known information and partially unknown information" and "small samples and poor information" as the research object. During the process of a laser fuze imaging on the ground, information such as scanning parameters and the target to be recognized in the knowledge base is determined, while information such as the scene of the scanning object and the missile-target encounter angle is unknown. And due to the weak hardware environment of the fuze, the amount of information obtained by the fuze is limited, which is a typical uncertain system with "small samples and poor information". Grey relational analysis is an important branch of the grey system theory, which can judge whether the relationship between different sequences is close according to the geometric shape and similarity degree of the sequence curves, and has a small amount of calculation, which is very suitable for the field of fuze target recognition. However, the current research on grey relational analysis is mainly one-dimensional grey relational analysis, and there is less research on two-dimensional grey relational analysis. Summary of the Invention

[0005] The object of the present invention is to provide a fast recognition method for point clouds of laser push-broom imaging fuzes based on two-dimensional GRA, which can quickly and effectively complete the recognition of specific targets. A multi-dimensional prior knowledge base is constructed by actual measurement and simulation methods according to the imaging method and target characteristics of the laser push-broom imaging fuze. A recognition model is constructed by two-dimensional grey relational analysis, and the model is loaded into the laser fuze to perform feature matching on the two-dimensional feature sequences obtained and processed in real time by the fuze push-broom, so as to complete the fast recognition of battlefield targets.

[0006] The technical solution for realizing the present invention is as follows: A fast recognition method for point clouds of laser push-broom imaging fuzes based on two-dimensional GRA includes the following steps:

[0007] Step 1: Taking a moving object as a preset reference target, collecting data through actual measurement by a laser linear array push-broom imaging fuze or simulating the radar detection process using a laser point cloud imaging virtual simulation platform. By changing the missile-target encounter angle, the target linear array distance information and intensity information in the current state are obtained, and the minimum value operation is performed on the above information. The minimum value is continuously obtained through missile-target encounter, and sorting and normalization processing are carried out to form a two-dimensional prior knowledge base of target point clouds with labeled information.

[0008] Step 2: The laser linear array imaging fuze performs push-broom detection on the target, and performs linear array push-broom on the front lower part of the laser fuze to sequentially obtain the target linear array distance information and intensity information in real time.

[0009] Step 3: Preprocess the target linear array distance information and intensity information obtained in real time, filter out ground points and abnormal noise points, reduce the influence of interference points on the recognition result, and obtain clearer and more representative target point cloud data.

[0010] Step 4: Perform minimum value taking, sorting, and data generation alignment operations on the target point cloud data obtained in Step 3 to obtain the aligned target linear array distance information and intensity information.

[0011] Step 5: Design a two-dimensional grey relational analysis method, establish a classification and matching model based on two-dimensional grey relational analysis, and load the two-dimensional prior knowledge base of target point clouds with labeled information and the classification and matching model into the MCU of the laser linear array push-broom imaging fuze.

[0012] Step 6: Input the aligned target linear array distance information and intensity information obtained in Step 4 into the classification and matching model obtained in Step 5 to achieve fast target recognition; for different targets, the fuze controls different detonation points to detonate, forming different warheads.

[0013] The remarkable effect of the present invention compared with the prior art is:

[0014] 1) Make full use of the time interval between two frames of data obtained in real time by the laser push-broom imaging fuse, which shortens the extraction time of the two-dimensional feature sequence. After being deployed to an embedded system such as an FPGA, the parallel data processing method can be adopted, greatly improving the processing speed of the algorithm.

[0015] 2) By using three-dimensional point cloud data, the influence of illumination and scale changes on the recognition effect can be avoided, and it has a certain robustness to occlusion.

[0016] 3) The recognition algorithm adopts grey relational analysis, which greatly reduces the scale of the knowledge base, reduces the number of parameters of the matching model, and reduces the hardware requirements of the fuse.

