A drone identification method and system for airspace monitoring

Through the cascading processing of deep learning cascading network units, the problem of low drone recognition and positioning accuracy is solved, and fast and accurate drone recognition and counter assistance is achieved, which improves recognition efficiency and accuracy, and reduces the labor costs.

CN118038219BActive Publication Date: 2024-11-15HUNAN HOWAT DEFENSE TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202311805544.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-12-26
Publication Date
2024-11-15
Estimated Expiration
2043-12-26

AI Technical Summary

Technical Problem

The existing radio interference countermeasures have problems such as untimely identification and positioning, incorrect identification and low positioning accuracy in drone identification and positioning, which affects the countermeasures and may cause electromagnetic environment pollution.

Method used

Deep learning cascading global convolution recognition network unit and local convolution recognition network unit are adopted to achieve fast and accurate drone recognition results through cascading processing of fuzzy recognition network units and precisely identifying network units. Get it.

Benefits of technology

It improves the accuracy and efficiency of drone identification, reduces labor costs and judgment time, realizes automated drone screening and counter assistance, and reduces the complexity of human-computer interaction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118038219B_ABST
    Figure CN118038219B_ABST
Patent Text Reader

Abstract

The present invention provides a method and system for drone identification for airspace monitoring, which imports the SAR map of the airspace scene to be identified into a fuzzy identification network unit to obtain candidate drone identification results; then the candidate drone identification results and the SAR map of the airspace scene to be identified are imported layer by layer into the precise identification network unit to obtain multiple intermediate drone identification results; finally, the intermediate drone identification results are filtered layer by layer to obtain accurate drone identification results; the intermediate drone identification results are local SAR maps. The present invention can quickly and accurately obtain drone inference results, improve screening efficiency while ensuring screening accuracy, and can better achieve a balance between accuracy and time efficiency, providing a basis for the calculation of drone-related indicators.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of airspace scene SAR image processing, and in particular to a drone identification method and system for airspace monitoring. Background Art

[0002] In recent years, China's drone industry has developed rapidly, and its application scenarios have been continuously broadened and deepened. However, the surge in the number of drones and the imperfection of their regulatory system have also led to a series of problems such as the leakage of personal privacy and confidential information, posing a serious threat to social and military security. Therefore, it is crucial to achieve countermeasures against low-altitude drone targets.

[0003] Compared with countermeasures such as laser weapons and nets, radio jamming can be used to counter without damaging drones, and has the advantages of low implementation difficulty and low cost. Most existing radio jamming countermeasures rely on manual or optoelectronic equipment to detect and locate drone targets, and then use jamming systems to counter. Once the drone is not discovered in time, misidentified, or the positioning accuracy is not high, it will affect the countermeasure effect and cause electromagnetic environmental pollution. Therefore, there is an urgent need for an automated drone countermeasure method that can quickly and accurately detect and locate drones and perform directional jamming. Summary of the invention

[0004] The purpose of the present invention is to provide a drone identification method and system for airspace monitoring, which can quickly and accurately obtain drone screening inference results by cascading global convolutional recognition network units and local convolutional recognition network units through deep learning. While ensuring the screening accuracy, the screening efficiency can be improved, thereby providing a basis for the calculation of drone centerline related indicators and effectively countering drones.

[0005] To achieve the above object, the present invention provides a method for identifying drones for airspace monitoring, comprising:

[0006] Import the SAR image of the airspace scene to be identified into the fuzzy recognition network unit to obtain the candidate drone identification result;

[0007] Importing the candidate drone identification results and the SAR image of the airspace scene to be identified into the precise identification network unit layer by layer to obtain multiple intermediate drone identification results;

[0008] Filtering the intermediate drone identification results layer by layer to obtain accurate drone identification results;

[0009] Among them, the SAR map of the airspace scene to be identified, the candidate drone identification result and the accurate drone identification result are all global SAR maps; the intermediate drone identification result is a local SAR map.

