Flapping wing target detection and identification method based on combination of laser micro-motion and polarization-infrared characteristics
By combining laser micro-motion with polarization-infrared features and machine learning algorithms, the problem of bird and flapping-wing UAV identification was solved, achieving efficient and accurate target recognition.
Patent Information
- Application Number
- CN202511048991.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-29
- Publication Date
- 2025-11-07
AI Technical Summary
Existing technologies struggle to effectively distinguish between birds and flapping-wing drones. Traditional millimeter-wave radar detection methods lack sufficient resolution, and single-mode analytical frameworks neglect polarization scattering and optical characteristics, making them susceptible to environmental interference and resulting in identification difficulties.
A method combining laser micro-motion and polarization-infrared features, along with machine learning algorithms, is employed to select multimodal features and assign weights to construct a dataset for target recognition.
It significantly improves the recognition accuracy of birds and flapping-wing UAVs. The multimodal feature recognition method improves the accuracy by more than 30% compared with the single micro-motion modal recognition method.
Smart Images
Figure CN120912866A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to a flapping target detection and recognition method based on multi-modal feature fusion, in particular to a flapping target detection and recognition method based on laser micro-motion and polarization-infrared feature combination, and belongs to the field of laser detection and recognition. BACKGROUND
[0002] With the development of unmanned aerial vehicle detection and recognition, inspired by the excellent flight skills of insects and birds in nature, the bionic performance of flapping unmanned aerial vehicles has been greatly improved. Flapping unmanned aerial vehicles have the advantages of strong concealment, strong maneuverability, strong adaptability, low noise, etc. At the same time, it also brings greater challenges to social public safety and military defense.
[0003] The flapping unmanned aerial vehicles such as "feather-PVDF biological hybrid mechanoreceptor" developed by Shanghai Jiaotong University team and "Carrier Pigeon" developed by Northwestern Polytechnical University can mix in the bird group to deceive the detection of microwave radar system. If some technical means are used to integrate deceptive feathers or coating with flapping unmanned aerial vehicles, it can even reach the level of being difficult to distinguish the authenticity, and truly become "bird spy" that cannot be identified by optical systems and radar detection. Therefore, the identification of flying birds and flapping unmanned aerial vehicles faces great challenges, mainly in the high similarity of the two in flight mode and motion characteristics. The flapping unmanned aerial vehicle imitates the flapping of the wings of the flying bird to generate lift and propulsion, which makes the micro-motion characteristics of the flapping unmanned aerial vehicle very similar to those of the flying bird in frequency, amplitude, etc. In addition, the signal changes of flapping unmanned aerial vehicles and flying birds in different flight modes are diverse, which further increases the difficulty of differentiation.
[0004] Therefore, how to accurately detect and identify flying birds and flapping unmanned aerial vehicles is a problem that needs to be explored in the field of detection and identification.
[0005] Currently, flying birds and bionic flapping unmanned aerial vehicles have a high degree of similarity in motion characteristics. Traditional millimeter wave radar detection methods are limited by physical characteristics such as longer wavelength and insufficient resolution, and are difficult to effectively capture the fine micro-motion characteristics of flapping targets and effectively distinguish flying birds from bionic flapping unmanned aerial vehicles. In contrast, laser micro-Doppler technology, with its high resolution advantage, can accurately analyze kinematic parameters such as flapping frequency and amplitude, thereby achieving high spatial and temporal resolution in target detection. However, existing methods are mostly limited to single-modal time-frequency feature analysis, which has certain limitations. First, it ignores the essential differences in polarization scattering characteristics between mechanical structures and biological tissues; second, it cannot integrate the optical properties of target materials; third, it is easily affected by complex environmental interference and background heat sources, resulting in distortion of the feature space.
[0006] Aiming at the above problems existing in the traditional single micro-motion modal detection method, the present application proposes a flapping target detection and identification method based on laser micro-motion and polarization-infrared feature combination, which can improve the accuracy of bird and flapping unmanned aerial vehicle identification. SUMMARY
[0007] The purpose of the present application is to solve the above problems existing in the background art, and to propose a flapping target detection and identification method based on laser micro-motion and polarization-infrared feature combination.
[0008] The present application utilizes the differences in micro-motion, polarization and infrared modal characteristics of flying birds and flapping unmanned aerial vehicles, and effectively detects and identifies flying birds and flapping unmanned aerial vehicles by combining machine learning algorithms, thereby providing a direction for the detection and identification of flying birds and flapping unmanned aerial vehicles.
