Target identification method based on polarization laser in smoke environment

Abstract

The invention discloses a target identification method based on polarization laser in a smoke environment, aims to solve the target identification problem under battlefield smoke interference, and is characterized in that a target identification model based on polarization laser in the smoke environment is established and comprises a Stokes vector simulation module and a Stokes vector identification module, the Stokes simulation module can be used for simulating to obtain a large number of simulation Stokes vectors in the smoke environment with / without targets, and the simulation Stokes vectors are used for training the Stokes vector recognition module, so that the trained Stokes vector recognition module is obtained, and the trained Stokes vector recognition module can be used for recognizing whether the targets exist in the smoke environment or not. The target identification method provided by the invention can judge whether the interference is smoke interference according to the echo polarization degree and the echo stokes vector by using the polarized laser, and can effectively improve the precision and stability of laser transmission and target identification in the smoke environment.

Description

Technical Field

[0001] The present invention relates to the field of polarized laser detection, and particularly to a method for target recognition based on polarized laser in a soot environment. Background Technique

[0002] With the rapid development of laser technology and laser devices, laser detection is increasingly widely used in military and civilian fields, especially performing excellently in military technology. Due to its high coherence, accurate ranging ability, and fast response speed, laser detection has significant advantages in the field of target detection. Pulse laser has become an effective supplement to radio detection due to its high instantaneous power, good directivity, strong anti-interference ability, high control accuracy, and small ranging error, and is widely used in various fields.

[0003] However, laser detection is easily affected by aerosols, resulting in strong backscattering, leading to misidentification or non-identification. As the most common passive interference, soot will have a strong scattering effect on the laser during laser detection, causing the laser receiving system to receive backscattered echo signals similar to the target echo, greatly reducing the correct target recognition rate.

[0004] Many scholars have carried out research on the principle of the influence of soot on laser detection and proposed various laser detection methods for anti-soot interference. The existing laser anti-soot interference methods are mainly based on LIDAR point cloud data processing, laser ranging principle, polarization characteristics, etc. Although methods based on point cloud or laser ranging can achieve the anti-soot effect to a certain extent, the effect is not good enough, and the algorithm of the LIDAR-based point cloud data processing model constructed by them is too complex and not conducive to hardware implementation. And through the difference in the polarization characteristics of the laser, although anti-smoke interference is achieved to a certain extent, the research on the existing polarization-based anti-soot interference methods is not deep enough, and the constructed models mostly identify from a single dimension of polarization degree and do not make good use of polarization characteristics. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for target recognition based on polarized laser in a soot environment, which can simulate the echo polarization characteristics of various polarized lasers with and without targets in different soot environments, and can analyze the polarization information of the measured echo according to the difference in polarization characteristics, so as to judge whether there is a target in the measured echo.

[0006] The technical solution for achieving the purpose of the present invention is: a method for target recognition based on a polarized laser echo classification model in a soot environment, the steps are as follows:

[0007] Step 1: Measure the soot particle parameters in the soot environment. The soot particle parameters include shape, size, and refractive index. A polarized laser emitter emits a beam of polarized laser towards a receiver to obtain the original Stokes vector of the polarized laser. Then, the polarized laser emitter emits a beam of polarized laser into the soot environment, which is received by the receiver to obtain the Stokes vector of the polarized laser echo to be recognized in the soot environment, and then proceed to Step 2.

[0008] Step 2: Construct a target recognition model based on polarized laser in the soot environment:

[0009] The target recognition model based on polarized laser includes a Stokes vector simulation module and an echo Stokes vector recognition module. The Stokes vector simulation module can obtain the simulated Stokes vectors with and without targets in the soot environment according to the input soot particle parameters and the original Stokes vector of the polarized laser. The echo Stokes vector recognition module can recognize the Stokes vector of the polarized laser echo to be recognized in the soot environment, and then proceed to Step 3.

[0010] Step 3: Input the measured soot particle parameters and the original Stokes vector of the polarized laser into the target recognition model based on polarized laser to obtain the trained target recognition model based on polarized laser in the soot environment:

[0011] Step 3-1: Input the soot particle parameters and the original Stokes vector of the polarized laser into the Stokes vector simulation module to obtain the simulated Stokes vectors with and without targets in the soot environment:

[0012] Step 3-1-1: Divide the polarized laser into photon packets containing T photons, initialize the photon positions and photon movement directions, and set the initial Stokes vector of each photon to the original Stokes vector of the measured polarized laser used, and select a photon, then proceed to Step 3-1-2.