[0010] Furthermore, before the candidate drone identification results and the to-be-identified airspace scene SAR images are imported layer by layer into the accurate identification network unit to obtain a plurality of intermediate drone identification results, the method further includes:

[0011] The first screening process is performed on the candidate identification SAR image of the UAV identified by the fuzzy identification network unit to obtain the candidate UAV identification result; the first screening process includes:

[0012] For each POI in the candidate drone identification SAR map, obtain the drone environment identification accuracy and non-drone environment identification accuracy of the POI; determine whether the drone environment identification accuracy is greater than or equal to the non-drone environment identification accuracy, and if so, set the POI value of the POI to the drone POI value; if not, set the POI value of the POI to the non-drone POI value to obtain the candidate drone identification result;

[0013] The candidate identification SAR image of the UAV is a SAR image containing a flying object;

[0014] Each POI in the candidate drone identification SAR image is a flying object in the corresponding SAR image.

[0015] Furthermore, filtering the intermediate drone identification results layer by layer to obtain accurate drone identification results includes:

[0016] Filtering the intermediate drone identification results layer by layer to obtain a third drone identification result; wherein the third drone identification result is a global SAR map;

[0017] The third UAV identification result is reconstructed and transformed to obtain the accurate UAV panoramic SAR image with the same scale as the SAR image of the airspace scene to be identified.

[0018] Furthermore, the fuzzy recognition network unit is a global convolutional recognition network unit, and the precise recognition network unit is a local convolutional recognition network unit.

[0019] Further, the fuzzy recognition network unit is modified by the following steps:

[0020] Step A10: obtaining a first original correction object, wherein the first original correction object includes a first original airspace scene correction SAR image and a first sequence SAR image corresponding to the first original airspace scene correction SAR image, wherein the first sequence SAR image is a first airspace scene SAR image in which the UAV is marked;

[0021] Step A20: Expanding the first original corrected object to obtain a first corrected object, wherein the first corrected object includes a first spatial scene corrected SAR image and a first sequence SAR image corresponding to the first spatial scene corrected SAR image;

[0022] Step A30: setting an initial value of a network element indicator of the fuzzy identification network element; and

[0023] Correcting the pre-built fuzzy identification network unit according to the first correction object and the initial value of the network unit indicator of the fuzzy identification network unit until a first correction end threshold is met;

[0024] Among them, the first original spatial scene corrected SAR map, the first sequence SAR map corresponding to the first original spatial scene corrected SAR map, the first spatial scene corrected SAR map, and the first sequence SAR map corresponding to the first spatial scene corrected SAR map are all global SAR maps.

[0025] Further, in step A30, the pre-built fuzzy identification network unit is corrected according to the first correction object and the initial value of the network unit indicator of the fuzzy identification network unit until the first correction end threshold is met, including:

[0026] Step A31: importing the first correction object into the fuzzy identification network unit, and obtaining the drone inference result of the first airspace scene correction SAR image according to the initial value of the network unit indicator of the fuzzy identification network unit;

[0027] Step A32: Calculate the value of a first cost function according to the drone inference result of the first airspace scene corrected SAR image and the first sequence SAR image corresponding to the first airspace scene corrected SAR image; and correct the network unit index of the fuzzy identification network unit according to the value of the first cost function and the first preliminary corrected deviation value to obtain a preliminary corrected fuzzy identification network unit;

[0028] Step A33: obtaining a first approval object, the first approval object comprising a first airspace scene approval SAR map and a second sequence SAR map corresponding to the first airspace scene approval SAR map, the second sequence SAR map being a second airspace scene SAR map with the drone marked; and importing the first approval object into the preliminarily corrected fuzzy recognition network unit to obtain a drone inference result of the first airspace scene approval SAR map;

[0029] Step A34: Calculate the value of the first cost function based on the UAV inference result of the first airspace scene approved SAR map and the second sequence SAR map corresponding to the first airspace scene approved SAR map; if the value of the first cost function regresses to the first deviation value or the number of corrections of the fuzzy identification network unit is greater than or equal to the first correction number, the correction ends; otherwise, correct the network unit indicator of the fuzzy identification network unit, and update the initial value of the network unit indicator of the fuzzy identification network unit to the corrected network unit indicator, and return to execute the step A31.

