Unmanned aerial vehicle classification method based on graph regularization and sparse representation
Through the drone classification method based on graph regularization and sparse representation, the objective function is trained using the graph Laplace matrix and ALM method, which solves the problem of low recognition accuracy in complex environments, and achieves higher accuracy and robustness of identification.
Patent Information
- Application Number
- CN202410965383.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-18
- Publication Date
- 2024-11-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing drone recognition technology has low recognition accuracy in complex environments, making it difficult to effectively utilize structural information in image data, and it is difficult to obtain precise labeling training data, resulting in insufficient recognition robustness and accuracy.
Using the drone classification method based on graph regularization and sparse representation, the objective function is trained through the graph Laplace matrix and the augmented Lagrangian method (ALM), and the l1 and 2,1 norm constraints are introduced to learn image data. The topological structure and label distribution of the scattered representation and graph regularization optimization model is constructed, key features of the drone image are extracted and structural relationships are maintained.
It improves the accuracy and robustness of drone identification, enhances the stability and consistency of the model under different conditions and backgrounds, and reduces the risk of overfitting.
Smart Images

Figure CN119027714A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of image processing and drone technology, and in particular to a drone classification method based on graph regularization and sparse representation. Background Art
[0002] In today's technological landscape, drones are widely used in a variety of critical areas, including military reconnaissance, traffic monitoring, agricultural surveillance, and disaster response. With the surge in the number of drones and the diversification of their application scenarios, effective and accurate drone identification has become crucial. Traditional drone identification technologies rely primarily on radar, infrared, and sonar systems. While these systems perform well in specific environments, they often require expensive equipment and complex deployment.
[0003] Image-based drone recognition offers a cost-effective alternative to traditional methods. However, this approach faces numerous challenges in practical application, primarily due to the difficulty drones can blend into their surroundings in complex environments, as well as factors such as lighting, weather conditions, image resolution, and drone speed that can affect recognition accuracy. Furthermore, existing image recognition algorithms often require extensive training data, and obtaining accurately labeled training data is a significant challenge in practice.
[0004] Furthermore, most image recognition systems fail to effectively utilize the inherent structural information in image data, such as the relative position of the drone and the background, and the geometric relationships between drone components. These are potential factors for improving recognition system performance. Fully utilizing this structural information in images would significantly improve recognition robustness and accuracy.
[0005] In view of this, we propose a drone classification method based on graph regularization and sparse representation. Summary of the Invention
[0006] The purpose of the present invention is to provide a drone classification method based on graph regularization and sparse representation to solve the problems raised in the above background technology.
[0007] To achieve the above objectives, the present invention provides the following technical solution: a drone classification method based on graph regularization and sparse representation, comprising the following steps:
[0008] S1. Collect a dataset containing different categories of drone objects.
[0009] S2, training the objective function based on the graph Laplacian matrix and label distribution and introducing the ALM method to update related parameters;
[0010] S3. Use the solved parameters to conduct classification tests on new samples.
[0011] Optionally, the S1 further includes:
[0012] where x i ∈X∈Rn ×d , n is the total number of samples, d is the feature dimension. i is the i-th sample, y i is the sample x i The dataset contains drone images in a variety of flight states, angles, and environmental conditions.
[0013] Optionally, the S2 further includes:
[0014] S21. Learn the internal topological structure of the sample on the original data and construct the optimization function of the label distribution as follows:
[0015]
[0016] StQ=H+M,XQ×1=1n,WTW=
[0017] S22. By approximating the non-convex constraints as convex functions and introducing variables for decoupling, the augmented Lagrangian function of the objective function is obtained, and the ALM method is used to solve and optimize each parameter;
[0018] S23. Calculate and update parameter H:
[0019]
[0020] in
[0021] Calculate and update the parameter M:
[0022]
[0023] in
[0024] Optionally, the S21 includes: where H is a projection matrix using l1 norm constraint, which is mainly used to learn the mapping relationship between specific features and label distribution D. In order to prevent overfitting during training, we introduce a matrix M using 2,1 norm constraint. L is the Laplace matrix calculated based on the sample, and λ1, λ2, λ3 are regularization parameters. In order to learn the label distribution of the sample, we use XQ×1 l =1 n To constrain.