[0009] To achieve the above purpose, the technical scheme adopted by the present application is as follows:
[0010] The flapping target detection and identification method based on laser micro-motion and polarization-infrared feature combination, the implementation steps of the method are as follows:
[0011] Step one, select the identification features of laser micro-motion, polarization and infrared three modalities;
[0012] Step two, propose a multi-modal feature fusion criterion, and assign weights to the features;
[0013] Step three, construct a data set and combine machine learning algorithms for target identification.
[0014] Further, in step one, the identification features of laser micro-motion, polarization and infrared three modalities are selected; the specific process is as follows:
[0015] (1) In the identification features of the laser micro-motion modality;
[0016] For a one-segment flapping unmanned aerial vehicle, select the double-peak consistency without wing sweep frequency shift as the one-segment flapping unmanned aerial vehicle discrimination index;
[0017] For a two-segment flapping unmanned aerial vehicle, select the double-peak consistency without wing sweep frequency shift as the two-segment flapping unmanned aerial vehicle discrimination index;
[0018] For a flying bird, select the double-peak consistency with wing sweep frequency shift as the flying bird discrimination index;
[0019] (2) In the identification features of the polarization modality;
[0020] Select natural biological materials with a depolarization value higher than 0.5 as the discrimination index of flying birds;
[0021] Select artificial composite materials with a depolarization value lower than 0.5 as the discrimination index of flapping unmanned aerial vehicles;
[0022] (3) in the identification feature of the infrared modal;
[0023] The texture structure feature of the target surface is selected as the discrimination index of the flying bird and the flapping-wing unmanned aerial vehicle, and a target locking and image intercepting method based on brightness maximization is used for pre-processing and target extraction of the infrared image.
[0024] Further, the texture structure feature of the target surface is selected as the discrimination index of the flying bird and the flapping-wing unmanned aerial vehicle, and a target locking and image intercepting method based on brightness maximization is used for pre-processing and target extraction of the infrared image, specifically:
[0025] (3.1) first read the experimental picture data, according to the self-defined target size, size tolerance, brightness threshold and center cropping expansion quantity; then convert the picture into a gray image, find all the brightest points, calculate the average value of the row coordinates of these brightest points, and round off to the nearest integer as the center point row coordinate, similarly, calculate the average value of the column coordinates of all the brightest points, and round off to the nearest integer as the center point column coordinate, thus obtaining the average brightest point center;
[0026] (3.2) lock the average brightest point center, obtain the original image size, determine the cropping range and adjust the boundary in combination with the above center cropping expansion quantity, and complete the image cropping with the average brightest point center as the center;
[0027] (3.3) target positioning: first, use adaptive histogram equalization to enhance contrast; then, dynamic threshold segmentation: take the top 10% of the pixels with the highest brightness of the image, set the threshold to the minimum value of the 10% pixels, and the threshold is not less than 200, generate a binary mask; then, morphological processing is performed, and noise is removed and the boundary is smoothed through open-close operation combination; finally, region analysis and filtering of images with size in the range of target size ± 50% and width-height ratio between 0.5-2 are performed; the target size is 120x120 pixels;
[0028] (3.4) finally, target extraction is performed, the candidate region closest to the image center is selected, the boundary box is expanded, boundary protection is performed to prevent out-of-bound, and final cropping is performed.
[0029] Further, the texture structure feature of the target surface includes contrast, inverse difference moment, energy and entropy;
[0030] For flying birds, the average range of contrast is 0.269-0.312, the average range of inverse difference moment is 0.931-0.946, the average range of energy is 0.398-0.538, and the average range of entropy is 1.683-2.225;
[0031] For flapping-wing UAV, the average range of contrast is 0.027-0.068, the average range of inverse difference moment is 0.969-0.989, the average range of energy is 0.599-0.916, and the average range of entropy is 0.391-1.122.
[0032] Further, the formula for calculating the depolarization degree is as follows:
[0033] (1)
[0034] In formula (1), represents the depolarization degree, represents the polarization degree; polarization degree (2)
[0035] In formula (2), S0 represents the total intensity of incident light, S1 represents the linearly polarized light intensity in 0° and 90° directions, S2 represents the linearly polarized light intensity in 45° and 135° directions, and S3 represents the left and right circularly polarized light intensity.
[0036] Further, , , , The formula for calculating is as follows:
[0037] (3)
[0038] (4)
[0039] ; (5)
[0040] (6)
[0041] In formula (3)-(6), i is the gray value of the reference pixel or the starting pixel, j is the gray value of the adjacent pixel paired with the reference pixel i under a certain spatial relationship, represents the probability of the pixel pair in the gray matrix.