[0017] 4) A two-dimensional grey relational analysis model is designed to reduce the computational load of the control chip. It considers the similarity and proximity between the real-time scanned target and the specific target in the knowledge base, as well as the difference in the surface area between the feature sequences of the real-time scanned target and the feature sequences of the specific target in the knowledge base, improving the accuracy of target recognition. Brief Description of the Drawings

[0018] Figure 1 It is a flowchart of a method for fast recognition of point cloud of a laser push-broom imaging fuse based on two-dimensional GRA disclosed by the present invention.

[0019] Figure 2 It is a schematic diagram of the ground push-broom of the laser push-broom imaging fuse.

[0020] Figure 3 It is a schematic diagram of the method for filtering ground and abnormal noise points. Detailed Embodiment

[0021] The present invention will be further described below with reference to the drawings. The described embodiments are only for understanding the present invention, and are only a part of the embodiments of the present invention, rather than all the embodiments.

[0022] The present invention makes full use of the time interval between two frames of data obtained in real time by the laser push-broom imaging fuse. The time interval mainly performs two tasks: linear array three-dimensional point cloud imaging and two-dimensional linear array feature extraction. Conducting the target recognition task based on three-dimensional point cloud data can avoid the influence of illumination and scale changes on the recognition effect, and has a certain robustness to occlusion. On the other hand, technologies such as splicing are used to obtain the two-dimensional linear array features composed of distance information and intensity information in real time. Since the method of processing while scanning the linear array point cloud image is adopted, the imaging and feature extraction tasks can be carried out simultaneously, greatly shortening the feature sequence extraction time of the traditional method. After being deployed to an embedded system such as an FPGA, when the parallel data processing method is adopted, the online processing speed of the algorithm is greatly improved.

[0023] The simultaneous recognition algorithm adopts the GRA method, and the GRA has good effects in dealing with "small sample and poor information" data. Therefore, it can greatly reduce the scale of the knowledge base, reduce the number of parameters of the matching model, and lower the embedded hardware requirements of the fuse.

[0024] In addition, the present invention considers the similarity and proximity between the real-time scanned target and the specific target in the knowledge base, as well as the difference in the surface area between the feature sequences of the real-time scanned target and the feature sequences of the specific target in the knowledge base, and designs a two-dimensional correlation analysis model suitable for the target recognition task, which can improve the accuracy of target recognition.

[0025] Combined Figure 1 , the method for rapid point cloud recognition of a laser pushbroom imaging fuse based on two-dimensional GRA according to the present invention includes the following steps:

[0026] Step 1: Taking a moving object (including but not limited to vehicles, pedestrians, buildings) as a preset reference target, actually collecting through a laser linear array pushbroom imaging fuse or simulating the radar detection process using a laser point cloud imaging virtual simulation platform, and obtaining the current state target linear array distance information and intensity information by changing the intersection angle between the imaging recognition platform (such as an aircraft, a ground vehicle, a drone, etc.) and the target, and performing a minimum value operation on the above information, continuously obtaining the minimum value through missile-target intersection, and performing sorting and normalization processing to form a two-dimensional prior knowledge base of the target point cloud with labeled information, specifically as follows:

[0027] Establish an interactive model of a virtual line scanning system, a target and a scene, generate the current state target linear array distance information and intensity information using a ray tracing algorithm, perform a minimum value operation on the above information, and obtain the current minimum distance information h min and the maximum intensity I max , according to the movement trajectory of missile-target intersection, continuously obtaining h min and the maximum intensity I max at different times, and forming a two-dimensional sequence A as follows:

[0028]

[0029] where h min (n) is the minimum distance information at the nth moment; I max (n) is the maximum intensity information at the nth moment; the nth moment represents the maximum moment, and also represents a two-dimensional sequence composed of n distance information and intensity information.

[0030] Perform a sorting operation on A to obtain a matrix B as follows:

[0031]

[0032] Among them, H(1) represents the minimum value of the H sequence, H(2) represents the second smallest value of the H sequence, and H(n) represents the maximum value of the H sequence; Q(1) represents the minimum value of the Q sequence, Q(2) represents the second smallest value of the Q sequence, and Q(n) represents the maximum value of the Q sequence. H represents the sorted result sequence of the minimum distance information, and Q represents the sorted result sequence of the maximum intensity information.

[0033] The normalization operation obtains the two-dimensional sequence Z 0 As follows, where the element x 0 (1) = 1, the element x 0 (2) = H(2) / H(1), the element x 0 (n) = H(n) / H(1), the element y 0 (1) = 1, the element y 0 (2) = Q(2) / Q(1), the element y 0 (n) = Q(n) / Q(1):

[0034]

[0035] Through the method of the label index table, the obtained Z 0 is accurately labeled to obtain a two-dimensional sequence containing the target category label, and finally a two-dimensional prior knowledge base of the target point cloud with labeled information is obtained.

[0036] Step 2: The laser linear array imaging fuse performs push-broom detection on the target. The front lower part of the laser fuse is push-broomed linearly, and the target linear array distance information and intensity information are obtained in real time successively, as follows:

[0037] The laser push-broom imaging fuse uses an N-element array laser transceiver system. For example, N = 24, 32, 48, etc. The larger N is, the higher the imaging resolution. When the fuse irradiates the target scene in the front lower part once, N data points are obtained. The data points include the distance information and intensity information of the target. The fuse performs linear array push-broom detection on the target scene and, combined with the missile-target intersection trajectory, obtains W frames of data successively, where W ≤ n. The value of W is as follows:

[0038]

[0039] where L is the length of the scanning scene, v is the relative speed of the missile-target intersection, and T is the time for one laser transceiver operation. is the floor operation.

[0040] Step 3: Preprocess the target linear array distance information and intensity information obtained in real time, filter out the ground points and abnormal noise points, reduce the influence of interference points on the recognition result, and obtain clearer and more representative target point cloud data to improve the target recognition effect; specifically as follows:

[0041] Step S31: For the data of the W-th frame, take the 3 points with the largest distance values in the W-th frame array, and calculate the average value of these 3 points. Take the 3 points with the largest intensity values in the frame array, and calculate the average value of these 3 points.

[0042] Step S32: Determine the distance filtering threshold Th H and the intensity filtering threshold Th I sizes. For any predicted point j among the N data points, where j ≤ N, subtract the distance value and intensity of point j from and respectively, and calculate the absolute value ΔH j and ΔI j after subtraction. When (ΔH j < Th H ) & (ΔI j < Th I ) = 1, it is determined as a ground or abnormal noise point, and the predicted point j is filtered out to obtain the target point cloud data.

[0043] Step 4: Perform minimum value extraction, sorting, and data generation alignment operations on the target point cloud data obtained in Step 3 to obtain the aligned target linear array distance information and intensity information; specifically as follows:

[0044] Step S41: Perform minimum value extraction operation on the target point cloud data obtained in Step 3 to obtain the minimum distance information h 2min (i) and the maximum intensity I 2max (i) at the current moment i. According to the motion trajectory of the missile-target encounter, continuously obtain the minimum distance information and maximum intensity at different moments, and form a two-dimensional sequence A 2 as follows:

[0045]

[0046] where h 2min (1) is the minimum distance information at the 1st moment, h 2min (2) is the minimum distance information at the 2nd moment, h 2min (W) is the minimum distance information at the n-th moment, I 2max (1) is the maximum intensity information at the 1st moment, I 2max (2) is the maximum intensity information at the 2nd moment, and I 2max (W) is the maximum intensity information at the n-th moment.