[0030] Further, the accurate identification of the network unit is corrected by the following steps:

[0031] Step B10: obtaining a second original corrected object, wherein the second original corrected object includes a second original airspace scene corrected SAR image and a third sequence SAR image corresponding to the second original airspace scene corrected SAR image; and importing the second original corrected object into a fuzzy recognition network unit to obtain a UAV inferred SAR image corresponding to the second original airspace scene corrected SAR image; the third sequence SAR image is a third airspace scene SAR image in which the UAV has been marked;

[0032] Step B20: respectively obtaining the second original airspace scene corrected SAR image and an axial sub-image of the UAV inferred SAR image corresponding to the second original airspace scene corrected SAR image in the Z-axis direction, to obtain a second airspace scene corrected SAR image;

[0033] Step B30: preprocessing the second spatial scene corrected SAR image to obtain a second corrected object, wherein the second corrected object includes the preprocessed second spatial scene corrected SAR image and a third sequence SAR image corresponding to the preprocessed second spatial scene corrected SAR image;

[0034] Step B40: Setting an initial value of a network element indicator that accurately identifies the network element; and

[0035] Modifying the pre-built accurate identification network unit according to the second modification object and the initial value of the network unit indicator of the accurate identification network unit until a second modification end threshold is met;

[0036] The third sequence SAR images corresponding to the preprocessed second spatial scene SAR image and the preprocessed second spatial scene corrected SAR image are both local SAR images.

[0037] Further, the pre-built accurate identification network unit is corrected according to the second correction object and the initial value of the network unit indicator of the accurate identification network unit until the second correction end threshold is met, including:

[0038] Step B41: importing the second correction object into the precise identification network unit, and obtaining the drone inference result of the second airspace scene corrected SAR image according to the initial value of the network unit indicator of the precise identification network unit;

[0039] Step B42: Calculate the value of the second cost function according to the drone inference result of the second airspace scene corrected SAR image and the third sequence SAR image corresponding to the second airspace scene corrected SAR image, and correct the network unit index of the accurately identified network unit according to the value of the second cost function and the second preliminary corrected deviation value to obtain a preliminary corrected accurately identified network unit;

[0040] Step B43: obtaining a second approval object, wherein the second approval object includes a second airspace scene approval SAR map and a fourth sequence SAR map corresponding to the second airspace scene approval SAR map, wherein the fourth sequence SAR map is a fourth airspace scene SAR map in which the drone is marked; and importing the second approval object into the preliminary corrected precise identification network unit to obtain a drone inference result of the second airspace scene approval SAR map;

[0041] Step B44: Calculate the value of the second cost function based on the UAV inference result of the second airspace scenario approved SAR map and the fourth sequence SAR map corresponding to the second airspace scenario approved SAR map; if the value of the second cost function regresses to the second deviation value or the number of corrections of the accurately identified network unit is greater than or equal to the second number of corrections, the correction ends; otherwise, correct the network unit indicator of the accurately identified network unit, and update the initial value of the network unit indicator of the accurately identified network unit to the corrected network unit indicator, and return to execute step B41.

[0042] According to a second aspect of the present invention, the present invention claims protection for a drone identification system for airspace monitoring, comprising:

[0043] The global SAR image screening module is configured to import the SAR image of the airspace scene to be identified into the fuzzy recognition network unit to obtain the candidate UAV identification result;

[0044] A local SAR image screening module is configured to import the candidate drone identification results and the airspace scene SAR image to be identified into the accurate identification network unit layer by layer to obtain multiple intermediate drone identification results;

[0045] A global SAR image reconstruction module is configured to filter the intermediate drone identification results layer by layer to obtain accurate drone identification results;

[0046] The SAR map of the airspace scene to be identified, the candidate drone identification result and the accurate drone identification result are all global SAR maps; the intermediate drone identification result is a local SAR map;

[0047] The drone identification system for airspace monitoring is used to execute the drone identification method for airspace monitoring.

