Airspace prevention patrol intelligent supervision system and method based on unmanned aerial vehicle

Abstract

The invention relates to the technical field of airspace patrol supervision, in particular to an airspace prevention patrol intelligent supervision system and method based on an unmanned aerial vehicle, and the system comprises the steps: constructing a flyer image recognition model through collected unmanned aerial vehicle patrol airspace sample image feature data; based on the flyer image recognition model, scientifically analyzing whether a flyer appears in the unmanned aerial vehicle patrol airspace real-time image data after data preprocessing, and if not, continuing to execute patrol operation by the unmanned aerial vehicle; otherwise, generating image feature data of the flyer in the patrol airspace, matching the image feature data with preset image feature data of the flyer in legal access, accurately identifying whether the flyer appearing in the patrol airspace belongs to legal access or not, if not, constructing unmanned aerial vehicle patrol supervision data, and pushing the unmanned aerial vehicle patrol supervision data to an unmanned aerial vehicle airspace patrol supervision platform. And an abnormal flyer intrusion alarm is given out, so that intelligent supervision of airspace patrol of the unmanned aerial vehicle and timely early warning when the abnormal flyer is found are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of airspace inspection and supervision, and specifically to an intelligent supervision system and method for airspace prevention and inspection based on unmanned aerial vehicles (UAVs). Background Art

[0002] With the continuous development of science and technology, UAV technology has gradually been applied in many fields. Therefore, the requirements for airspace inspection and supervision are getting higher and higher. Illegal flying objects pose a severe challenge to airspace safety. The existing technology lacks efficient linkage for rapid identification, accurate determination, and timely feedback of flying objects, resulting in poor quality of airspace inspection and supervision operations.

[0003] The Chinese invention patent with the publication number CN111640272B introduces a method and system for UAV monitoring and early warning based on blockchain. By collecting real-time information of UAVs through monitoring nodes, it analyzes whether the current position of the UAV is within the controlled airspace according to the real-time information. If so, a violation identifier V is added before the UAV information, and an early warning message is sent; otherwise, a normal identifier N is added before the UAV information; the identified UAV information is fed back to the blockchain for judgment and early warning, and the controlled airspace information and the information of UAVs illegally entering the controlled airspace are synchronized in real time among multiple different institutions, thereby improving the data sharing and emergency response efficiency of the corresponding institutions. However, it cannot monitor illegal flying objects appearing in the airspace, and the monitoring and early warning of the controlled airspace are not comprehensive enough. Summary of the Invention

[0004] (I) Technical Problems to be Solved To solve the deficiencies in the background art, the present invention provides an intelligent supervision system and method for airspace prevention and inspection based on UAVs, realizing intelligent supervision of UAV airspace inspection and timely early warning when abnormal flying objects are found.

[0005] (II) Technical Solutions An intelligent supervision method for airspace prevention and inspection based on UAVs includes the following steps: S1. Collect the characteristic data of the sample images of the airspace inspected by the UAV; S2. Build an image recognition model of flying objects based on the characteristic data of the sample images of the airspace inspected by the UAV; S3. Collect the real-time image data of the airspace inspected by the UAV; S4. Perform data preprocessing on the real-time image data of the airspace inspected by the UAV to generate the characteristic data of the real-time images of the airspace inspected by the UAV; S5. Perform object recognition processing based on the object image recognition model and the real-time image feature data of the UAV patrol airspace to generate object recognition data for the patrol airspace. If the object recognition data for the patrol airspace indicates the presence of an object, generate object image feature data for the patrol airspace and proceed to S6; otherwise, do not perform any operations. S6. Establish legally accessible object image feature data, perform object feature matching processing on the object image feature data of the patrol airspace and the legally accessible object image feature data to generate object access analysis data for the patrol airspace. If the object access analysis data for the patrol airspace indicates legal access, do not perform any operations; otherwise, generate real-time position coordinate data of the patrol UAV and proceed to S7. S7. Construct UAV patrol supervision data based on the real-time image feature data of the UAV patrol airspace, the object image feature data of the patrol airspace, the object access analysis data of the patrol airspace, and the real-time position coordinate data of the patrol UAV, and push it to the UAV airspace patrol supervision platform.

[0006] The present invention constructs an object image recognition model through the collected sample image feature data of the UAV patrol airspace, scientifically analyzes whether an object appears in the real-time image data of the UAV patrol airspace after data preprocessing based on the object image recognition model. If not, the UAV continues to perform patrol operations; otherwise, generate object image feature data for the patrol airspace, and match it with the preset legally accessible object image feature data to accurately identify whether the object appearing in the patrol airspace belongs to legal access. If it does not belong to legal access, construct UAV patrol supervision data and push it to the UAV airspace patrol supervision platform to realize intelligent supervision of UAV airspace patrol.

[0007] Preferably, the specific steps for collecting sample image feature data of the UAV patrol airspace are as follows: S11. Online collect the image feature data captured by several UAVs during airspace patrol operations through the UAV airspace patrol supervision platform to obtain a sample image feature data set of the UAV patrol airspace , where represents the th sample image feature data of the UAV patrol airspace, represents the total number of sample image feature data of the UAV patrol airspace, and the image feature data includes image feature data with an object and image feature data without an object.

[0008] Preferably, the specific steps for constructing an object image recognition model based on the sample image feature data of the UAV patrol airspace are as follows: S21. Set the training data ratio and the test data ratio, and divide the drone patrol airspace sample image feature dataset according to the training data ratio and the test data ratio to obtain a drone patrol airspace sample image feature training dataset and a drone patrol airspace sample image feature test dataset respectively; S22. Construct an initial convolutional neural network model, and set the number of input nodes of the initial convolutional neural network model to be , the number of hidden nodes to be , the number of output nodes to be , the initial weight to be and the initial bias to be ; S23. Set the training error threshold and the maximum number of training times, input the drone patrol airspace sample image feature training data in the drone patrol airspace sample image feature training dataset into the initial convolutional neural network model for training. If the training error during the training process is less than the training error threshold or the number of training times is greater than the maximum number of training times, stop training to obtain a trained convolutional neural network model; otherwise, continue training until the training error during the training process is less than the training error threshold or the number of training times is greater than the maximum number of training times; S24. Set the test accuracy threshold, input the drone patrol airspace sample image feature test data in the drone patrol airspace sample image feature test dataset into the trained convolutional neural network model for testing, calculate the test accuracy. When the test accuracy is greater than or equal to the test accuracy threshold, use the trained convolutional neural network model as the flying object image recognition model; otherwise, use the grid search algorithm to optimize the parameters of the trained convolutional neural network model to obtain the flying object image recognition model.

[0009] By collecting drone patrol airspace sample image feature data online through the drone airspace patrol supervision platform and training and testing the initial convolutional neural network model, a flying object image recognition model is obtained, providing an accurate and reliable flying object recognition tool for airspace patrol operations, improving the speed and efficiency of the flying object recognition process, and ensuring the timeliness of the flying object recognition results.

[0010] Preferably, the specific steps for collecting real-time image data of the drone patrol airspace are as follows: S31. Collect the preset patrol route of the target prevention airspace online through the drone airspace patrol supervision platform, control the target drone to perform airspace patrol operations in the target prevention airspace according to the preset patrol route, and use the image shooting device installed on the target drone to regularly shoot the real-time image data of the target prevention airspace to generate real-time image data of the drone patrol airspace , the image capturing device represents any one of a high-definition camera, a panoramic camera, and a low-light camera.

[0011] Preferably, the specific steps for preprocessing the real-time image data of the drone patrol airspace to generate the real-time image feature data of the drone patrol airspace are as follows: S41. Perform image data noise reduction processing on the real-time image data of the drone patrol airspace through a weighted average filtering method to generate the real-time image feature data of the flying objects in the patrol airspace ; the process of image data noise reduction is as follows: S411. Set a filtering window, select any pixel point in the real-time image data of the drone patrol airspace , and align the center point of the filtering window with this pixel point; S412. Calculate the weight coefficients of each pixel point in the filtering window through a weighted kernel function formula; the weighted kernel function formula is as follows: , where, represents the weight coefficient corresponding to the pixel point , represents the Euclidean distance from the pixel point after Gaussian weighting to the center point of the filtering window, and represents the smoothing parameter of the weighted kernel function; , where, represents the weighted average value of all pixel points in the filtering window, represents the weight coefficient of the pixel point , represents the pixel value of the pixel point , represents the length of the filtering window; S414. Repeat the steps in S411 to S413 until the filtering window traverses all pixel points in the real-time image data of the drone patrol airspace to generate the real-time image feature data of the flying objects in the patrol airspace .

[0012] Performing image data noise reduction processing on the real-time image data of the drone patrol airspace through a weighted average filtering method can effectively avoid the influence of noise in the image data, highlight the image features in the acquired data, and provide a reliable data basis for the flying object recognition process.

[0013] Preferably, perform flying object recognition processing based on the flying object image recognition model and the real-time image feature data of the drone's patrol airspace to generate flying object recognition data for the patrol airspace. If the flying object recognition data for the patrol airspace indicates the existence of a flying object, generate flying object image feature data for the patrol airspace and enter S6; otherwise, the specific steps for not performing any operation are as follows: S51. Input the real-time image feature data of the drone's patrol airspace into the flying object image recognition model for flying object recognition processing to generate flying object recognition data for the patrol airspace ; If the flying object image recognition model does not recognize the existence of a flying object in the real-time image feature data of the drone's patrol airspace then output that the flying object recognition data for the patrol airspace indicates the non-existence of a flying object, and the drone will continue to perform airspace patrol operations according to the preset patrol route; If the flying object image recognition model recognizes the existence of a flying object in the real-time image feature data of the drone's patrol airspace then output that the flying object recognition data for the patrol airspace indicates the existence of a flying object, and the flying object image recognition model will extract flying object image feature data from the real-time image feature data of the drone's patrol airspace to generate flying object image feature data for the patrol airspace and enter S6.

[0014] Perform a preliminary analysis on the real-time image feature data of the drone's patrol airspace through the flying object image recognition model. If a flying object is recognized, accurately extract the image feature data of the flying object from the real-time image feature data of the drone's patrol airspace.

[0015] Preferably, establish legal access to flying object image feature data, perform flying object feature matching processing on the flying object image feature data of the patrol airspace and the legal access flying object image feature data to generate flying object access analysis data for the patrol airspace. If the flying object access analysis data for the patrol airspace indicates legal access, no operation is performed; otherwise, the specific steps for generating the real-time position coordinate data of the patrol drone and entering S7 are as follows: S61. Establish a dataset of legal access flying object image features , where represents the image feature data of the th legal access flying object stored in the drone airspace patrol supervision platform, and S62. Use the jellyfish optimization algorithm to perform flying object feature matching processing on the flying object image feature data in the inspected airspace and the legal access flying object image feature data in the legal access flying object image feature data set to generate flying object access analysis data for the inspected airspace ; S621. Construct a legal access flying object search jellyfish population, and set the population size to , the current iteration number to , the maximum iteration number to and the dimension of the legal access flying object image feature data search space to ; Use the legal access flying object image feature data set as the legal access flying object image feature data search space, and randomly generate pieces of legal access flying object image feature data in the legal access flying object image feature data search space. Each piece of legal access flying object image feature data corresponds to a legal access flying object search jellyfish individual in the legal access flying object search jellyfish population; S622. Calculate the fitness value of each legal access flying object search jellyfish individual in the legal access flying object search jellyfish population, arrange each legal access flying object search jellyfish individual in the legal access flying object search jellyfish population from largest to smallest according to the fitness value, and select the legal access flying object search jellyfish individual with the highest fitness value as the current optimal individual; the fitness value calculation formula is as follows: , where represents the fitness value of the th legal access flying object search jellyfish individual in the legal access flying object search jellyfish population, represents the value of the rd dimension of the -dimensional feature vector of the legal access flying object image feature data corresponding to the th legal access flying object search jellyfish individual in the legal access flying object search jellyfish population, represents the value of the -dimensional feature vector of the flying object image feature data in the inspected airspace at the rd dimension, represents the correction value; S623. The movement behavior of each legal access flying object search jellyfish individual in the legal access flying object search jellyfish population in the legal access flying object image feature data search space is controlled by the time control function ; the time control function formula is as follows: , Among them, represents the time control function value during the -th iteration of the legal access flying object searching jellyfish population, represents a random number uniformly distributed between (0, 1); represents the maximum number of iterations; If , then each legal access flying object searching jellyfish individual in the legal access flying object searching jellyfish population will update its position following the ocean current movement direction in the legal access flying object image feature data search space; the position update formula is as follows: , Among them, represents the position after the -th legal access flying object searching jellyfish individual in the legal access flying object searching jellyfish population updates its position, represents the current position of the -th legal access flying object searching jellyfish individual in the legal access flying object searching jellyfish population, represents the position of the current optimal individual, represents the average position of the legal access flying object searching jellyfish population, and both represent random numbers uniformly distributed between (0, 1), represents the jellyfish position distribution coefficient; If and , then each legal access flying object searching jellyfish individual in the legal access flying object searching jellyfish population will adopt a passive movement strategy to randomly walk and update its position in the legal access flying object image feature data search space; the position update formula is as follows: , Among them, and both represent random numbers uniformly distributed between (0, 1), and respectively represent the upper search limit and the lower search limit of the search space, represents the jellyfish movement coefficient; If and , then each legal access flying object searching jellyfish individual in the legal access flying object searching jellyfish population will adopt an active movement strategy to randomly select a legal access flying object searching jellyfish individual in the legal access flying object image feature data search space and determine the movement direction to update its position; the position update formula is as follows: , Among them, and both represent random numbers that follow a uniform distribution between (0, 1), represents the movement direction of the th legitimate access flying object search jellyfish individual in the legitimate access flying object search jellyfish population adopting an active movement strategy; S624. Calculate the fitness value of each legitimate access flying object search jellyfish individual in the legitimate access flying object search jellyfish population after position update. If the fitness value of a legitimate access flying object search jellyfish individual after position update is greater than the fitness value of the original position, replace the original position with the new position; otherwise, retain the original position; Re - arrange all the legitimate access flying object search jellyfish individuals in the legitimate access flying object search jellyfish population from largest to smallest according to the fitness value, and select the legitimate access flying object search jellyfish individual with the highest fitness value as the new current optimal individual; S625. Judge whether the current iteration number is less than the maximum iteration number . If the current iteration number is less than the maximum iteration number , then increment the current iteration number by 1 and return to S623; otherwise, take the current optimal individual as the global optimal solution; S626. Set a fitness threshold. If the fitness value of the global optimal solution is greater than or equal to the fitness threshold, it means that the flying object found by the target UAV in the patrol airspace is a legitimate access flying object, and output the flight object access analysis data for legitimate access, and the UAV will continue to perform airspace patrol operations according to the preset patrol route; if the fitness value of the global optimal solution is less than the fitness threshold, it means that the flying object found by the target UAV in the patrol airspace is an illegal access flying object, and output the flight object access analysis data for illegal access, locate the real - time position of the UAV through the position sensor installed on the UAV, and generate the real - time position coordinate data of the patrol UAV .

[0016] The jellyfish optimization algorithm performs flying object feature matching processing on the flight object image feature data in the patrol airspace and the legitimate access flying object image feature data, which improves the speed and efficiency of the matching process, ensures the accuracy of the matching result, and scientifically analyzes whether the flying object appearing in the patrol airspace belongs to a legitimate access flying object according to the matching result. When it does not belong to a legitimate access flying object, it accurately locates the position of the UAV. At the same time, the jellyfish optimization algorithm has good robustness, which can ensure the stability of the matching process.

[0017] Preferably, the specific steps of constructing the UAV patrol supervision data based on the real-time image feature data of the UAV patrol airspace, the image feature data of the flying objects in the patrol airspace, the access analysis data of the flying objects in the patrol airspace, and the real-time position coordinate data of the patrol UAV, and pushing it to the UAV airspace patrol supervision platform are as follows: S71. Combine the real-time image feature data of the UAV patrol airspace, the image feature data of the flying objects in the patrol airspace, the access analysis data of the flying objects in the patrol airspace, and the real-time position coordinate data of the patrol UAV to construct the UAV patrol supervision data ; S72. Push the UAV patrol supervision data to the UAV airspace patrol supervision platform through a wireless communication network and issue an alarm for the intrusion of abnormal flying objects.

[0018] The present invention further includes an intelligent supervision system for airspace prevention and patrol based on UAVs, including a module for collecting image feature data of UAV patrol airspace samples, a module for constructing a flying object image recognition model, a module for collecting real-time image data of UAV patrol airspace, a module for preprocessing real-time image data of UAV patrol airspace, a module for identifying flying objects in the patrol airspace, a module for analyzing the access of flying objects in the patrol airspace, and a module for constructing UAV patrol supervision data; The module for collecting image feature data of UAV patrol airspace samples collects, through the UAV airspace patrol supervision platform, image feature data taken by a number of UAVs during airspace patrol operations to obtain image feature data of UAV patrol airspace samples; The module for constructing a flying object image recognition model trains and tests an initial convolutional neural network model based on the image feature data of UAV patrol airspace samples to obtain a flying object image recognition model When the target UAV performs airspace patrol operations in the target prevention airspace according to the preset patrol route, the module for collecting real-time image data of UAV patrol airspace regularly takes real-time image data of the target prevention airspace through an image capturing device installed on the target UAV to generate real-time image data of UAV patrol airspace; The module for preprocessing real-time image data of UAV patrol airspace performs image data noise reduction processing on the real-time image data of UAV patrol airspace through a weighted average filtering method to generate real-time image feature data of flying objects in the patrol airspace; The module for identifying flying objects in the patrol airspace inputs the real-time image feature data of UAV patrol airspace into the flying object image recognition model for flying object recognition processing to generate flying object recognition data in the patrol airspace; The flight object access analysis module for the inspected airspace performs flight object feature matching on the flight object image feature data of the inspected airspace and the established legal access flight object image feature data through the jellyfish optimization algorithm, and generates flight object access analysis data for the inspected airspace; The UAV inspection and supervision data construction module combines the real-time image feature data of the UAV inspection airspace, the flight object image feature data of the inspected airspace, the flight object access analysis data of the inspected airspace, and the real-time position coordinate data of the inspection UAV to construct UAV inspection and supervision data, which is pushed to the UAV airspace inspection and supervision platform through a wireless communication network, and an abnormal flight object intrusion alarm is issued.

[0019] (III) Beneficial effects 1. The present invention constructs a flight object image recognition model based on the collected UAV inspection airspace sample image feature data. Based on the flight object image recognition model, it scientifically analyzes whether there are flight objects in the real-time image data of the UAV inspection airspace after data preprocessing. If not, the UAV continues to perform inspection operations; otherwise, it generates flight object image feature data for the inspected airspace and matches it with the preset legal access flight object image feature data to accurately identify whether the flight objects appearing in the inspected airspace belong to legal access. If not, it constructs UAV inspection and supervision data, pushes it to the UAV airspace inspection and supervision platform, and issues an abnormal flight object intrusion alarm, realizing intelligent supervision of UAV airspace inspection and timely early warning when abnormal flight objects are found; 2. By online collecting UAV inspection airspace sample image feature data through the UAV airspace inspection and supervision platform and training and testing the initial convolutional neural network model, a flight object image recognition model is obtained, providing an accurate and reliable flight object recognition tool for airspace inspection operations, improving the speed and efficiency of the flight object recognition process, and ensuring the timeliness of the flight object recognition results; 3. The image data of the real-time image data of the UAV inspection airspace is denoised by the weighted average filtering method, effectively avoiding the influence brought by the noise in the image data, highlighting the image features in the obtained data, and providing a reliable data basis for the flight object recognition process; through the flight object image recognition model, the real-time image feature data of the UAV inspection airspace is preliminarily analyzed. If a flight object is recognized, the image feature data of the flight object is accurately extracted from the real-time image feature data of the UAV inspection airspace; 4. The jellyfish optimization algorithm is used to perform feature matching processing on the image feature data of the flying objects in the inspected airspace and the image feature data of the legally accessed flying objects, which improves the speed and efficiency of the matching process, ensures the accuracy of the matching result, and scientifically analyzes whether the flying objects appearing in the inspected airspace belong to the legally accessed flying objects according to the matching result. When they do not belong to the legally accessed flying objects, the position of the UAV is accurately located. At the same time, the jellyfish optimization algorithm has good robustness and can ensure the stability of the matching process. Brief Description of the Drawings

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0021] Figure 1 It is a flowchart of an intelligent supervision method for airspace prevention and inspection based on UAV provided by the present invention; Figure 2 It is a schematic diagram of the modules of an intelligent supervision system for airspace prevention and inspection based on UAV provided by the present invention. Detailed Embodiments

[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the protection scope of the present invention.

[0023] In the description of the present invention, it should be understood that the terms "opening", "upper", "lower", "top", "middle", "inner", etc. indicating orientation or positional relationships are only for the convenience of describing the invention and simplifying the description, rather than indicating or implying that the components or elements referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation of the invention.

[0024] The first embodiment is as follows: Please refer to Figure 1 , an intelligent supervision method for airspace prevention and inspection based on UAV, including the following steps: S1. Collect the image feature data of the UAV inspected airspace samples; S11. Online collect the image feature data taken by several UAVs during airspace inspection operations through the UAV airspace inspection and supervision platform to obtain the UAV inspected airspace sample image feature data set , where represent the th feature data of the sample images of the UAV patrol airspace, represent the total number of the feature data of the sample images of the UAV patrol airspace. The image feature data includes the image feature data with flying objects and the image feature data without flying objects.

[0025] S2. Build a flying object image recognition model based on the feature data of the sample images of the UAV patrol airspace; S21. Set the training data ratio and the test data ratio, and divide the dataset of the feature data of the sample images of the UAV patrol airspace according to the training data ratio and the test data ratio to obtain the training dataset of the feature data of the sample images of the UAV patrol airspace and the test dataset of the feature data of the sample images of the UAV patrol airspace respectively; S22. Build an initial convolutional neural network model, and set the number of input nodes of the initial convolutional neural network model to be , the number of hidden nodes to be , the number of output nodes to be , the initial weight to be and the initial bias to be ; S23. Set the training error threshold and the maximum number of training times, and input the training data of the feature data of the sample images of the UAV patrol airspace in the training dataset of the feature data of the sample images of the UAV patrol airspace into the initial convolutional neural network model for training. If the training error in the training process is less than the training error threshold or the number of training times is greater than the maximum number of training times, stop training to obtain the trained convolutional neural network model; otherwise, continue training until the training error in the training process is less than the training error threshold or the number of training times is greater than the maximum number of training times; S24. Set the test accuracy threshold, input the test data of the feature data of the sample images of the UAV patrol airspace in the test dataset of the feature data of the sample images of the UAV patrol airspace into the trained convolutional neural network model for testing, calculate the test accuracy. When the test accuracy is greater than or equal to the test accuracy threshold, regard the trained convolutional neural network model as the flying object image recognition model; otherwise, use the grid search algorithm to optimize the parameters of the trained convolutional neural network model to obtain the flying object image recognition model.

[0026] S3. Collect the real-time image data of the UAV patrol airspace; S31. Online collect the preset inspection route of the target prevention airspace through the UAV airspace inspection and supervision platform, operate the target UAV to perform airspace inspection operations in the target prevention airspace according to the preset inspection route, and regularly capture the real-time image data of the target prevention airspace through the image capture device installed on the target UAV to generate the real-time image data of the UAV inspection airspace. The image capture device represents any one of a high-definition camera, a panoramic camera, and a low-light camera.

[0027] S4. Perform data preprocessing on the real-time image data of the UAV inspection airspace to generate the real-time image feature data of the UAV inspection airspace; S41. Perform image data noise reduction processing on the real-time image data of the UAV inspection airspace through the weighted average filtering method to generate the real-time image feature data of the flying objects in the inspection airspace ; The process of image data noise reduction is as follows: S411. Set the filtering window, select any pixel point in the real-time image data of the UAV inspection airspace and align the center point of the filtering window with this pixel point; S412. Calculate the weight coefficients of each pixel point in the filtering window through the weighted kernel function formula; The weighted kernel function formula is as follows: where represents the weight coefficient corresponding to the pixel point , represents the Euclidean distance from the pixel point after Gaussian weighting to the center point of the filtering window, represents the smoothing parameter of the weighted kernel function; S413. Calculate the weighted average value of all pixel points in the filtering window, and replace the pixel value of the pixel point corresponding to the center point of the current filtering window with the weighted average value; The calculation formula is as follows: where represents the weighted average value of all pixel points in the filtering window, represents the weight coefficient of the pixel point , represents the pixel value of the pixel point , represents the length of the filtering window; S414. Repeat the steps in S411 to S413 until the filtering window traverses the real-time image data of the UAV inspection airspace ​​After all the pixels in it, real-time image feature data of flying objects in the inspected airspace is generated .

[0028] S5. Perform flying object recognition processing based on the flying object image recognition model and the real-time image feature data of the drone's inspected airspace to generate flying object recognition data for the inspected airspace. If the flying object recognition data for the inspected airspace indicates the existence of a flying object, generate real-time image feature data of the flying object in the inspected airspace and proceed to S6; otherwise, no operation is performed; S51. Input the real-time image feature data of the drone's inspected airspace into the flying object image recognition model for flying object recognition processing to generate flying object recognition data for the inspected airspace ; If the flying object image recognition model fails to recognize the existence of a flying object in the real-time image feature data of the drone's inspected airspace , output the flying object recognition data for the inspected airspace as indicating the non-existence of a flying object, and the drone will continue to perform airspace inspection operations along the preset inspection route; If the flying object image recognition model recognizes the existence of a flying object in the real-time image feature data of the drone's inspected airspace , output the flying object recognition data for the inspected airspace as indicating the existence of a flying object. The flying object image recognition model will extract the flying object image feature data from the real-time image feature data of the drone's inspected airspace to generate real-time image feature data of the flying object in the inspected airspace , and proceed to S6.

[0029] S6. Establish legally accessible flying object image feature data, perform flying object feature matching processing on the real-time image feature data of the flying object in the inspected airspace and the legally accessible flying object image feature data to generate flying object access analysis data for the inspected airspace. If the flying object access analysis data for the inspected airspace indicates a legal access, no operation is performed; otherwise, generate real-time position coordinate data of the inspected drone and proceed to S7; S61. Establish a dataset of legally accessible flying object image features , where represents the image feature data of the th legally accessible flying object stored in the drone airspace inspection and supervision platform, represents the total number of legally accessible flying object image feature data; S62. Perform flying object feature matching processing on the real-time image feature data of the flying object in the inspected airspace and the legally accessible flying object image feature data in the dataset of legally accessible flying object image features through the jellyfish optimization algorithm to generate flying object access analysis data for the inspected airspace ; S621. Construct a legal access flying object to search for jellyfish populations, and set the population size to , the current iteration number to , the maximum iteration number to and the dimension of the search space for the image feature data of legal access flying objects to ; Take the legal access flying object image feature data set as the search space for the legal access flying object image feature data, and randomly generate legal access flying object image feature data in the search space for the legal access flying object image feature data. Each legal access flying object image feature data corresponds to a legal access flying object search jellyfish individual in the legal access flying object search jellyfish population; S622. Calculate the fitness value of each legal access flying object search jellyfish individual in the legal access flying object search jellyfish population, arrange the legal access flying object search jellyfish individuals in the legal access flying object search jellyfish population in descending order according to the fitness value, and select the legal access flying object search jellyfish individual with the highest fitness value as the current optimal individual; the fitness value calculation formula is as follows: , where represents the fitness value of the th legal access flying object search jellyfish individual in the legal access flying object search jellyfish population, represents the value of the th dimension of the -dimensional feature vector of the legal access flying object image feature data corresponding to the th legal access flying object search jellyfish individual in the legal access flying object search jellyfish population, represents the value of the th dimension of the -dimensional feature vector of the patrol airspace flying object image feature data, represents the correction value; S623. The movement behavior of each legal access flying object search jellyfish individual in the legal access flying object search jellyfish population in the search space for the legal access flying object image feature data is controlled by the time control function ; the time control function formula is as follows: , where represents the time control function value during the th iteration of the legal access flying object search jellyfish population, denotes a random number uniformly distributed between (0, 1); denotes the maximum number of iterations; If , then each legal access flying object search jellyfish individual in the legal access flying object search jellyfish population will update its position following the ocean current movement direction in the legal access flying object image feature data search space; the position update formula is as follows: , wherein, denotes the position of the th legal access flying object search jellyfish individual in the legal access flying object search jellyfish population after position update, denotes the current position of the th legal access flying object search jellyfish individual in the legal access flying object search jellyfish population, denotes the position of the current optimal individual, denotes the average position of the legal access flying object search jellyfish population, and both denote random numbers uniformly distributed between (0, 1), denotes the jellyfish position distribution coefficient; If and , then each legal access flying object search jellyfish individual in the legal access flying object search jellyfish population will adopt a passive movement strategy to randomly walk and update its position in the legal access flying object image feature data search space; the position update formula is as follows: , wherein, and both denote random numbers uniformly distributed between (0, 1), and respectively denote the search upper limit and search lower limit of the search space, denotes the jellyfish movement coefficient; If and , then each legal access flying object search jellyfish individual in the legal access flying object search jellyfish population will adopt an active movement strategy to randomly select a legal access flying object search jellyfish individual in the legal access flying object image feature data search space and determine the movement direction to update its position; the position update formula is as follows: , wherein, and both denote random numbers uniformly distributed between (0, 1), Indicates the movement direction of the th legal access flying object searching jellyfish individual in the jellyfish population adopting an active movement strategy; S624. Calculate the fitness value of each legal access flying object searching jellyfish individual in the legal access flying object searching jellyfish population after position update. If the fitness value of a legal access flying object searching jellyfish individual after position update is greater than the fitness value of the original position, replace the original position with the new position; otherwise, retain the original position; Re - arrange all the legal access flying object searching jellyfish individuals in the legal access flying object searching jellyfish population in descending order of fitness value, and select the legal access flying object searching jellyfish individual with the highest fitness value as the new current optimal individual; S625. Judge whether the current iteration number is less than the maximum iteration number . If the current iteration number is less than the maximum iteration number , then increment the current iteration number by 1 and return to S623; otherwise, take the current optimal individual as the global optimal solution; S626. Set a fitness threshold. If the fitness value of the global optimal solution is greater than or equal to the fitness threshold, it indicates that the flying object found by the target UAV in the patrol airspace is a legal access flying object, and output the flight object access analysis data for legal access, and the UAV will continue to perform airspace patrol operations along the preset patrol route; if the fitness value of the global optimal solution is less than the fitness threshold, it indicates that the flying object found by the target UAV in the patrol airspace is an illegal access flying object, and output the flight object access analysis data for illegal access, locate the real - time position of the UAV through the position sensor installed on the UAV, and generate the real - time position coordinate data of the patrol UAV .

[0030] S7. Based on the real - time image feature data of the UAV patrol airspace, the flight object image feature data of the patrol airspace, the flight object access analysis data of the patrol airspace, and the real - time position coordinate data of the patrol UAV, construct UAV patrol supervision data and push it to the UAV airspace patrol supervision platform; S71. Combine the real - time image feature data of the UAV patrol airspace, the flight object image feature data of the patrol airspace, the flight object access analysis data of the patrol airspace, and the real - time position coordinate data of the patrol UAV to construct the UAV patrol supervision data ; S72. Transmit the UAV patrol supervision data through the wireless communication network Push it to the UAV airspace inspection and supervision platform and issue an alarm for the intrusion of abnormal flying objects.