[0047] Perform sorting operation on A 2 to obtain the matrix B 2 as follows:

[0048]

[0049] Among them, H 2 ′(1) represents the minimum value of the H′ 2 sequence, H 2 ′(2) represents the second smallest value of the H′ 2 sequence, H 2 ′(W) represents the maximum value of the H′ 2 sequence; Q 2 ′(1) represents the minimum value of the Q′ 2 sequence, Q 2 ′(2) represents the second smallest value of the Q′ 2 sequence, Q 2 ′(W) represents the maximum value of the Q′ 2 sequence, H′ 2 The sorted result sequence representing the minimum distance information, Q′ 2 represents the sequence output after sorting the maximum intensity information sequence.

[0050] Since each row of the two-dimensional sequence of the knowledge base has n elements, and after B 2 the element size W ≤ n, it is necessary to use a gray operator to generate missing data to align the feature sequences in the knowledge base; to improve the operation speed, the adjacent mean generation operator x(k)d is used, and the generation method is as follows:

[0051] x(k)d = 0.5(x(k) + x(k - 1))

[0052] Among them, x(k) represents the k-th element in H′ 2 or Q′ 2 where k > 1.

[0053] Obtain the two-dimensional sequence C with data complemented 2 :

[0054]

[0055] Among them, H 2 (1) represents the minimum value of the H 2 sequence, H 2 (2) represents the second smallest value of the H 2 sequence, H 2 (W) represents the maximum value of the H 2 sequence; Q 2 (1) represents the minimum value of the Q 2 sequence, Q 2 (2) represents the second smallest value of the Q 2 sequence, Q 2 (W) represents the maximum value of the Q 2 sequence, H 2 The sorted result sequence representing the minimum distance information, Q2 The sorted result sequence representing the maximum intensity information.

[0056] Step S42: Assume that the two-dimensional sequence obtained in real time is Z 2 , which is defined as follows. The two-dimensional feature sequence of the target in the prior knowledge base is Z 0 . Let Z 2 and Z 0 have equal lengths.

[0057]

[0058] Among them, x 2 (1) = 1, the element x 2 (2) = H 2 (2) / H 2 (1), the element x 2 (n) = H 2 (n) / H 2 (1), the element y 2 (1) = 1, the element y 2 (2) = Q 2 (2) / Q 2 (1), the element y 2 (n) = Q 2 (n) / Q 2 (1).

[0059] Step S43: Calculate the initial point zero-imagined feature sequences Z' 2 and Z' 0 of Z 2 and Z' 0 , and the calculation method is as follows:

[0060]

[0061] Step S43: Calculate the similarity distance |s 2 - s 0 | between Z' 2 and Z' 0 , and the calculation method is as follows:

[0062]

[0063] Among them and both represent the start point zero-imagined processing operation, which is convenient for subsequent data alignment, comparison and other operations.

[0064]

[0065] Step S44: Calculate the proximity distance |S 2 and Z 0 between2 -S 0 |, the calculation method is as follows;

[0066]

[0067] Step S45: Calculate the surface areas M 2 and M 0 of Z 2 and Z 0 , respectively. The calculation method is as follows:

[0068]

[0069] Step S46: Calculate the grey similarity correlation degree ε 2 between Z 0 and Z 20 , and the calculation method is as follows:

[0070]

[0071] Step S47: Calculate the grey proximity correlation degree ρ 2 between Z 0 and Z 20 , and the calculation method is as follows:

[0072]

[0073] Step S48: Calculate the grey area correlation degree 2 between Z 0 and Z , and the calculation method is as follows:

[0074]

[0075] Step S49: Classification matching model based on two-dimensional grey correlation analysis:

[0076]

[0077] Among them, α, γ, and λ correspond to the grey similarity correlation coefficient, grey proximity correlation coefficient, and grey area correlation coefficient respectively, and α + γ + λ = 1. Different values of α, γ, and λ can be set for different scenarios. For example, when the scenario is that the recognition platform flies over the target for recognition, α can be appropriately increased at this time; if the recognition platform conducts the recognition task during the process of gradually approaching the target, γ can be appropriately increased at this time; and when there is more noise in the data collected by the sensor, λ can be appropriately reduced to improve the robustness of the recognition algorithm to noise. ξ is the degree of similarity between Z i and Z 0 . The larger ξ is, the greater the probability that the target category detected by real-time scanning is a specific target.

[0078] Step 5: Design a two-dimensional grey relational analysis method, establish a classification and matching model based on two-dimensional grey relational analysis, and load the two-dimensional prior knowledge base of the target point cloud with annotation information and the classification and matching model into the MCU of the laser line array push-broom imaging fuse.

[0079] Step 6: Input the aligned target line array distance information and intensity information obtained in Step 4 into the classification and matching model obtained in Step 5 to achieve rapid target recognition; for different targets, the fuse controls different detonation points to form different warheads.

[0080] Embodiment 1

[0081] Combined with Figure 1 、 Figure 2 and Figure 3 For a method for rapid point cloud target recognition of a laser push-broom imaging fuse based on two-dimensional grey relational analysis according to the present invention, the steps are as follows:

[0082] Step 1: Taking moving objects (such as buses, cars, bicycles, and pedestrians, etc.) as preset reference targets, a prior knowledge base is jointly constructed by actual measurement and a laser point cloud imaging virtual simulation platform to make the knowledge base richer. By changing the intersection angle between the imaging and recognition platform (such as aircraft, ground vehicles, drones, etc.) and the target, the angle setting values are shown in Table 1, the target line array distance information and intensity information in the current state are obtained, and the minimum value operation is performed on the information. The minimum value is continuously obtained through missile-target intersection, and sorting and normalization processing are carried out to form a two-dimensional prior knowledge base of the target point cloud with annotation information.

[0083] Table 1 Intersection angle parameter settings

[0084]

[0085] Step 2: Combining the motion trajectory, the laser line array imaging fuse pushes and scans the target, performs line array push-broom on the front lower part of the laser fuse, and sequentially obtains the target line array distance information and intensity information. The schematic diagram of the laser push-broom imaging fuse's ground push-broom is as Figure 2 shown. When the fuse irradiates the target scene in the front lower part once, N data points (the data points include the distance information, intensity information, etc. of the target) can be obtained. The fuse performs line array push-broom detection on the target scene, and in combination with the missile-target intersection trajectory, W frames of data can be obtained successively.

[0086] Step 3: Preprocess the scanned point cloud, filter out the ground points and abnormal noise points, reduce the influence of interference points on the recognition result. The ground and abnormal noise point filtering method is as Figure 3 shown, and clearer and more representative target point cloud data is obtained.

[0087] Take the 3 points with the largest array distance value in one frame and calculate the average value of the 3 points Select the 3 points with the largest array intensity values, and calculate the average value of the 3 points Determine the distance filtering threshold Th h and the intensity filtering threshold Th i size, and subtract the distance value and intensity of the subsequent predicted point from and respectively, and take the absolute value to obtain ΔH j and ΔI j . When (ΔH j < Th H ) & (ΔI j < Th I ) = 1, it is determined as a ground or abnormal noise point, and the predicted point j is filtered to obtain the target point cloud data.

[0088] Step 4: Perform minimum value extraction, sorting, and data generation alignment operations on the target point cloud data to obtain the aligned target linear array distance information and intensity information.

[0089] In this embodiment, perform a minimum value extraction operation on the target point cloud data preprocessed in step 3 to obtain the minimum distance information h min and the maximum intensity I max at the current moment. According to the motion trajectory of the missile-target intersection, continuously obtain the sum and maximum intensity at different moments to form a two-dimensional sequence A 2 :

[0090]

[0091] where h 2min (i) and I 2max (i) respectively represent the minimum distance information and the maximum intensity information of the i-th frame.