[0048] Compared with the prior art, the drone identification method and system for airspace monitoring provided by the present invention has the following advantages:

[0049] The present invention provides a drone identification method for airspace monitoring, which includes importing a SAR map of an airspace scene to be identified into a fuzzy identification network unit to obtain a candidate drone identification result; then importing the candidate drone identification result and the airspace scene SAR map to be identified into a precise identification network unit layer by layer to obtain a plurality of intermediate drone identification results; finally, filtering the intermediate drone identification results layer by layer to obtain an accurate drone identification result; wherein the airspace scene SAR map to be identified, the candidate drone identification result and the accurate drone identification result are all global SAR maps; and the intermediate drone identification result is a local SAR map. Therefore, the panoramic drone screening method provided by the present invention first performs overall screening through the fuzzy recognition network unit, and then fine-tunes each layer of the screening results of the fuzzy recognition network unit through the precise recognition network unit, thereby realizing the automatic screening of panoramic drones; further, the drone identification method for airspace monitoring provided by the present invention can quickly and accurately obtain drone inference results through deep learning cascade global convolution recognition network units and local convolution recognition network units, and can improve the screening efficiency while ensuring the screening accuracy, and can better achieve the balance between accuracy and time efficiency; further, the drone inference results obtained by the drone identification method for airspace monitoring provided by the present invention can provide a basis for the calculation of drone-related indicators, and can significantly reduce the time and labor cost of pilot judgment. In addition, the drone identification method for airspace monitoring provided by the present invention can not only improve the overall algorithm accuracy, but also reduce the cumbersome operations of human-computer interaction, realize end-to-end algorithm flow, and can better assist pilots to improve judgment accuracy.

[0050] Since the drone identification system for airspace monitoring provided by the present invention and the drone identification method for airspace monitoring provided by the present invention belong to the same inventive concept, they have at least the same beneficial effects and will not be described in detail herein. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of the overall process of a drone identification method for airspace monitoring provided by one embodiment of the present invention;

[0052] Figure 2 A second flow chart of a method for identifying drones for airspace monitoring provided by one embodiment of the present invention;

[0053] Figure 3 A third flow chart of a method for identifying drones for airspace monitoring provided by one embodiment of the present invention;

[0054] Figure 4 A structural module diagram of a drone identification system for airspace monitoring provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0055] The following is a further detailed description of the drone identification method, system, electronic device and storage medium for airspace monitoring proposed by the present invention in conjunction with the accompanying drawings. According to the following description, the advantages and features of the present invention will be clearer. It should be noted that the drawings are in a very simplified form and use non-precise proportions, which are only used to conveniently and clearly assist in explaining the purpose of the implementation method of the present invention. In order to make the purposes, features and advantages of the present invention more obvious and easy to understand, please refer to the accompanying drawings. It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the threshold for the implementation of the present invention. Any structural modification, change in proportional relationship or size correction, when the effects and purposes that can be achieved by the present invention are the same or similar, should still fall within the scope of the technical content disclosed by the present invention.

[0056] It should be noted that the drone identification method for airspace monitoring of the embodiment of the present invention can be applied to the electronic device of the embodiment of the present invention, which can be a personal computer, a mobile terminal, etc. The mobile terminal can be a mobile phone, a tablet computer, and other hardware devices with various operating systems.

[0057] An important application of drones in the security field is reconnaissance and surveillance. By carrying various sensors and camera equipment, drones can effectively obtain real-time images and data of target areas. This information can help security personnel to identify and analyze potential threats in a timely manner so as to formulate corresponding security measures.

[0058] Drones are also widely used in border inspection and monitoring. Traditional patrol and monitoring tasks usually require a lot of manpower and material resources, while drones can cover a wider range through a high-altitude bird's-eye view, reduce labor costs, and improve monitoring efficiency. Drones can also detect and monitor nighttime activities or potential illegal intrusions by carrying infrared sensors and other equipment.

[0059] The application of drone technology in the detection and removal of explosives is also of great significance. Drones equipped with explosive detection sensors can quickly detect and locate suspicious explosives without touching the target object by analyzing data such as gas, heat and magnetism, providing security personnel with necessary detection information. Drones can also be used to dismantle and remove explosives, preventing security personnel from being exposed to potential threats.

[0060] With the rapid development of drone technology, anti-drone technology has also become one of the key areas of research in the security field. The core goal of anti-drone technology is to identify, interfere with and eliminate potential drone threats. The following are several common anti-drone technology applications:

[0061] Radio spectrum jamming is a common anti-drone technology application. By sending targeted electromagnetic signals, the wireless communications of drones can be disrupted, making it unable to receive commands or send back data, thereby interfering with or stopping the normal operation of drones.