[0025] Optionally, the S22 includes: the augmented Lagrangian function corresponding to the model is:
[0026]
[0027] Compared with the existing technology, the present invention provides a drone classification method based on graph regularization and sparse representation, which has the following beneficial effects:
[0028] This drone classification method based on graph regularization and sparse representation constructs an optimization model that incorporates sparse representation and graph regularization to extract key features from drone image data and maintain the intrinsic structural relationships of the sample data to enhance recognition accuracy and robustness. The method first standardizes the input image through a preprocessing step, then uses sparse representation technology to extract the essential features of the image, and then uses graph regularization technology to strengthen the structural similarity between these feature vectors, ultimately forming a robust feature representation framework. In addition, the present invention also uses the Laplacian matrix to impose smoothness constraints in the feature space to promote the consistency and stability of the model when processing drone images from different conditions and backgrounds. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION
[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0031] like Figure 1 As shown, the present invention provides a technical solution: a drone classification method based on graph regularization and sparse representation, comprising the following steps:
[0032] Step 1: Collect a dataset containing different categories of drone objects where x i ∈X∈Rn ×d , n is the total number of samples, d is the feature dimension. i is the i-th sample, y i is the sample x i Ensure that the dataset contains drone images in a variety of flight states, angles, and environmental conditions.
[0033] Step 2: Train the objective function based on the graph Laplacian matrix and label distribution and introduce the ALM method to update related parameters.
[0034] Step 1: Learn the internal topological structure of the sample on the original data and construct the optimization function of the label distribution as follows:
[0035]
[0036] stQ=H+M,XQ×1=1,WTW=
[0037] Here, H is a projection matrix constrained by the l1 norm, which is used to learn the mapping relationship between specific features and the label distribution D. To prevent overfitting during training, we introduce a matrix M constrained by the 2,1 norm. L is the Laplacian matrix calculated based on the samples, and λ1, λ2, and λ3 are regularization parameters. To learn the label distribution of the samples, we use the constraint XQ×1l=1n.
[0038] Step 2: By approximating the non-convex constraints as convex functions and introducing variables for decoupling, we obtain the augmented Lagrangian function of the objective function, and then use the ALM method to solve and optimize each parameter. The corresponding augmented Lagrangian function of the model is:
[0039]
[0040] Step 3, calculate and update the parameter H:
[0041]
[0042] in
[0043] Calculate and update the parameter M:
[0044]
[0045] in
[0046] Step 3: Use the solved parameters to conduct classification tests on new samples.
[0047] The above generally describes the present invention in detail. However, it is obvious to those skilled in the art that modifications or improvements may be made based on the present invention. Therefore, modifications or improvements that do not depart from the spirit of the present invention are within the scope of protection of the present invention.
Claims
1. A drone classification method based on graph regularization and sparse representation, Features: The steps include: S1. Collect a dataset containing different categories of drone objects. S2, training the objective function based on the graph Laplacian matrix and label distribution and introducing the ALM method to update related parameters; S3. Use the solved parameters to conduct classification tests on new samples.
2. A drone classification method based on graph regularization and sparse representation according to claim 1, Features: The S1 further comprises: where x i ∈X∈R n×d , n is the total number of samples, d is the feature dimension. i is the i-th sample, y i is the sample x i The dataset contains drone images in a variety of flight states, angles, and environmental conditions.
3. The drone classification method based on graph regularization and sparse representation according to claim 1, Features: The S2 further comprises: S21. Learn the topological structure inside the sample on the original data and construct the optimization function of label distribution as follows: s.t.Q=H+M,XQ×1 l =1 n ,W T W=I S22. By approximating the non-convex constraints as convex functions and introducing variables for decoupling, the augmented Lagrangian function of the objective function is obtained, and the ALM method is used to solve and optimize each parameter; S23, calculate and update parameter H: in Calculate and update the parameter M: in 4. A method for drone classification based on graph regularization and sparse representation according to claim 3, Features: The S21 includes: wherein H is used l 1 The projection matrix with norm constraints is mainly used to learn the mapping relationship between specific features and label distribution D. In order to prevent overfitting during training, we introduce a matrix M with 2,1 norm constraints. L is the Laplace matrix calculated based on the sample, λ 1 ,λ 2 ,λ 3 is the regularization parameter. In order to learn the label distribution of the sample, we use XQ×1 l =1 n To constrain.
5. The method for drone classification based on graph regularization and sparse representation according to claim 1, Features: The S22 includes: the augmented Lagrangian function corresponding to the model is:
