A UAV resource scheduling method for small target identification

Through multi-dimensional information analysis and decision function calculation, comprehensively considering the size, movement trajectory and semantic information of the drone target, the problems of high false alarm rate and resource waste when identifying small targets are solved, and more efficient resources are achieved Scheduling and identification accuracy.

CN114638490BActive Publication Date: 2024-11-12THE 54TH RESEARCH INSTITUTE OF CHINA ELECTRONICS TECHNOLOGY GROUP CORPORATION
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210231824.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-09
Publication Date
2024-11-12
Estimated Expiration
2042-03-09

Smart Images

  • Figure CN114638490B_ABST
    Figure CN114638490B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for dispatching UAV resources for small target identification, which belongs to the field of UAV intelligent decision-making. The method comprehensively considers target size information, target movement trajectory information, target movement speed information, and target activity area image semantic segmentation information, designs multiple thresholds, and provides reference information for dispatching flight resources of UAVs, so as to reduce the false alarm rate of small target detection, reduce the cost of resource dispatching, and improve the utilization rate of UAV resources.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of unmanned aerial vehicle (UAV) intelligent decision-making, and in particular relates to a UAV resource scheduling method for small target identification. Technical Background

[0002] A drone is a controllable, reusable unmanned aerial vehicle that can carry multiple mission payloads and perform multiple tasks. Drones can carry visible light zoom equipment to implement intelligent monitoring and scene perception within the jurisdiction area. They have the advantages of flexibility, high cost performance, etc. in modern agriculture, smart cities, smart patrols, etc. Due to the limited battery capacity of drones, the resource scheduling and use of drones is particularly important for the drones to successfully complete their flight missions.

[0003] In actual situations, the distance between the drone and the target is far, and the scale of the target in the drone image is very small. Some targets are often only a dozen pixels, and the features such as shape and contour are not significant. The target contains weak information, which makes detection more difficult. Further accurate identification requires the drone to mobilize load resources or flight resources. A single small target detection algorithm has a certain false alarm rate for the detection results of a specific category of targets. If the drone load or flight resource is scheduled for each detection result, it will cause a certain waste of resources. Therefore, it is very important to establish a drone resource scheduling method for small target identification. Summary of the invention

[0004] Aiming at the problems existing in the process of existing UAVs identifying small targets, the present invention proposes a UAV resource scheduling method for small target identification. After detecting the target, multi-dimensional information analysis is performed to increase the accuracy of small target determination and reduce resource waste.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] A method for scheduling UAV resources for small target identification includes the following steps:

[0007] Step 1: Use the target detection algorithm to detect specific small targets in the image taken by the drone, and after locking the target, obtain the target position and size information;

[0008] Step 2: Use the target tracking algorithm to track the specific small target and obtain the target's moving trajectory information and moving speed information;

[0009] Step 3, intercepting the target activity area image according to the target's moving trajectory information, and performing semantic segmentation processing on the target activity area image;

[0010] Step 4, analyzing the semantic segmentation processing result of the target active area image to obtain the semantic coefficient and texture coefficient in the decision function;

[0011] Step 5, calculating the decision function according to the size information, movement trajectory information, movement speed information, semantic coefficient and texture coefficient of the target;

[0012] Step 6: UAV resource scheduling is performed based on the calculation result of the decision function. If the calculation result of the decision function is greater than the scheduling threshold, the camera payload will adjust the attitude and zoom or reduce the flight altitude of the UAV. Otherwise, UAV resource scheduling will not be performed.

[0013] Furthermore, the specific formula of the decision function f is as follows:

[0014]

[0015] Among them, ε1 represents the size hyperparameter, ε2 represents the semantic coefficient, ε3 represents the texture coefficient, S 阈 is the small target size determination threshold, d 阈 is the target movement displacement threshold, v 阈 is the target moving speed threshold, d is the moving displacement of the specific small target obtained according to the moving trajectory information, v is the moving speed of the specific small target, and w, h are the size information of the specific small target.

[0016] Furthermore, the displacement d of a specific small target is calculated based on the target position (x, y) at the end of tracking and the initial position (x0, y0). The specific formula is as follows:

[0017]

[0018] Furthermore, the moving speed v of a specific small target is based on the average of the position coordinate differences (Δx, Δy) at intervals of Δt: The specific formula is as follows:

[0019]

[0020] Furthermore, the size hyperparameter ε1 is set according to the application scenario requirements, the semantic coefficient ε2 is determined according to the region category of the target in the semantic segmentation processing result, and the texture coefficient ε3 is determined according to the number of categories in the semantic segmentation processing result.