[0092] Sort the minimum distance h 2min (i) and the maximum intensity I 2max (i) for each time frame to generate a sorted matrix B 2 :

[0093]

[0094] where H′ 2 and Q′ 2 respectively represent the sorted minimum distance information and maximum intensity information sequences.

[0095] Since the length W of the feature sequence obtained in real time is less than the length n of the sequence in the knowledge base, use the adjacent mean generation operator to complete the missing data pairs to align the feature sequences in the knowledge base, and finally obtain the completed two-dimensional sequence C 2 :

[0096]

[0097] Finally, perform a standardization operation so that the aligned data has the same scale and unit as the feature sequence of the prior knowledge base. For each dimension of data (minimum distance and maximum intensity), through x 2 (n) = H 2 (n) / H 2 (1) and y 2 (n) = Q 2 (n) / Q 2 (1) formula for standardization, so as to obtain the aligned two-dimensional feature sequence Z 2 :

[0098]

[0099] Step 5: Design a two-dimensional grey relational analysis method and establish a classification and matching model based on two-dimensional grey relational analysis. For the two-dimensional sequence Z 2 obtained in real time and the two-dimensional feature sequence Z 0 of the target in the prior knowledge base, calculate the grey similarity correlation degree ε 2 between Z 0 and Z 20 , the grey proximity correlation degree ρ 20 and the grey area correlation degree to construct a classification and matching model for two-dimensional grey relational analysis:

[0100]

[0101] where α + γ + λ = 1, and the values of α, γ, and λ can be selected according to the actual application scenario.

[0102] Load the two-dimensional prior knowledge base of the target point cloud with annotation information and the classification and matching model into the MCU of the laser line array push-broom imaging fuse for real-time target recognition.

[0103] Step 6: Input the aligned target line array distance information and intensity information obtained in Step 4 into the classification and matching model obtained in Step 5 to achieve fast target recognition; for different targets, the fuse controls different detonation points to form different warheads.

Claims

1. A method for rapid identification of laser push-scanning imaging fuze point cloud based on two-dimensional GRA, characterized in that: The steps include: Step 1: Taking the moving object as the preset reference target, the laser linear array push-scan imaging fuze is used for actual measurement and acquisition or the laser point cloud imaging virtual simulation platform is used to simulate the radar detection process. By changing the projectile-target intersection angle, the current state target linear array distance information and intensity information are obtained, and the above information is minimized. The minimum value is continuously obtained through projectile-target intersection, and sorting and standardization are performed to form a two-dimensional prior knowledge base of the target point cloud with labeled information; Step 2: Push-scanning the laser linear array imaging fuze to detect the target, and push-scanning the linear array in front of and below the laser fuze to obtain the target linear array distance information and intensity information in real time; Step 3: Preprocess the target linear array distance information and intensity information acquired in real time, filter out ground points and abnormal noise points, reduce the influence of interference points on the recognition results, and obtain clearer and more representative target point cloud data; Step 4: performing minimum value, sorting, and data generation alignment operations on the target point cloud data obtained in step 3 to obtain the aligned target line array distance information and intensity information; Step 5: Design a two-dimensional grey correlation analysis method, establish a classification matching model based on the two-dimensional grey correlation analysis, and load the two-dimensional prior knowledge base of the target point cloud with annotation information and the classification matching model into the MCU of the laser linear array push-scan imaging fuze; Step 6: Input the aligned target array distance information and strength information obtained in step 4 into the classification matching model obtained in step 5 to achieve rapid target identification; for different targets, the fuze controls the detonation of different detonation points to form different warheads.

2. The method for rapid identification of fuze point cloud based on two-dimensional GRA by laser push-scanning imaging according to claim 1 is characterized in that: The moving objects in step 1 include but are not limited to vehicles and pedestrians.