[0062] Optical detection and tracking systems are also widely used in anti-drone technology. By using high-resolution camera equipment, infrared sensors and other technologies, security personnel can promptly detect and track potential intruding drones. These systems can also provide positioning information and target tracking support for countermeasures.

[0063] High-power microwave jamming technology is an effective anti-drone technology. By sending high-energy microwave signals, it can interfere with the electronic equipment of drones, causing communication interruption or operational failure, thus causing the drone to lose control or return.

[0064] In the field of security, the application of drone technology provides more convenience and possibilities, but it also brings some potential threats and risks. In addition to continuing to promote the development of drone technology, it is also necessary to accelerate the research and application of anti-drone technology to ensure safety and stability.

[0065] To realize the above idea, the present invention provides a drone identification method for airspace monitoring, please refer to Figure 1 , which schematically shows the overall flow diagram of a drone identification method for airspace monitoring provided by an embodiment of the present invention. Figure 1 It can be seen that the drone identification method for airspace monitoring provided in this embodiment includes:

[0066] Import the SAR image of the airspace scene to be identified into the fuzzy recognition network unit to obtain the candidate drone identification result;

[0067] Importing the candidate drone identification results and the SAR image of the airspace scene to be identified into the precise identification network unit layer by layer to obtain multiple intermediate drone identification results;

[0068] The intermediate drone recognition results are filtered layer by layer to obtain accurate drone recognition results.

[0069] Among them, the SAR map of the airspace scene to be identified, the candidate drone identification result and the accurate drone identification result are all global SAR maps; the intermediate drone identification result is a local SAR map.

[0070] Therefore, the drone identification method for airspace monitoring provided by the present invention first screens the interlayer as a whole through the fuzzy recognition network unit, and then fine-tunes each layer of the screening result of the fuzzy recognition network unit through the precise recognition network unit, thereby realizing the automatic screening of panoramic drones; further, the drone identification method for airspace monitoring provided by the present invention can quickly and accurately obtain drone inference results through deep learning cascade global convolution recognition network units and local convolution recognition network units, and can improve the screening efficiency while ensuring the screening accuracy, and can better achieve the balance between accuracy and time efficiency; further, the drone inference results obtained by the drone identification method for airspace monitoring provided by the present invention can provide a basis for the calculation of drone-related indicators, significantly reducing the time and labor cost of pilot judgment. In addition, the drone identification method for airspace monitoring provided by the present invention can not only improve the overall algorithm accuracy, but also reduce the cumbersome operations of human-computer interaction, realize end-to-end algorithm flow, and better assist pilots in improving judgment accuracy.

[0071] Specifically, in one preferred embodiment, before the SAR image of the airspace scene to be identified is imported into the fuzzy identification network unit to obtain the candidate drone identification result, the method further includes:

[0072] The panoramic SAR image is preprocessed to obtain the airspace scene SAR image to be identified; wherein the panoramic SAR image is obtained from the original airspace scene SAR image using a panoramic screening method.

[0073] Specifically, those skilled in the art should be able to understand that although the present invention is described by taking a panoramic global CTASAR image as an example, it is obvious that this is only an exemplary description and not a limitation of the present invention;

[0074] The size of the panoramic SAR image to be screened can be set according to specific circumstances, and the present invention is not limited to this. For example, the size of the airspace scene SAR image to be identified can be 512×512×130 POIs;

[0075] The POI is a feature point of interest in a geographic airspace.

[0076] Preferably, in one preferred embodiment, before importing the candidate drone identification results and the to-be-identified airspace scene SAR map layer by layer into the accurate identification network unit to obtain a plurality of intermediate drone identification results, the method further includes:

[0077] The first screening process is performed on the candidate identification SAR image of the UAV identified by the fuzzy identification network unit to obtain the candidate UAV identification result; the first screening process includes:

[0078] For each POI in the candidate drone identification SAR map, the drone environment identification accuracy and non-drone environment identification accuracy of the POI are obtained; it is determined whether the drone environment identification accuracy is greater than or equal to the non-drone environment identification accuracy, and if so, the POI value of the POI is set to the drone POI value; if not, the POI value of the POI is set to the non-drone POI value to obtain the candidate drone identification result.

[0079] The UAV candidate identification SAR image is a SAR image containing a flying object; each POI in the UAV candidate identification SAR image is a flying object in the corresponding SAR image.