[0042] Further, in step two, the multi-modal feature fusion criterion is proposed, and the features are weighted and distributed; specifically:
[0043] The features of each of the above modes are combined with machine learning, and a multi-dimensional information fusion criterion is adopted, which takes the target micro-motion modal feature as the leading and the polarization and infrared features as the auxiliary, to realize efficient and accurate classification of the target; and the weight distribution is based on the physical significance of the features.
[0044] Further, the feature-based physical saliency weight distribution is combined with machine learning algorithm training, and the weight distribution principle is as follows:
[0045] The total weight of the micro-motion feature is 0.70, wherein the weight of the bimodal consistency is 0.55, and the weight of the sweep wing frequency shift is 0.15.
[0046] The weight of the polarization feature is 0.10.
[0047] The total weight of the image texture feature is 0.20, wherein the weight of the contrast is 0.15, the weight of the inverse difference moment is 0.02, the weight of the energy is 0.01, and the weight of the entropy is 0.02.
[0048] Further, in step three, the data set is constructed, and the machine learning algorithm is used for target recognition, specifically:
[0049] First, the training data set containing the features of flying birds, one-segment unmanned aerial vehicles and two-segment unmanned aerial vehicles is imported, the data order is shuffled to randomize the data, the data is normalized, the feature weights are set, and the weights are 0.55, 0.15, 0.10, 0.15, 0.02, 0.01 and 0.02; the normalized features are applied with weights, the test set is imported, the test set is processed using the normalization parameters of the training set, and the same feature weight preprocessing is applied, then three algorithm models are trained, which are MLP, RF and SVM, finally, the trained model is used to predict the test set, and the accuracy is calculated using the following formula:
[0050] Correct prediction number / total sample number=classification accuracy.
[0051] The correct prediction number and the total number of samples can be reflected in the program, and the total number of samples is the total number of the test set, and the correct prediction number is obtained in the training of the trained model to predict the test set.
[0052] The beneficial effects of the present application relative to the prior art are: the flapping target detection and identification method of the present application takes the micro-motion characteristics contained in complex flapping motion targets as the core, combines polarization and infrared characteristics, and combines machine learning algorithms to fuse the characteristics of the three modalities and perform target identification. The flapping target detection and identification method of the present application based on the combination of laser micro-motion and polarization-infrared characteristics significantly improves the recognition accuracy compared to the accuracy of the feature recognition method with only a single micro-motion modality. In the case of no noise, the accuracy of the single micro-motion modality feature under the RF, SVM and MLP algorithms is 58.83%, 59.83% and 57.5% respectively; while the accuracy of the multi-modal feature after the fusion of the RF, SVM and MLP algorithms is 91.83%, 93.17% and 93.50% respectively (the above data is the accuracy calculated by training the algorithm with the training set combined with the weight, and applying the trained algorithm to the test set. The specific calculation process of the multi-modal feature algorithm is as follows: first, import the training data set, shuffle the data order to randomize the data, normalize the data, set the feature weight, the weight is 0.55, 0.15, 0.10, 0.15, 0.02, 0.01, 0.02. Apply the weight to the normalized features, import the test set, process the test set using the normalization parameters of the training set, and apply the same feature weight preprocessing, then train three models, namely Random Forest (Random Forest, RF), Support Vector Machine (Support Vector Machine, SVM)
[0053] and Multilayer Perceptron (Multilayer Perceptron, MLP), which are currently available machine learning models, and finally use the trained models to predict the test set to calculate the above accuracy).
[0054] Compared with the existing flapping target recognition method, the flapping target detection and identification method based on the combination of laser micro-motion and polarization-infrared characteristics of the present application significantly improves the detection and identification accuracy of flying birds and flapping drones. BRIEF DESCRIPTION OF DRAWINGS
[0055] Figure 1 is a flying bird and drone feature time-frequency graph; wherein: Figure 1 (a) is a one-segment flapping drone feature time-frequency graph, Figure 1 (b) is a flying bird feature time-frequency graph, Figure 1 (c) is a two-segment flapping drone feature time-frequency graph.
[0056] Figure 2 is a natural biological material and artificial composite material depolarization graph; wherein: Figure 2 (a) is a downy feather biological material depolarization graph, Figure 2(b) is the depolarization map of hard feather biological materials, Figure 2 (c) is the depolarization map of artificial composite materials.
[0057] Figure 3 The texture characteristic parameter maps of real birds and bionic birds at different distances are shown in the following figures: Figure 3 (a) is the texture characteristic parameter map of real birds and bionic birds at a distance of 5 m, Figure 3 (b) is the texture characteristic parameter map of real birds and bionic birds at a distance of 10 m, Figure 3 (c) is the texture characteristic parameter map of real birds and bionic birds at a distance of 15 m.
[0058] Figure 4 The confusion matrix maps of the three algorithms are shown in the following figures: Figure 4 (a) is the confusion matrix map of the RF algorithm, Figure 4 (b) is the confusion matrix map of the SVM algorithm, Figure 4 (c) is the confusion matrix map of the MLP algorithm.
[0059] Figure 5 The technical roadmap of the flapping wing target detection and recognition method based on laser micro-motion and polarization-infrared feature combination of the present application is shown in the following figure. DETAILED DESCRIPTION
[0060] The technical solutions of the present application will be further described below in combination with the accompanying drawings, but are not limited thereto. Any modifications or equivalent replacements to the technical solutions of the present application without departing from the spirit and scope of the present application shall be covered in the protection scope of the present application. DETAILED DESCRIPTION
[0062] The present embodiment provides a flapping wing target detection and recognition method based on laser micro-motion and polarization-infrared feature combination. The implementation steps of the method are as follows:
[0063] Step one, selecting identification features of three modalities of laser micro-motion, polarization and infrared;
[0064] Step two, proposing a multi-modal feature fusion criterion and assigning weights to the features;
[0065] Step three, constructing a data set and combining a machine learning algorithm to perform target recognition.
[0066] Further, in step one, the identification features of three modalities of laser micro-motion, polarization and infrared are selected. The specific process is as follows:
[0067] (1) in the identification features of the laser micro-motion modality (as shown in the figure); Figure 1
[0068] For a one-segment ornithopter, the double-peak consistency non-sweep wing frequency shift is selected as the one-segment ornithopter discrimination index;
[0069] For a two-segment ornithopter, the non-double-peak consistency non-sweep wing frequency shift is selected as the two-segment ornithopter discrimination index;
[0070] For a flying bird, the non-double-peak consistency sweep wing frequency shift is selected as the flying bird discrimination index;
[0071] (4) In the identification characteristics of the polarization mode (as shown in Figure 2 );
[0072] Select natural biological materials (such as mountain chicken big points (a kind of chicken feather name), pearl chicken feathers, silver chicken phoenix feathers, goose feathers, hen nest feathers, or flower duck feathers) with a depolarization value higher than 0.5 as the discrimination index of flying birds;
[0073] Select artificial composite materials (such as epoxy carbon fiber or peptide lipid carbon fiber) with a depolarization value lower than 0.5 as the discrimination index of ornithopter; the depolarization value of 0.5 is obtained from experimental data analysis.
[0074] (5) In the identification characteristics of the infrared mode (as shown in Figure 3 );
[0075] Select the texture structure features of the target surface as the discrimination index of flying birds and ornithopters, and use the target locking and image cutting method based on brightness maximization to preprocess the infrared image (the purpose is to reduce the background influence and facilitate subsequent processing) and target extraction.
[0076] Further, the texture structure features of the target surface are selected as the discrimination index of flying birds and ornithopters, and the target locking and image cutting method based on brightness maximization is used for preprocessing and target extraction of the infrared image, specifically:
[0077] (3.1) First, read the experimental picture data, and according to the self-defined target size, size tolerance, brightness threshold, and center cropping expansion amount; then convert the picture to a grayscale image, find all the brightest points, calculate the average value of the row coordinates of these brightest points, and round it to the nearest integer as the center point row coordinate. Similarly, calculate the average value of the column coordinates of all the brightest points, and round it to the nearest integer as the center point column coordinate, thereby obtaining the average brightest point center;
[0078] (3.2) Lock the average brightest point center, obtain the original image size, determine the cropping range and adjust the boundary in combination with the above center cropping expansion amount, and complete the image cropping centered on the average brightest point center;
[0079] (3.3) Target localization: First, adaptive histogram equalization is used to enhance contrast; then dynamic thresholding is performed: the top 10% of pixels with the highest image brightness are selected, and the threshold is set to the minimum value of these 10% of pixels, and the threshold is not less than 200, generating a binary mask; then morphological processing is performed, and noise is removed and the boundaries are smoothed by combining opening and closing operations; finally, region analysis and screening are performed on images whose size is within ±50% of the target size and whose aspect ratio is between 0.5 and 2; the target size is 120×120 pixels;
[0080] Morphological opening and closing operations are existing algorithms. Reference: Zhang Ying. Research on the application of opening and closing operations in eliminating image noise [J]. Journal of Weifang University, 2002(2).