3. The method for rapid identification of fuze point cloud based on two-dimensional GRA according to claim 2 is characterized in that: In step 1, the moving object is used as the preset reference target, and the radar detection process is simulated by actual measurement and acquisition of the laser linear array push-scan imaging fuze or by using the laser point cloud imaging virtual simulation platform. By changing the projectile-target intersection angle, the current state target linear array distance information and intensity information are obtained, and the above information is minimized. The minimum value is continuously obtained through projectile-target intersection, and sorting and standardization are performed to form a two-dimensional prior knowledge base of the target point cloud with labeled information, as follows: The interactive model of the virtual line scanning system, target and scene is established, and the ray tracing algorithm is used to generate the current state target line array distance information and intensity information. The above information is minimized to obtain the current minimum distance information h min and maximum intensity I max , according to the trajectory of the projectile-target intersection, h at different times is continuously obtained min and maximum intensity I max , forming a two-dimensional sequence A as follows: Among them, h min (n) is the minimum distance information at time n; I max (n) is the maximum intensity information at time n; time n represents the maximum moment, and also represents a two-dimensional sequence consisting of n distance information and intensity information; Sorting A, we get the matrix B as follows: Wherein, H(1) represents the minimum value of the H sequence, H(2) represents the second minimum value of the H sequence, and H(n) represents the maximum value of the H sequence; Q(1) represents the minimum value of the Q sequence, Q(2) represents the second minimum value of the Q sequence, and Q(n) represents the maximum value of the Q sequence. H represents the sorting result sequence of the minimum distance information, and Q represents the sorting result sequence of the maximum strength information. The normalization operation yields the following two-dimensional sequence Z0, where element x0(1) = 1, element x0(2) = H(2) / H(1), element x0(n) = H(n) / H(1), element y0(1) = 1, element y0(2) = Q(2) / Q(1), and element y0(n) = Q(n) / Q(1): Through the label index table method, the obtained Z0 is accurately labeled to obtain a two-dimensional sequence containing the target category label, and finally a two-dimensional prior knowledge base of the target point cloud with labeled information is obtained.

4. The method for rapid identification of fuze point cloud based on two-dimensional GRA according to claim 3 is characterized by: In step 2, the laser linear array imaging fuze push-scans the target, and performs linear array push-scans in front of and below the laser fuze to obtain the target linear array distance information and intensity information in real time, as follows: The laser push-scan imaging fuze uses an N-element array laser transceiver system, N = 24, 32, 48...; the fuze irradiates the target scene below once, and obtains N data points, which contain the distance information and intensity information of the target; the fuze performs linear push-scan detection on the target scene, and combines the trajectory of the missile intersection to obtain W frames of data one by one, W≤n, and the value of W is as follows: Where L is the length of the scanned scene, v is the relative speed of the missile-target intersection, and T is the time it takes for the laser to be sent and received once. This is a floor operation.

5. The method for rapid identification of fuze point cloud based on two-dimensional GRA according to claim 4 is characterized in that: In step 3, the target linear array distance information and intensity information acquired in real time are preprocessed to filter out ground points and abnormal noise points, reduce the influence of interference points on the recognition results, and obtain clearer and more representative target point cloud data, as follows: Step S31: For the W-frame data, take the three points with the largest array distance values ​​in the W-th frame and calculate the average value of these three points. Take the three points with the largest intensity values ​​in the frame array and calculate the average value of these three points Step S32: Determine the distance filtering threshold Th H and intensity filtering threshold Th I For any predicted point j among N data points, j≤N, the distance value and intensity of point j are respectively and Subtract, and find the absolute value after subtraction ΔH j and ΔI j , when (ΔH j <Th H )&(ΔI j <Th I )=1, it is identified as a ground or abnormal noise point, and the predicted point j is filtered out to obtain the target point cloud data.