[0080] Specifically, the first screening process is to process the similar parts of the inference results of drones and non-drones. Specifically, in one embodiment, for the same POI, it is assumed that the recognition accuracy of the drone environment is p1 and the recognition accuracy of the non-drone environment is p2. If p1≥p2, the overlapping POI is judged as a drone; otherwise, if p1<p2, the POI is judged as a non-drone. That is, when a POI may be a drone or a non-drone, the one with a higher probability is taken as its inference result, thereby further improving the accuracy of drone inference.

[0081] The panoramic SAR image (such as a SAR image) drone screening method provided by the present invention multiplies the mask and the drone and non-drone inference results, and uses the obtained mask SAR image to remove irrelevant areas of the drone and non-drone inference results, thereby limiting all inference results within the area.

[0082] Furthermore, filtering the intermediate drone identification results layer by layer to obtain accurate drone identification results includes:

[0083] Filtering the intermediate drone identification results layer by layer to obtain a third drone identification result; wherein the third drone identification result is a global SAR map;

[0084] The third UAV identification result is reconstructed and transformed to obtain the accurate UAV panoramic SAR image with the same scale as the SAR image of the airspace scene to be identified.

[0085] The drone identification method for airspace monitoring provided by the present invention fine-tunes each layer of the screening results through the precise identification network unit, thereby further improving the screening accuracy while realizing the automatic screening of the sandwich. At the same time, by reconstructing the third drone identification result, the accurate drone panoramic SAR image with the same scale as the airspace scene SAR image to be identified can be obtained, ensuring the uniformity of the airspace scene SAR image before and after screening.

[0086] Preferably, the fuzzy recognition network unit is a global convolution recognition network unit, and the precise recognition network unit is a local convolution recognition network unit. In order to facilitate understanding and avoid redundant description, the specific structures of the fuzzy recognition network unit and the precise recognition network unit will be described in detail below, and will not be described here for the time being.

[0087] Further, in one preferred embodiment, see Figure 2 , the fuzzy recognition network unit is modified by the following steps:

[0088] Step A10: obtaining a first original correction object, wherein the first original correction object includes a first original airspace scene correction SAR image and a first sequence SAR image corresponding to the first original airspace scene correction SAR image, wherein the first sequence SAR image is a first airspace scene SAR image in which the UAV is marked;

[0089] Step A20: Expanding the first original corrected object to obtain a first corrected object, wherein the first corrected object includes a first spatial scene corrected SAR image and a first sequence SAR image corresponding to the first spatial scene corrected SAR image;

[0090] Step A30: Set the initial value of the network unit indicator of the fuzzy identification network unit, and correct the pre-built fuzzy identification network unit according to the first correction object and the initial value of the network unit indicator of the fuzzy identification network unit until the first correction end threshold is met.

[0091] Among them, the first original spatial scene corrected SAR map, the first sequence SAR map corresponding to the first original spatial scene corrected SAR map, the first spatial scene corrected SAR map, and the first sequence SAR map corresponding to the first spatial scene corrected SAR map are all global SAR maps.

[0092] It should be noted that, as a preference, in order to ensure the generalization of the correction results of the deep learning network unit, the first original airspace scene correction SAR map should come from different regions. The obtained first original airspace scene correction SAR map is annotated, and the drone area position is marked in each layer of the first original airspace scene correction SAR map to obtain a first sequence of SAR maps. As those skilled in the art can understand, the present invention does not limit the specific annotation method of the first sequence of SAR maps, including but not limited to the standards in the prior art.