[0031] The second embodiment is as follows: Please refer to Figure 2 , an intelligent supervision system for airspace prevention and inspection based on UAVs, including a UAV inspection airspace sample image feature data acquisition module, a flying object image recognition model construction module, a UAV inspection airspace real-time image data acquisition module, a UAV inspection airspace real-time image data preprocessing module, an inspection airspace flying object recognition module, an inspection airspace flying object access analysis module, and a UAV inspection supervision data construction module; The UAV inspection airspace sample image feature data acquisition module online acquires image feature data of several images taken by a UAV during airspace inspection operations through the UAV airspace inspection and supervision platform to obtain UAV inspection airspace sample image feature data; The flying object image recognition model construction module trains and tests an initial convolutional neural network model based on the UAV inspection airspace sample image feature data to obtain a flying object image recognition model When the target UAV performs airspace inspection operations in the target prevention airspace according to the preset inspection route, the UAV inspection airspace real-time image data acquisition module regularly captures real-time image data of the target prevention airspace through the image capture device installed on the target UAV to generate UAV inspection airspace real-time image data; The UAV inspection airspace real-time image data preprocessing module performs image data noise reduction processing on the UAV inspection airspace real-time image data through a weighted average filtering method to generate inspection airspace flying object real-time image feature data; The inspection airspace flying object recognition module inputs the UAV inspection airspace real-time image feature data into the flying object image recognition model for flying object recognition processing to generate inspection airspace flying object recognition data; The inspection airspace flying object access analysis module performs flying object feature matching processing on the inspection airspace flying object image feature data and the established legal access flying object image feature data through a jellyfish optimization algorithm to generate inspection airspace flying object access analysis data; The UAV inspection supervision data construction module combines the UAV inspection airspace real-time image feature data, the inspection airspace flying object image feature data, the inspection airspace flying object access analysis data, and the inspection UAV real-time position coordinate data to construct UAV inspection supervision data, which is pushed to the UAV airspace inspection and supervision platform through a wireless communication network and an alarm for the intrusion of abnormal flying objects is issued.

[0032] In the description of this specification, the descriptions referring to the terms "one embodiment", "example", "specific example", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0033] The preferred embodiments of the invention disclosed above are only used to help explain the invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification in order to better explain the principles and practical applications of the invention, so that those skilled in the art can well understand and utilize the invention.

Claims

1. An intelligent supervision method for airspace prevention and inspection based on unmanned aerial vehicles, characterized in that, It includes the following steps: S1. Collect the characteristic data of the sample images in the airspace patrolled by the drone; S2. Build an image recognition model for flying objects based on the characteristic data of the sample images in the airspace patrolled by the drone; S3. Collect the real-time image data in the airspace patrolled by the drone; S4. Perform data preprocessing on the real-time image data in the airspace patrolled by the drone to generate the characteristic data of the real-time images in the airspace patrolled by the drone; S5. Perform flying object recognition processing according to the flying object image recognition model and the characteristic data of the real-time images in the airspace patrolled by the drone to generate the recognition data of the flying objects in the patrolled airspace. If the recognition data of the flying objects in the patrolled airspace indicates the existence of flying objects, generate the characteristic data of the flying object images in the patrolled airspace and proceed to S6; otherwise, no operation is performed; S6. Establish the legal access flying object image characteristic data, perform flying object characteristic matching processing on the characteristic data of the flying object images in the patrolled airspace and the legal access flying object image characteristic data to generate the access analysis data of the flying objects in the patrolled airspace. If the access analysis data of the flying objects in the patrolled airspace indicates legal access, no operation is performed; otherwise, generate the real-time position coordinate data of the patrolling drone and proceed to S7; S7. Build the drone patrol supervision data based on the characteristic data of the real-time images in the airspace patrolled by the drone, the characteristic data of the flying object images in the patrolled airspace, the access analysis data of the flying objects in the patrolled airspace, and the real-time position coordinate data of the patrolling drone, and push it to the drone airspace patrol supervision platform.

2. The intelligent supervision method for airspace prevention and inspection based on unmanned aerial vehicles according to claim 1, characterized in that, The S1 includes the following steps: S11. Online collect the image feature data captured by several drones during airspace inspection operations through the drone airspace inspection and supervision platform to obtain the drone inspection airspace sample image feature data set , where represents the th drone inspection airspace sample image feature data, represents the total number of drone inspection airspace sample image feature data.

3. The intelligent supervision method for airspace prevention and inspection based on unmanned aerial vehicles according to claim 2, characterized in that, The S2 includes the following steps: S21. Set the training data ratio and the test data ratio, and divide the characteristic data set of the sample images in the airspace patrolled by the drone according to the training data ratio and the test data ratio to obtain the training data set of the characteristic data of the sample images in the airspace patrolled by the drone and the test data set of the characteristic data of the sample images in the airspace patrolled by the drone respectively; S22. Build an initial convolutional neural network model; S23. Set the training error threshold and the maximum number of training times, and input the training data of the characteristic data of the sample images in the airspace patrolled by the drone in the training data set of the characteristic data of the sample images in the airspace patrolled by the drone into the initial convolutional neural network model for training. If the training error during the training process is less than the training error threshold or the number of training times is greater than the maximum number of training times, stop the training and obtain the trained convolutional neural network model; otherwise, continue the training until the training error during the training process is less than the training error threshold or the number of training times is greater than the maximum number of training times; S24. Set the test accuracy threshold, input the test data of the characteristic data of the sample images in the airspace patrolled by the drone in the test data set of the characteristic data of the sample images in the airspace patrolled by the drone into the trained convolutional neural network model for testing, calculate the test accuracy. When the test accuracy is greater than or equal to the test accuracy threshold, use the trained convolutional neural network model as the flying object image recognition model; otherwise, use the grid search algorithm to optimize the parameters of the trained convolutional neural network model to obtain the flying object image recognition model.