[0013] Step 3-1-2: Analyze the random walk step size of the selected photon and update the photon position, then proceed to Step 3-1-3.

[0014] Step 3-1-3: According to the photon position, judge the photon collision situation; if it is determined that a collision occurs with a soot particle, update the Stokes vector of photon soot scattering, and then proceed to Step 3-1-4; if it is determined that a collision occurs with a target, update the Stokes vector of photon target reflection, and then proceed to Step 3-1-4.

[0015] Step 3-1-4: Update the direction of the collided photons, and determine whether the photons are extinct according to the optimized semi-analytical method; if it is determined that the photons are not extinct, go to Step 3-1-5; if it is determined that the photons are extinct, select the next photon and return to Step 3-1-2.

[0016] Step 3-1-5: Determine whether the photons enter the receiving field of view; if the photons enter the receiving field of view, go to Step 3-1-6; if the photons do not enter the receiving field of view, select the next photon and return to Step 3-1-2.

[0017] Step 3-1-6: Statistically analyze the Stokes vectors of the photons, and determine whether it is the last photon; if so, obtain the simulated Stokes vectors in the soot environment with and without targets according to the Stokes vectors of all photons, and go to Step 3-2; if not, select the next photon and return to Step 3-1-2.

[0018] Step 3-2: Use the simulated Stokes vectors in the soot environment with and without targets to train the echo Stokes vector recognition module, obtain the trained echo Stokes vector recognition module, and further obtain the trained target recognition model based on polarized laser in the soot environment:

[0019] Step 3-2-1: Normalize the simulated Stokes vectors in the soot environment with and without targets to obtain the normalized simulated Stokes vectors in the soot environment with and without targets, denoted as the sample set D, and go to Step 3-2-2.

[0020] Step 3-2-2: Process the sample set D using the optimized CART algorithm to determine each optimal classification feature of the decision tree, and then construct a single decision tree, and go to Step 3-2-3.

[0021] Step 3-2-3: If the number of generated decision trees is equal to 100, obtain the trained target recognition model based on polarized laser in the soot environment composed of all the generated decision trees, and go to Step 4; otherwise, return to Step 3-2-2.

[0022] Step 4: Input the Stokes vector of the polarized laser echo to be recognized in the soot environment into the trained target recognition model based on polarized laser in the soot environment, and the result of whether there is a target is obtained by the joint voting of all the decision trees.

[0023] Compared with the prior art, the remarkable advantage of the present invention is that:

[0024] (1) The present invention uses the Stokes vector, a multi-dimensional tool, to describe the polarization state of light. Compared with the single-dimensional degree of polarization, the Stokes vector provides more comprehensive and detailed polarization information through its four components I, Q, U, and V, making full use of multi-dimensional features.

[0025] (2) Different algorithms for the transmission of the Stokes vector by colliding soot particles and colliding targets are introduced into the traditional Monte Carlo simulation. These algorithms consider the physical properties of soot particles and colliding targets such as (shape, size, refractive index) and their interaction mechanisms with the laser (such as absorption, scattering, refraction, reflection), and establish a soot scattering module, significantly improving the authenticity of the simulation process.

[0026] (3) The Stokes vector simulation module has significant advantages in saving the time and financial costs required for actual data acquisition, providing strong support for the research and development of related applications.

[0027] (4) An improved random forest model is introduced when processing simulation and actual data, greatly reducing the influence of subjective factors, being able to efficiently process a large amount of Stokes vector data, and improving the recognition accuracy of the target. Description of the Drawings

[0028] Figure 1 is the overall flowchart.

[0029] Figure 2 is the flowchart of the Stokes vector simulation of the laser echo.

[0030] Figure 3 is the target recognition model based on polarized laser in a soot environment.

[0031] Figure 4 is the schematic diagram of actual polarized laser detection. Detailed Embodiment

[0032] The present invention will be further elaborated in detail with reference to the accompanying drawings.

[0033] Combined with Figures 1 to 4 , a method for target recognition based on polarized laser in a soot environment according to the present invention is as follows:

[0034] Step 1: Measure the soot particle parameters in the soot environment. The soot particle parameters include shape, size, and refractive index. A polarized laser emitter emits a beam of polarized laser towards a receiver to obtain the original Stokes vector of the polarized laser. Then, the polarized laser emitter emits a beam of polarized laser into the soot environment, which is received by the receiver to obtain the Stokes vector of the polarized laser echo to be recognized in the soot environment, and then proceed to Step 2.