6. The method for rapid identification of fuze point cloud based on two-dimensional GRA according to claim 5 is characterized in that: In step 4, the target point cloud data obtained in step 3 is subjected to minimum value taking, sorting, and data generation alignment operations to obtain the aligned target linear array distance information and intensity information, as follows: Step S41: perform a minimum value operation on the target point cloud data obtained in step 3 to obtain the minimum distance information h at the current time i. 2min (i) and maximum intensity I 2max (i) According to the trajectory of the missile-target intersection, the minimum distance information and maximum intensity at different times are continuously obtained to form a two-dimensional sequence A2 as follows: Among them, h 2min (1) is the minimum distance information at the first moment, h 2min (2) is the minimum distance information at the second moment, h 2min (W) is the minimum distance information at time n, I 2max (1) is the maximum intensity information I at the first moment 2max (2) is the maximum intensity information at the second moment, I 2max (W) is the maximum intensity information at time n; Sorting A2, we get the matrix B2 as follows: Wherein, H2′(1) represents the minimum value of the H′2 sequence, H2′(2) represents the second minimum value of the H′2 sequence, and H2′(W) represents the maximum value of the H′2 sequence; Q2′(1) represents the minimum value of the Q′2 sequence, Q2′(2) represents the second minimum value of the Q′2 sequence, and Q2′(W) represents the maximum value of the Q′2 sequence. H′2 represents the sorting result sequence of the minimum distance information, and Q′2 represents the sorting result sequence of the maximum intensity information. Since each row of the two-dimensional sequence of the knowledge base has n elements, and the size of the B2 element W≤n, it is necessary to use a gray operator to generate vacant data to align the feature sequence in the knowledge base; to improve the operation speed, the adjacent mean generation operator x(k)d is used, and the generation method is as follows: x(k)d=0.5(x(k)+x(k-1)) Where x(k) represents the kth element in H′2 or Q′2, k>1; Get the two-dimensional sequence C2 with data completion: Among them, H2(1) represents the minimum value of the H2 sequence, H2(2) represents the second smallest value of the H2 sequence, and H2(W) represents the maximum value of the H2 sequence; Q2(1) represents the minimum value of the Q2 sequence, Q2(2) represents the second smallest value of the Q2 sequence, and Q2(W) represents the maximum value of the Q2 sequence. H2 represents the sorting result sequence of the minimum distance information, and Q2 represents the sorting result sequence of the maximum intensity information. Step S42, assuming that the two-dimensional sequence obtained in real time is Z2, which is defined as follows, the two-dimensional feature sequence of the target in the prior knowledge base is Z0, and the lengths of Z2 and Z0 are equal; Where x2(1)=1, element x2(2)=H2(2) / H2(1), element x2(n)=H2(n) / H2(1), element y2(1)=1, element y2(2)=Q2(2) / Q2(1), element y2(n)=Q2(n) / Q2(1); Step S43, calculate the characteristic sequences Z′2 and Z′0 of the zero-image of the initial points of Z2 and Z0, and the calculation method is as follows: Step S43, calculate the similarity distance |s2-s0| between Z′2 and Z′0, the calculation method is as follows: in and Both represent the zero-image processing operation of the starting point; Step S44, calculating the approach distance |S2-S0| between Z2 and Z0, the calculation method is as follows; Step S45, respectively calculate the surface areas M2 and M0 of Z2 and Z0, the calculation method is as follows: Step S46: Calculate the grey similarity correlation degree ε between Z2 and Z0 20 , the calculation method is as follows: Step S47, calculate the gray proximity correlation degree ρ between Z2 and Z0 20 , the calculation method is as follows: Step S48: Calculate the gray area correlation between Z2 and Z0 The calculation method is as follows: Step S49: Classification matching model based on two-dimensional grey relational analysis: Among them, α, γ, and λ correspond to the gray similarity correlation coefficient, gray proximity correlation coefficient, and gray area correlation coefficient, respectively. α+γ+λ=1, ξ is Z i The degree of similarity between ξ and Z0, the larger the ξ, the greater the probability that the target category detected by real-time scanning is a specific target.

Images