4. The intelligent supervision method for airspace prevention and inspection based on unmanned aerial vehicles according to claim 3, characterized in that, The S3 includes the following steps: S31. Online collect the preset inspection route of the target prevention airspace through the UAV airspace inspection and supervision platform, operate the target UAV to perform airspace inspection operations in the target prevention airspace according to the preset inspection route, and regularly capture the real-time image data of the target prevention airspace through the image capturing device installed on the target UAV to generate the real-time image data of the UAV inspection airspace .

5. The intelligent supervision method for airspace prevention and inspection based on unmanned aerial vehicles according to claim 4, characterized in that, The S4 includes the following steps: S41. Perform image data noise reduction processing on the real-time image data of the airspace inspected by the UAV through the weighted average filtering method to generate real-time image feature data of flying objects in the inspected airspace .

6. The intelligent supervision method for airspace prevention and inspection based on unmanned aerial vehicles according to claim 5, characterized in that, The said S5 includes the following steps: S51. Input the into the flying object image recognition model for flying object recognition processing to generate flying object recognition data for the patrol airspace ; If the flying object image recognition model fails to recognize the existence of a flying object in the , output that the indicates the absence of a flying object, and the drone will continue to perform airspace inspection operations along the preset inspection route; If the flying object image recognition model recognizes that there is a flying object in the , it outputs that the has a flying object. The flying object image recognition model extracts flying object image feature data from the real-time image feature data of the airspace patrolled by the drone, and generates flying object image feature data for the patrolled airspace , and enters S6.

7. The intelligent supervision method for airspace prevention and inspection based on unmanned aerial vehicles according to claim 6, characterized in that, The said S6 includes the following steps: S61. Establish a legal access dataset of UAV image features , where represents the image feature data of the th legal access UAV in the UAV airspace inspection and supervision platform, represents the total number of legal access UAV image feature data; S62. Using the jellyfish optimization algorithm to perform flight object feature matching processing on the legal access flight object image feature data in the and the to generate flight object access analysis data for the inspected airspace .

8. The intelligent supervision method for airspace prevention and inspection based on unmanned aerial vehicles according to claim 7, characterized in that, The said S62 includes the following steps: S621. Construct a legal access to search for jellyfish populations by flying objects, and set the population size to , the current iteration number is , the maximum iteration number is and the dimension of the search space for the image feature data of legal access flying objects is ; Taking the as the legal access flying object image feature data search space, randomly generate legal access flying object image feature data in the legal access flying object image feature data search space, and each legal access flying object image feature data corresponds to a legal access flying object search jellyfish individual; S622. Calculate the fitness values of each legal access flying object search jellyfish individual, arrange each legal access flying object search jellyfish individual in descending order according to the fitness value, and select the legal access flying object search jellyfish individual with the highest fitness value as the current optimal individual; S623. The movement behaviors of each legal access flying object search jellyfish individual in the legal access flying object image feature data search space among the legal access flying object search jellyfish population are controlled by a time control function . If , each legally accessible flying object searches for jellyfish individuals and updates their positions following the direction of ocean current movement in the search space of the image feature data of the legally accessible flying objects; If and , then each legal access flying object search jellyfish individual will adopt a passive movement strategy to randomly walk in the legal access flying object image feature data search space for position update; If and , then each legal access flying object search jellyfish individual will adopt an active motion strategy to randomly select a legal access flying object search jellyfish individual in the legal access flying object image feature data search space, and determine the motion direction for position update; Among them, and both represent random numbers subject to a uniform distribution between (0, 1); S624. Calculate the fitness values of each legal access flying object search jellyfish individual after position update. If the fitness value of a legal access flying object search jellyfish individual after position update is greater than the fitness value of the original position, replace the original position with the new position; otherwise, retain the original position; Rearrange each legal access flying object search jellyfish individual in descending order according to the fitness value again, and select the legal access flying object search jellyfish individual with the highest fitness value as the new current optimal individual; S625. Judgment Is it less than ? If is less than , then increment by 1 and return to S623; otherwise, take the current optimal individual as the global optimal solution; S626. Set a fitness threshold. If the fitness value of the global optimal solution is greater than or equal to the fitness threshold, then output the as a legal access, and the drone will continue to perform airspace inspection operations along the preset inspection route; if the fitness value of the global optimal solution is less than the fitness threshold, then output the as an illegal access, locate the real-time position of the drone through the position sensor installed on the drone, and generate real-time position coordinate data of the inspection drone .

9. The intelligent supervision method for airspace prevention and inspection based on unmanned aerial vehicles according to claim 8, characterized in that, The said S7 includes the following steps: S71. Combine the , the , the and the to perform data combination and construct the UAV inspection and supervision data ; S72. Push the through the wireless communication network to the UAV airspace inspection and supervision platform, and issue an alarm for the intrusion of abnormal flying objects. ​ 10. A system of the intelligent supervision method for airspace prevention and inspection based on unmanned aerial vehicles, characterized in that, Implement the intelligent supervision method for airspace prevention and inspection based on drones as described in any one of claims 1-9.

Images