[0035] Step 2: Construct a target recognition model based on polarized laser in a soot environment:

[0036] The target recognition model based on polarized laser includes a Stokes vector simulation module and an echo Stokes vector recognition module. The Stokes vector simulation module can obtain the simulated Stokes vectors in the soot environment with and without targets based on the input soot particle parameters and the original Stokes vector of the polarized laser. The echo Stokes vector recognition module can recognize the Stokes vector of the polarized laser echo to be recognized in the soot environment, and then proceed to Step 3.

[0037] Step 3: Input the measured soot particle parameters and the original Stokes vector of the polarized laser into the target recognition model based on polarized laser to obtain a trained target recognition model based on polarized laser in a soot environment:

[0038] Step 3-1: Input the soot particle parameters and the original Stokes vector of the polarized laser into the Stokes vector simulation module to obtain the simulated Stokes vectors in the soot environment with and without targets:

[0039] Step 3-1-1: Divide the polarized laser into photon packets containing T photons, initialize the photon positions and photon movement directions, set the initial Stokes vector of each photon to the original Stokes vector of the measured polarized laser, and select a photon, then proceed to Step 3-1-2.

[0040] Step 3-1-2: Analyze the random walk step size of the selected photon, update the photon position, and then proceed to Step 3-1-3.

[0041] Step 3-1-3: According to the photon position, judge the photon collision situation. If it is determined that the photon collides with a soot particle, then update the Stokes vector of photon soot scattering, and the update of the Stokes vector of photon soot scattering is expressed as:

[0042] S out =[C(x,n,θ)·L(-γ)·M(θ)·L(φ)]·S in

[0043] where, S out represents the Stokes vector of the photon after collision, S inrepresents the photon Stokes vector before collision, C(x, n, θ) represents the scattering correction matrix, L represents the rotation scattering matrix, M(θ) is the Mueller matrix, θ is the scattering angle, x represents the size parameter of the soot particle, φ is the angle that converts the Stokes vector of the incident light from the reference plane to the scattering plane before scattering, γ is the rotation angle that rotates back to the reference plane after scattering, and its cosine is:

[0044]

[0045] where, u zN represents the displacement of the photon in the z direction during the Nth scattering, θ N represents the scattering angle of the photon during the Nth scattering, and N represents the number of photon scatterings.

[0046] The scattering correction matrix C(x, n, θ) can be expressed as:

[0047]

[0048] where, x represents the size parameter of the soot particle 2πr / λ, r represents the radius of the soot particle, and λ represents the wavelength of the incident light.

[0049] f is expressed as the Mie scattering correction factor, which is expressed as:

[0050]

[0051] where, a n and b n represent the Mie scattering coefficients, and n represents the corresponding scattering angle serial number.

[0052] L is also called the coordinate transformation matrix, which is used to transform the Stokes vector between different reference frames and can be expressed as:

[0053]

[0054] After the Stokes vector is updated, it proceeds to step 3-1-4.

[0055] If it is determined that the photon collides with the target, then the Stokes vector update of the photon target reflection is performed. When it is determined that the photon collides with the target, the Stokes vector update of the photon target reflection is expressed as:

[0056] S out = M · S in · f r (θ i , φ i , θ r , φ r ) · K(θ r , φ r )

[0057] Among them, S out represents the Stokes vector of the photon after collision, and S in represents the Stokes vector of the photon before collision. M represents the Mueller matrix of the target material, and f r (θ i , φ i , θ r , φ r ) represents the bidirectional reflectance distribution function (BRDF) of the target. θ i and φ i represent the zenith angle and azimuth angle of the incident light, θ r represents the zenith angle of the reflected light, and φ r represents the azimuth angle of the reflected light. K(θ r , φ r ) represents the correction coefficient of the diffuse reflection Stokes vector, which can be expressed as:

[0058] K(θ r , φ r ) = K R (θ r )·K P (φ r )·K S (θ r )

[0059] 1) Reflectance correction coefficient K R (θ r ), which is expressed as:

[0060] K R (θ r ) = R(θ r )

[0061] Among them, R(θ r ) represents the reflectance of the target material.