[0081] (3.4) Finally, target extraction is performed. The candidate region closest to the center of the image is selected, and the bounding box is expanded (left 50px, top 40px, width and height each +100px) to protect the boundary and prevent it from going out of bounds. The final cropping is then performed.
[0082] Furthermore, the texture structure features of the target surface include contrast, inverse difference moment, energy, and entropy;
[0083] For birds, the mean contrast ratio ranges from 0.269 to 0.312, the mean inverse moment ranges from 0.931 to 0.946, the mean energy ranges from 0.398 to 0.538, and the mean entropy ranges from 1.683 to 2.225.
[0084] For flapping-wing UAVs, the average contrast ratio ranges from 0.027 to 0.068, the average inverse moment ranges from 0.969 to 0.989, the average energy ranges from 0.599 to 0.916, and the average entropy ranges from 0.391 to 1.122.
[0085] The infrared images obtained in the experiment were processed using a method based on the gray-level co-occurrence matrix (a current technology) to obtain the corresponding values for the above features. It can be seen that there are significant differences between the four texture features of birds and flapping-wing UAVs.
[0086] The method based on the gray-level co-occurrence matrix is cited in this paper: Liu Guangyu, Huang Yi, Cao Yu, et al. Research on image texture feature extraction based on gray-level co-occurrence matrix [J]. Science and Technology Wind, 2021, (12): 61-64.
[0087] Furthermore, the formula for calculating the deflection degree is as follows:
[0088] (1)
[0089] In equation (1), Indicates the degree of deflection. polarization degree; polarization degree (2)
[0090] In formula (2), S0 represents the total intensity of incident light, S1 represents the linearly polarized light intensity in 0° and 90° directions, S2 represents the linearly polarized light intensity in 45° and 135° directions, and S3 represents the left and right circularly polarized light intensity. Most of the circularly polarized components can be ignored in the range of instrument detection, so the circularly polarized component is taken as 0 in multi-beam imaging polarization detection.
[0091] Further, contrast), inverse difference moment, energy, entropy) are as follows:
[0092] (3)
[0093] (4)
[0094] ; (5)
[0095] (6)
[0096] In formula (3)-formula (6), i is the gray value of a reference pixel or a starting pixel, j is the gray value of an adjacent pixel paired with the reference pixel i under a certain spatial relationship, represents the probability of the pixel pair in the gray matrix.
[0097] Further, in step two, the multi-modal feature fusion criterion (target classification criterion based on feature hierarchical fusion strategy) is proposed, and the features are weighted and distributed; specifically:
[0098] The features of each of the above modes are combined with machine learning, a multi-dimensional information fusion criterion is adopted, the target micro-motion modal feature is dominated, the polarization and infrared features are auxiliary, and efficient and accurate classification of the target is realized; and the weight distribution is based on the physical significance of the features.
[0099] Further, the weight distribution based on the physical significance of the features is combined with machine learning algorithm training, and the weight distribution principle is as follows:
[0100] The total weight of the micro-motion feature is 0.70, wherein the weight of the double-peak consistency is 0.55, and the weight of the sweep wing frequency shift is 0.15;
[0101] The weight of the polarization feature is 0.10;
[0102] The total weight of the image texture features is 0.20, wherein the contrast weight is 0.15, the inverse difference moment weight is 0.02, the energy weight is 0.01, and the entropy weight is 0.02.
[0103] The above data is obtained and distributed according to the following: according to the significant physical characteristics, combined with the training set and the data set, the SVM, MLP and RF algorithms are used to train to find the weights of the features corresponding to the best recognition accuracy.
[0104] Further, in step three, the data set is constructed, and a machine learning algorithm is used for target recognition, specifically:
[0105] First, the training data set containing the characteristics of flying birds, one-section unmanned aerial vehicles and two-section unmanned aerial vehicles is imported, the data order is shuffled to randomize the data, the data is normalized, the feature weights are set, and the weights are 0.55, 0.15, 0.10, 0.15, 0.02, 0.01, and 0.02; the normalized features are applied with weights, the test set is imported, the test set is processed using the normalization parameters of the training set, and the same feature weights are applied for preprocessing, then three algorithm models are trained, which are MLP, RF and SVM (all three algorithms are known algorithms), finally, the trained model is used to predict the test set, and the accuracy is calculated using the following formula:
[0106] Correct prediction number / total sample number=classification accuracy;
[0107] The correct prediction number and the total sample number can be reflected in the program, and the total sample number is the total number of the test set, and the correct prediction number is obtained in the training of the trained model to predict the test set.
[0108] Example 1:
[0109] As shown in Figure 5 , the present embodiment discloses a multi-modal feature fusion flapping wing target detection and recognition method based on laser micro-motion features and polarization-infrared features, which comprises the following steps:
[0110] First, select the micro-motion, polarization and infrared modal features, and the selection criteria are as follows:
[0111] (1) Micro-motion modal features: the specific feature time-frequency graph is as shown in Figure 1 , according to the results of the feature time-frequency graph, select the double-peak consistency and wing sweep frequency shift as the discrimination index of flying birds and flapping wing unmanned aerial vehicles. The specific difference lies in that flying birds have no double-peak consistency and have wing sweep frequency shift, one-section flapping wing unmanned aerial vehicles have double-peak consistency and have no wing sweep frequency shift, and two-section flapping wing unmanned aerial vehicles have no double-peak consistency and have no wing sweep frequency shift.
[0112] (2) Polarization modal features: the depolarization degree results of different materials are as shown in Figure 2As shown, the depolarization value of 0.5 is selected as the discrimination index of flying birds and flapping-wing drones according to the depolarization result. The specific difference lies in that the depolarization of natural biological materials (such as flying birds) is generally higher than 0.5, while the depolarization of artificial composite materials (such as flapping-wing drones) is basically lower than 0.5.
[0113] (3) Infrared modal characteristics: the texture characteristic parameters of true birds and bionic birds at different distances are as shown in the table. Figure 3 As shown, the texture structure characteristics of the target surface, i.e. contrast, inverse difference moment, energy and entropy value, are selected as the discrimination index of flying birds and flapping-wing drones. The specific difference lies in that flying birds show high contrast, low inverse difference moment, low energy and high entropy, reflecting the complexity and dynamics of natural texture. While flapping-wing drones show low contrast, high inverse difference moment and low entropy, reflecting the uniformity and static nature of artificial materials.
[0114] Next, a target classification criterion based on feature hierarchical fusion strategy is proposed, which combines the features of each modal with machine learning, adopts a multi-modal information fusion method dominated by target micro-motion features and assisted by polarization and infrared features, and realizes efficient and accurate classification of targets. The micro-motion level contains the physical unforgeability of the motion dynamics characteristics, the polarization level reflects the characteristic difference of the material, and the infrared level improves the accuracy of discrimination.
[0115] Based on the physical significance of features, the weight distribution is combined with the machine learning algorithm training, and the best weight distribution is as follows: the total weight of micro-motion features is 0.70, and its sub-feature double-peak consistency weight is 0.55, because it has strong characterization ability for periodic motion of mechanical structure and directly determines the classification of one-segment drones; the wing frequency shift weight is 0.15, because it realizes the effective separation of flying birds and two-segment drones by quantifying the frequency spectrum difference between biological flexible motion and mechanical rigid motion. The polarization feature weight is 0.10, and the degree of polarization as an intrinsic property of the material shows significant separability between biological feathers and polymer materials in laboratory measurement. The total weight of image texture features is 0.20, of which the contrast weight is 0.15, the difference between the gray gradient distribution is used to describe the sharpness of the target contour, which can obviously distinguish flying birds from flapping-wing drones, the inverse difference moment weight is 0.02, which quantifies the difference in surface uniformity; the energy weight is 0.01, which reflects the periodic characteristics of texture; the entropy weight is 0.02, which reflects the statistical characteristics of random distribution.
[0116] Example 2:
[0117] As shown in Figure 5 The embodiment provides a multi-modal feature fusion flapping-wing target detection and identification method based on laser micro-motion features and polarization-infrared features, and the specific implementation steps are as follows:
[0118] Firstly, a hybrid dataset is constructed based on the combination of physical mechanism modeling and measured data collection. The training set is generated by numerical simulation of three types of targets (one-segment ornithopter, two-segment ornithopter and flying bird), each generating 1000 numerical simulation samples (3000 in total), and randomizing processing to eliminate sequence bias. The test set contains 200 samples for each type (600 in total), in which: micro-motion feature data is obtained by model simulation, and depolarization and infrared texture features are derived from measured data in a laboratory environment.
[0119] Secondly, weight distribution is based on the physical significance of features, combined with machine learning training, and the best weights are as follows: the total weight of micro-Doppler features is 0.70, among which the weight of double-peak consistency is 0.55, and the weight of sweep-wing frequency shift is 0.15; the weight of polarization features is 0.10; the total weight of image texture features is 0.20, among which the weight of contrast is 0.15, the weight of inverse difference moment is 0.02, the weight of energy is 0.01, and the weight of entropy is 0.02.
[0120] Next, in order to compare the recognition accuracy under single micro-Doppler modality, the micro-Doppler feature weight is first set to 1. The recognition accuracy of each algorithm under single micro-Doppler modality is as follows: the recognition accuracy of RF algorithm is 58.83%, the recognition accuracy of SVM algorithm is 59.83%, and the recognition accuracy of MLP algorithm is 57.5%. The recognition accuracy of each algorithm based on multi-modal feature fusion is as follows: the recognition accuracy of RF algorithm is 91.83%, the recognition accuracy of SVM algorithm is 93.17%, and the recognition accuracy of MLP algorithm is 93.50%, which significantly improves the accuracy. Further combined with the confusion matrix, the confusion matrix of three-dimensional feature fusion is as shown in Figure 4 The analysis found from the confusion matrix that most of the recognition errors are that the two-segment unmanned aerial vehicle is incorrectly identified as a one-segment unmanned aerial vehicle, and the main reason may be that the features of the two types of targets overlap. In the micro-Doppler modality, both one-segment unmanned aerial vehicles and two-segment unmanned aerial vehicles have no sweep-wing frequency shift, but they are both man-made materials, and are similar in the polarization modality. MLP enhances feature fusion through deep nonlinear modeling, SVM strengthens the classification boundary with kernel function, and RF reduces accuracy due to its random feature sampling which ignores feature differences.
[0121] Double-peak consistency: indicates that in the micro-Doppler time-frequency graph, the micro-Doppler frequency shift values of the main peak and the secondary peak are consistent.
[0122] Sweep-wing frequency shift: indicates the frequency shift value below the base frequency value at time 0 in the micro-Doppler time-frequency graph.
[0123] Degree of depolarization: a physical quantity representing the degree of change in the polarization state of light during propagation or scattering. It measures the degree of change from a completely polarized state to a non-polarized state, and its value ranges from 0 to 1, and its relationship with the degree of polarization is:
[0124] (1)
[0125] In formula (1), represents the degree of depolarization, represents the degree of polarization.
[0126] Degree of polarization (2)
[0127] In formula (2), S0 represents the total intensity of incident light, S1 represents the intensity of linearly polarized light in the 0° and 90° directions, S2 represents the intensity of linearly polarized light in the 45° and 135° directions, and S3 represents the intensity of left and right circularly polarized light. Most of the circularly polarized component can be ignored in the range of instrument detection, so it is taken as 0 in the multi-beam imaging polarization detection.
[0128] In addition, it should be understood that although the present specification is described in terms of embodiments, not every embodiment contains only one independent technical solution, and the description of the specification is only for the sake of clarity, and those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be combined appropriately to form other embodiments that those skilled in the art can understand.
Claims
1. A flapping-wing target detection and identification method based on laser micro-motion and polarization-infrared feature combination, characterized in that: The implementation steps of the method are as follows: Step one, select the identification features of laser micro-motion, polarization and infrared three modalities; Step two, propose a multi-modal feature fusion criterion, and assign weights to the features; Step three, build a data set and combine machine learning algorithms for target recognition.
2. The flapping target detection and recognition method based on laser micro-motion and polarization-infrared feature combination according to claim 1, characterized in that: In step one, the identification features of laser micro-motion, polarization and infrared three modalities are selected. The specific process is as follows: (1) In terms of identification features of the laser micro-motion modality; For a one-segment flapping-wing unmanned aerial vehicle, select the double-peak consistency non-sweep-wing frequency shift as the one-segment flapping-wing unmanned aerial vehicle discrimination index; For a two-segment flapping-wing unmanned aerial vehicle, select the non-double-peak consistency non-sweep-wing frequency shift as the two-segment flapping-wing unmanned aerial vehicle discrimination index; For a flying bird, select the non-double-peak consistency sweep-wing frequency shift as the flying bird discrimination index; (2) In terms of identification features of the polarization modality; Select natural biological materials with a depolarization value higher than 0.5 as the discrimination index for flying birds; Select artificial composite materials with a depolarization value lower than 0.5 as the discrimination index for flapping-wing unmanned aerial vehicles; (3) In terms of identification features of the infrared modality; Select the texture structure features of the target surface as the discrimination index for flying birds and flapping-wing unmanned aerial vehicles, and use the target locking and image cropping method based on brightness maximization to preprocess the infrared image and extract the target.