[0021] The beneficial effects of the present invention are:

[0022] The method of the present invention comprehensively considers target size information, target movement trajectory information, target movement speed information, and target activity area image semantic segmentation information, designs multiple thresholds, and provides decision information for UAV scheduling flight resources and payload resources, thereby achieving the purpose of reducing the false alarm rate of small target detection, reducing the cost of resource scheduling, and improving the utilization rate of UAV resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 The figure is a flow chart of a method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to facilitate those skilled in the art to understand and implement the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0025] See also Figure 1 , a method for scheduling UAV resources for small target identification. In this embodiment, the UAV flight altitude is set to 150 meters, the initial focal length of the camera device is 4.3 mm, and the hyperparameter ε1, the small target size determination threshold S 阈 , target movement displacement threshold d 阈 、Target moving speed threshold v 阈 and the decision function f 阈 . According to the application scenario, the semantic segmentation categories are divided into sensitive categories, interesting categories and general categories to provide a basis for determining the category of the target area. The target detection algorithm is implemented using the YOLOv4-tiny algorithm, the target tracking algorithm is implemented using the improved KCF algorithm, and the semantic segmentation algorithm is implemented using the Deeplabv3+ algorithm. It includes the following steps:

[0026] Step 1: Use the target detection algorithm to detect specific small targets in the image taken by the drone. After locking the target, obtain the target position (x0, y0) and size information (w, h);

[0027] Step 2: Use the target tracking algorithm to track the specific small target and obtain the target's trajectory information and movement speed information. The tracking duration can be set according to the actual situation. In this embodiment, the tracking duration is set to t. Among them, the target's movement displacement d is calculated based on the target position (x, y) at the end of tracking and the initial position (x0, y0). The specific formula is as follows:

[0028]

[0029] Set the interval time Δt, calculate the target's moving speed v, based on the average of the interval time position coordinate difference (Δx, Δy) The specific formula is as follows:

[0030]

[0031] Step 3, intercepting the target activity area image according to the target movement trajectory information, and performing semantic segmentation processing on the image;

[0032] Step 4: Analyze the semantic segmentation results of the target activity area image; determine the semantic coefficient ε2 according to the category of the target area in the semantic segmentation result, and the formula is as follows:

[0033]

[0034] The texture coefficient ε3 is determined according to the number of categories m in the semantic segmentation result. The formula is as follows:

[0035]

[0036] Step 5: Determine the threshold S based on the known hyperparameter ε1 and the small target size 阈 , target movement displacement threshold d 阈 、Target moving speed threshold v 阈 The decision function is calculated based on the obtained target size information (w, h), target movement displacement information d and speed information v, target activity area image semantic coefficient ε2 and texture coefficient ε3. The specific formula of the decision function f is as follows:

[0037]

[0038] This formula comprehensively considers the application scenarios, combines the semantic coefficient with the target displacement, and combines the texture coefficient with the target speed.

[0039] Step 6: Make a UAV resource scheduling decision based on the result of the decision function calculation. If f>f 阈 , the camera payload will adjust the attitude and zoom or reduce the flight altitude of the UAV. If f<f 阈 , no resources need to be scheduled.

[0040] In summary, the present invention comprehensively considers target size information, target movement trajectory information, target movement speed information, and target activity area image semantic segmentation information, designs multiple thresholds, and provides reference information for UAV flight resource scheduling, thereby achieving the purpose of reducing the false alarm rate of small target detection, reducing the cost of resource scheduling, and improving the utilization rate of UAV resources.

[0041] The above is a detailed introduction to a drone resource scheduling method for small target identification proposed in the present invention, in which specific examples are used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention, and does not limit the scope of rights of the present invention. All equivalent changes made using the contents of the present invention specification and its drawings are included in the scope of rights of the present invention.

Claims

1. A UAV resource scheduling method for small target identification, It is characterized in that The following steps are involved: Step 1: Use the target detection algorithm to detect specific small targets in the image taken by the drone, and after locking the target, obtain the target position and size information; Step 2: Use the target tracking algorithm to track the specific small target and obtain the target's moving trajectory information and moving speed information; Step 3, intercepting the target activity area image according to the target's moving trajectory information, and performing semantic segmentation processing on the target activity area image; Step 4, analyzing the semantic segmentation processing result of the target active area image to obtain the semantic coefficient and texture coefficient in the decision function; Step 5, calculate the decision function according to the size information, movement trajectory information, movement speed information, semantic coefficient and texture coefficient of the target; the specific formula of the decision function f is as follows: Among them, ε 1 represents the size hyperparameter, ε 2 represents the semantic coefficient, ε 3 represents the texture coefficient, S 阈 is the small target size determination threshold, d 阈 is the target movement displacement threshold, v 阈 is the target moving speed threshold, d is the moving displacement of the specific small target obtained according to the moving trajectory information, v is the moving speed of the specific small target, w, h are the size information of the specific small target; The displacement d of a specific small target is based on the target position (x, y) at the end of tracking and the initial position (x 0 ,y 0 ) is calculated, the specific formula is as follows: The moving speed v of a specific small target is based on the average of the position coordinate differences (Δx, Δy) at intervals of Δt. The specific formula is as follows: Step 6: UAV resource scheduling is performed based on the calculation result of the decision function. If the calculation result of the decision function is greater than the scheduling threshold, the camera payload will adjust the attitude and zoom or reduce the flight altitude of the UAV. Otherwise, UAV resource scheduling will not be performed.

2. A method for scheduling unmanned aerial vehicle resources for small target identification according to claim 1, It is characterized in that Size hyperparameter ε 1 Set according to the application scenario requirements, the semantic coefficient ε 2 The texture coefficient ε is determined according to the region category of the target in the semantic segmentation processing result. 3 Determined by the number of categories in the semantic segmentation processing results.