[0062] 2) Polarization state correction coefficient K P (φ r ), which is expressed as:

[0063] K P (φ r ) = P·cos(2φ r )

[0064] Among them, P represents the degree of polarization of the incident light.

[0065] 3) Scattering distribution correction coefficient K S (θ r ), which is expressed as:

[0066]

[0067] Among them, n s represents the exponential parameter of the scattering direction, which determines the concentration degree of the direction distribution.

[0068] After the Stokes vector is updated, go to step 3-1-4.

[0069] Step 3-1-4: Update the direction of the collided photons, and determine whether the photons disappear according to the optimized semi-analytical method. The optimized semi-analytical method is expressed as:

[0070]

[0071] Among them, N s represents the cumulative scattering times of the photons, N s,max represents the set maximum photon scattering times, p abs represents the absorption probability when the photons collide, p thresh represents the absorption probability threshold when the photons collide, θ represents the scattering angle, θ max represents the maximum scattering angle of the photons, θ min represents the minimum scattering angle of the photons, W represents the energy weight of the photons, W thresh represents the lowest energy threshold of the photons. When any of the above conditions is met, it is considered that the photons disappear.

[0072] If it is determined that the photons do not disappear, go to step 3-1-5; if it is determined that the photons disappear, select the next photon and return to step 3-1-2.

[0073] Step 3-1-5: Determine whether the photons enter the receiving field of view: If the photons enter the receiving field of view, go to step 3-1-6; if the photons do not enter the receiving field of view, return to step 3-1-2.

[0074] Step 3-1-6: Statistically analyze the Stokes vectors of the photons and determine whether it is the last photon. If so, obtain the simulated Stokes vectors with and without targets in the soot environment according to the Stokes vectors of all photons, and go to step 3-2; if not, select the next photon and return to step 3-1-2.

[0075] Step 3-2: Use the simulated Stokes vectors with and without targets in the soot environment to train the echo Stokes vector recognition module, obtain the trained echo Stokes vector recognition module, and further obtain the trained target recognition model based on polarized laser in the soot environment:

[0076] Step 3-2-1: Normalize the simulated Stokes vectors with and without the target in the soot environment to obtain the normalized simulated Stokes vectors with and without the target in the soot environment, denoted as the sample set D, and then proceed to Step 3-2-2.

[0077] Step 3-2-2: Process the sample set D using the optimized CART algorithm, which uses the optimized smoothed Gini index as the judgment criterion:

[0078]

[0079] where D represents the sample set, m represents the sample label serial number, C L represents the number of categories; Q represents the total number of samples in the sample set D, and Q m represents the number of samples with the m-th sample label in the sample set, w m represents the category weight, α represents the dynamic smoothing factor, k represents the regularization strength, and R represents the regularization penalty term.

[0080] The category weight w m can be expressed as:

[0081]

[0082] where Δ represents the minimization term.

[0083] The dynamic smoothing factor α can be expressed as:

[0084]

[0085] where β is the smoothing strength.

[0086] The regularization penalty term R can be expressed as:

[0087]

[0088] where Gini(D j ) represents the Gini index of the split sample subset, and D j represents the j-th split sample subset.

[0089] Use the above optimized CART algorithm to determine each best classification feature of the decision tree, and then construct a single decision tree, and proceed to Step 3-2-3.

[0090] Step 3-2-3: If the number of generated decision trees is equal to 100, obtain the trained target recognition model based on polarized laser in the soot environment composed of all the generated decision trees, and proceed to Step 4; otherwise, return to Step 3-2-2.

[0091] Step 4: Input the Stokes vector of the polarization laser echo to be recognized in the soot environment into the trained target recognition model based on polarization laser in the soot environment, and the result of whether there is a target is obtained by the joint voting of all decision trees.