3. The flapping target detection and recognition method based on laser micro-motion and polarization-infrared feature combination according to claim 2, characterized in that: The specific process is as follows: (3.1) First, read the experimental picture data, and define the target size, size tolerance, brightness threshold and center cropping expansion amount; then convert the picture to a grayscale image, find all the brightest points, calculate the average value of the row coordinates of these brightest points, and round it to the nearest integer as the center point row coordinate. Similarly, calculate the average value of the column coordinates of all the brightest points, and round it to the nearest integer as the center point column coordinate, thus obtaining the average brightest point center; (3.2) Lock the average brightest point center, obtain the original image size, determine the cropping range and adjust the boundary according to the above center cropping expansion amount, and complete the image cropping centered on the average brightest point center; (3.3) Target positioning: first, use adaptive histogram equalization to enhance contrast; then perform dynamic threshold segmentation: take the top 10% of the brightest pixels in the image, set the threshold to the minimum value of the 10% pixels, and the threshold is not less than 200, generate a binary mask; then perform morphological processing to remove noise and smooth the boundary through open-close operation combination; finally, perform region analysis and select images with a size within the target size ± 50% range and a width-to-height ratio between 0.5 and 2; the target size is 120x120 pixels; (3.4) Finally, perform target extraction, select the candidate region closest to the image center, expand the boundary box, perform boundary protection to prevent boundary overflow, and perform final cropping.
4. The flapping target detection and recognition method based on laser micro-motion and polarization-infrared feature combination according to claim 2 or 3, characterized in that: The texture structure features of the target surface include contrast, inverse difference moment, energy and entropy. For flying birds, the average range of contrast is 0.269-0.312, the average range of inverse difference moment is 0.931-0.946, the average range of energy is 0.398-0.538, and the average range of entropy is 1.683-2.225; For flapping-wing drones, the average range of contrast is 0.027-0.068, the average range of inverse difference moment is 0.969-0.989, the average range of energy is 0.599-0.916, and the average range of entropy is 0.391-1.
122.
5. The flapping target detection and recognition method based on laser micro-motion and polarization-infrared feature combination according to claim 2, characterized in that: The formula for calculating the degree of depolarization is as follows: （1） In formula (1), denotes the degree of depolarization, denotes the degree of polarization; Polarization degree (2) In formula (2), S0 represents the total intensity of incident light, S1 represents the linearly polarized light intensity in the 0° and 90° directions, S2 represents the linearly polarized light intensity in the 45° and 135° directions, and S3 represents the left and right circularly polarized light intensity.
6. The flapping target detection and recognition method based on laser micro-motion and polarization-infrared feature combination according to claim 4, characterized in that: , , , The calculation formulas of the above are as follows: （3） （4） ； （5） （6） In the formula (3) - formula (6), i is the gray value of a reference pixel or a starting pixel, j is the gray value of a neighboring pixel paired with the reference pixel i under a certain spatial relationship, represents the probability of a pixel pair in the gray matrix.
7. The flapping target detection and recognition method based on laser micro-motion and polarization-infrared feature combination according to claim 1, characterized in that: In step two, the multi-modal feature fusion criterion is proposed, and the features are weighted; specifically: The features of each modality are combined with machine learning, and a multi-dimensional information fusion criterion is adopted, which takes the target micro-motion modal feature as the main guide and the polarization and infrared features as auxiliary, to realize efficient and accurate classification of the target; and the weight distribution is based on the physical significance of the features.
8. The flapping target detection and recognition method based on laser micro-motion and polarization-infrared feature combination according to claim 7, characterized in that: The weight distribution based on the physical significance of the features is combined with machine learning algorithm training, and the weight distribution principle is as follows: The total weight of the micro-motion feature is 0.70, of which the weight of the double peak consistency is 0.55 and the weight of the wing frequency shift is 0.15; The weight of the polarization feature is 0.10; The total weight of the image texture feature is 0.20, of which the weight of the contrast is 0.15, the weight of the inverse difference moment is 0.02, the weight of the energy is 0.01, and the weight of the entropy is 0.
02.
9. The flapping target detection and recognition method based on laser micro-motion and polarization-infrared feature combination according to claim 1, characterized in that: In step three, the data set is constructed, and machine learning algorithm is used for target recognition, specifically: First, import the training data set containing the features of flying birds, one-segment drones and two-segment drones, shuffle the data order to randomize the data, normalize the data, set the feature weights, and the weights are 0.55, 0.15, 0.10, 0.15, 0.02, 0.01, and 0.02; apply the weights to the normalized features, import the test set, process the test set using the normalization parameters of the training set, and apply the same feature weights for preprocessing, then train three algorithm models, namely MLP, RF and SVM, finally, use the trained model to predict the test set, and calculate the accuracy rate using the following formula; Correct prediction number / total sample number=classification accuracy rate; The correct prediction number and the total number of samples can be reflected in the program, and the total number of samples is the total number of the test set, and the correct prediction number is obtained in the training of the trained model to predict the test set.