Claims

1. A target recognition method based on polarized laser in a smoke environment, characterized in that: Here are the steps: Step 1, measuring the parameters of the smoke particles in the smoke environment, the parameters of the smoke particles include shape, size, and refractive index, a polarized laser transmitter transmits a beam of polarized laser to a receiver, and obtains the original Stokes vector of the polarized laser, and then a polarized laser transmitter transmits a beam of polarized laser to the smoke environment, and the polarized laser is received by the receiver to obtain the Stokes vector of the polarized laser echo to be identified in the smoke environment, and then proceeds to step 2; Step 2: Construct a target recognition model based on polarized laser in smoke environment: The target recognition model based on polarized laser includes a Stokes vector simulation module and an echo Stokes vector recognition module. The Stokes vector simulation module can obtain the simulated Stokes vector with and without the target in the smoke environment according to the input smoke particle parameters and the polarized laser original Stokes vector. The echo Stokes vector recognition module can recognize the polarized laser echo Stokes vector to be recognized in the smoke environment, and then proceed to step 3. Step 3: Input the measured smoke particle parameters and the original Stokes vector of the polarized laser into the target recognition model based on polarized laser to obtain the trained target recognition model based on polarized laser in the smoke environment: Step 3-1, input the smoke particle parameters and the original Stokes vector of the polarized laser into the Stokes vector simulation module to obtain the simulated Stokes vector with and without the target in the smoke environment: Step 3-1-1: Divide the polarized laser into photon packets to obtain photon packets containing T photons, initialize the photon position and photon motion direction, set the initial Stokes vector of each photon to the original Stokes vector of the polarized laser used obtained by actual measurement, select a photon, and proceed to step 3-1-2; Step 3-1-2: Perform random walk step length analysis on the selected photons, update the positions of the photons, and proceed to step 3-1-3; Step 3-1-3: According to the position of the photon, determine the photon collision situation; if it is determined that the photon collides with the smoke particles, the photon smoke scattering Stokes vector is updated, and the process goes to step 3-1-4; if it is determined that the photon collides with the target, the photon target reflection Stokes vector is updated, and the process goes to step 3-1-4; Step 3-1-4: Update the direction of the colliding photon and determine whether the photon is extinct based on the optimized semi-analytical method; if the photon is determined not to be extinct, proceed to step 3-1-5; if the photon is determined to be extinct, select the next photon and return to step 3-1-2; Step 3-1-5: Determine whether the photon enters the receiving field of view; if the photon enters the receiving field of view, go to step 3-1-6; if the photon does not enter the receiving field of view, select the next photon and return to step 3-1-2; Step 3-1-6: Count the Stokes vectors of the photons to determine whether it is the last photon; if so, obtain the simulated Stokes vectors with and without targets in the smoke environment based on the Stokes vectors of all photons, and proceed to step 3-2; if not, select the next photon and return to step 3-1-2; Step 3-2: Use the simulated Stokes vectors in the smoke environment with and without targets to train the echo Stokes vector recognition module to obtain a trained echo Stokes vector recognition module, and then obtain a trained target recognition model based on polarized laser in the smoke environment: Step 3-2-1: normalize the simulated Stokes vectors with and without targets in the smoke environment to obtain the normalized simulated Stokes vectors with and without targets in the smoke environment, record them as sample set D, and proceed to step 3-2-2; Step 3-2-2: Use the optimized CART algorithm to process the sample set D to determine each optimal classification feature of the decision tree, and then construct a single decision tree and proceed to step 3-2-3; Step 3-2-3: If the number of decision trees generated is equal to 100, a trained target recognition model based on polarized laser in a smoke environment composed of all generated decision trees is obtained, and the process goes to step 4; otherwise, the process returns to step 3-2-2; Step 4: Input the Stokes vector of the polarized laser echo to be identified in the smoke environment into the trained target recognition model based on polarized laser in the smoke environment, and all decision trees jointly vote to obtain the result of whether the target exists.

2. The target recognition method based on polarized laser in smoke and dust environment according to claim 1 is characterized in that: In step 3-1-3, if it is determined that the photon collides with the smoke particle, the photon smoke scattering Stokes vector is updated as follows: When it is determined that a photon collides with a smoke particle, the photon smoke scattering Stokes vector is updated as: S out =[C(x,n,θ)·L(-γ)·M(θ)·L(φ)]·S in Among them, S out represents the Stokes vector of the photon after the collision, S in represents the Stokes vector of the photon before the collision, C(x,n,θ) represents the scattering correction matrix, n represents the corresponding scattering angle number, θ is the scattering angle, x represents the size parameter of the smoke particle, L represents the rotation scattering matrix, M(θ) is the Mueller matrix, φ is the angle of converting the Stokes vector of the incident light from the reference plane to the scattering plane before scattering occurs, γ is the rotation angle back to the reference plane after scattering, and its cosine is: Among them, u zN represents the displacement of the photon in the z direction when it is scattered for the Nth time, θ N It represents the scattering angle of the photon when it is scattered for the Nth time, where N represents the number of photon scattering times; The rotation scattering matrix L is also called the coordinate transfer matrix, which is used to transform the Stokes vector between different reference frames and is expressed as: 。 3. The target recognition method based on polarized laser in smoke environment according to claim 2 is characterized in that: The scatter correction matrix C(x,n,θ) is as follows: Where x represents the size parameter of the smoke particle 2πr / λ, r represents the radius of the smoke particle, and λ represents the wavelength of the incident light; f(x,n) is the Mie scattering correction factor, expressed as: Among them, a n With b n represents the Mie scattering coefficient, and n represents the corresponding scattering angle number.

4. The target recognition method based on polarized laser in smoke and dust environment according to claim 3 is characterized in that: In step 3-1-3, if it is determined that a collision occurs with the target, the photon target reflection Stokes vector is updated as follows: When it is determined that the photon collides with the target, the Stokes vector update of the photon target reflection is expressed as: S out =M·S in ·f r (i i ,f i ,i r ,f r )·K(θ r ,f r ) Among them, S out represents the Stokes vector of the photon after the collision, S in represents the Stokes vector of the photon before collision, M represents the Mueller matrix of the target material, and f r (θ i ,φ i ,θ r ,φ r ) represents the target bidirectional reflectance distribution function (BRDF), θ i and φ i represents the zenith angle and azimuth angle of the incident light, θ r represents the zenith angle of reflected light, φ r represents the azimuth of the reflected light, K(θ r ,φ r ) represents the diffuse reflection Stokes vector correction coefficient.

5. The target recognition method based on polarized laser in smoke and dust environment according to claim 4 is characterized in that: Diffuse reflection Stokes vector correction coefficient K(θ r ,φ r ), as follows: Diffuse reflection Stokes vector correction coefficient K(θ r ,φ r ) includes reflectivity correction factor, polarization state correction factor and scattering distribution correction factor; 1) Reflectivity correction factor K R (θ r ), expressed as: K R (i r )=R(θ r ) Among them, R(θ r ) represents the reflectivity of the target material; 2) Polarization correction factor K P (φ r ), expressed as: K P (f r )=P·cos(2φ r ) Where P represents the polarization degree of the incident light; 3) Scattering distribution correction coefficient K S (θ r ), expressed as: Among them, n s The exponential parameter representing the scattering direction determines the concentration of the directional distribution; Then the diffuse reflection Stokes vector correction coefficient K(θ r ,φ r ) is expressed as: K(θ r ,f r )=K R (i r )·K P (f r )·K S (i r )。 6. The target recognition method based on polarized laser in smoke and dust environment according to claim 1 is characterized in that: In step 3-1-4, the direction of the colliding photons is updated, and the photons are judged to be dead based on the optimized semi-analytical method, as follows: The optimized semi-analytical method is expressed as: Among them, N s represents the cumulative number of photon scattering, N s,max Indicates setting the maximum number of photon scattering times, p abs represents the absorption probability when a photon collides, p thresh represents the absorption probability threshold of photon collision, θ represents the scattering angle, and θ max represents the maximum scattering angle of photons, θ min represents the minimum scattering angle of photons, W represents the energy weight of photons, and W thresh Represents the minimum energy threshold of photons. When any of the above conditions is met, the photon is considered to be extinguished.

7. The target recognition method based on polarized laser in smoke environment according to claim 1, characterized in that: In step 3-2-2, the sample set D is processed using the optimized CART algorithm to determine each optimal classification feature of the decision tree, and then a single decision tree is constructed, as follows: The optimized CART algorithm is used to construct the random forest decision tree. In order to select the best classification features, the optimized CART algorithm uses the optimized smoothed Gini index as the judgment criterion: Among them, D represents the sample set, m represents the sample label number, and C L represents the number of categories; Q represents the total number of samples in the sample set D, Q m represents the number of samples with the mth sample label in the sample set, w m represents the category weight, α represents the dynamic smoothing factor, k represents the regularization strength, and R represents the regularization penalty term.

8. The target recognition method based on polarized laser in smoke and dust environment according to claim 7 is characterized in that: Generation category weight w m , dynamic smoothing factor α, regularization penalty term R, as follows: Class weight w m It is expressed as: Among them, Δ represents the minimization term; The dynamic smoothing factor α is expressed as: Among them, β is the smoothing strength; The regular penalty term R is expressed as: Among them, Gini (D j ) represents the Gini index of the sample subset after splitting, D j Represents the sample subset after the jth split.

Images