[0093] Further, those skilled in the art can understand that in step A20, since the data of the first original correction object is limited, and deep learning needs to be learned on certain data to have a certain robustness, in order to increase the robustness, data augmentation operations need to be performed to increase the generalization ability of the fuzzy recognition network unit. Specifically, the same random rigid transformation can be performed on the original airspace scene correction SAR image and the corresponding SAR image sequence of the unmanned aerial vehicle, including: rotation, scaling, translation, flipping and grayscale transformation. More specifically, the original airspace scene correction SAR image and the corresponding SAR image sequence of the unmanned aerial vehicle can be translated by -10 to 10 POIs, rotated by -30° to 30°, horizontally flipped, vertically flipped, scaled by 0.8 to 1.2 times, grayscale transformation, etc. to complete the data augmentation of the correction object. Through the above transformation, the original 20 objects, for example, can be expanded to 2000, 1600 of which can be used for network unit correction, and the remaining 400 objects can be used for network unit testing. It should be noted that since the random rigid transformation of the original airspace scene corrected SAR image will not affect its UAV results, the UAV's serial SAR image does not need to be transformed when the object is expanded, that is, the first sequence SAR images corresponding to multiple expanded airspace scene corrected SAR images obtained by performing different transformations on the same original airspace scene corrected SAR image are the same as the first sequence SAR images corresponding to the original airspace scene corrected SAR image.

[0094] Preferably, in order to improve the accuracy of the fuzzy recognition network unit, after the first correction object is generated and before the fuzzy recognition network unit is corrected, the first correction object may be preprocessed to remove noise in the SAR image and / or improve the correction efficiency of the fuzzy recognition network unit, thereby improving the SAR image quality of the first correction object. The preprocessing includes but is not limited to SAR image clipping, SAR image resampling, SAR image CT value truncation, and CT value normalization.

[0095] The SAR image resampling is to count the voxel spacing information of the data set, select the 50th percentile voxel spacing as the target voxel spacing, and interpolate all data to this voxel spacing. The SAR image CT value truncation is to count the POI values ​​of the environmental area, use the 0.5th percentile and 99.5th percentile of the POI value distribution as the upper and lower limits, and ensure that the CT value is within this range. The CT value normalization is to obtain a SAR image with a mean of 0 and a variance of 1.

[0096] In step A30, as those skilled in the art can understand, the network unit indicators of the fuzzy identification network unit include two categories: feature indicators and super indicators. Feature indicators are indicators used to learn the characteristics of drones, such as the POI value characteristics and location characteristics of drones. Super indicators are indicators set artificially during correction. Only by setting appropriate super indicators can feature indicators be learned from the object. Super indicators may include learning rate, the number of downsampling modules, the number of upsampling modules, the number of corrections, and the batch size of each correction. The learning rate can be regarded as the step size. For example, the present invention can set the learning rate to 0.001, the downsampling modules to 4, etc., the number of corrections to 10,000, and the batch size of each correction to 100, etc.

[0097] Preferably, in one exemplary embodiment, in step A30, the pre-built fuzzy identification network unit is corrected according to the first correction object and the initial value of the network unit indicator of the fuzzy identification network unit until the first correction end threshold is met, including:

[0098] Step A31: importing the first correction object into the fuzzy identification network unit, and obtaining the drone inference result of the first airspace scene correction SAR image according to the initial value of the network unit indicator of the fuzzy identification network unit;

[0099] Step A32: Calculate the value of a first cost function according to the drone inference result of the first airspace scene corrected SAR image and the first sequence SAR image corresponding to the first airspace scene corrected SAR image; and correct the network unit index of the fuzzy identification network unit according to the value of the first cost function and the first preliminary corrected deviation value to obtain a preliminary corrected fuzzy identification network unit;

[0100] Step A33: obtaining a first approval object, the first approval object comprising a first airspace scene approval SAR map and a second sequence SAR map corresponding to the first airspace scene approval SAR map, the second sequence SAR map being a second airspace scene SAR map with the drone marked; and importing the first approval object into the preliminarily corrected fuzzy recognition network unit to obtain a drone inference result of the first airspace scene approval SAR map;

[0101] Step A34: Calculate the value of the first cost function based on the UAV inference result of the first airspace scene approved SAR map and the second sequence SAR map corresponding to the first airspace scene approved SAR map; if the value of the first cost function regresses to the first deviation value or the number of corrections of the fuzzy identification network unit is greater than or equal to the first correction number, the correction ends; otherwise, correct the network unit indicator of the fuzzy identification network unit, and update the initial value of the network unit indicator of the fuzzy identification network unit to the corrected network unit indicator, and return to execute the step A31.

[0102] Preferably, in one exemplary embodiment, the first cost function is used to measure the area size of the similar part between the UAV inference result of the first airspace scene corrected SAR image and the first sequence SAR image corresponding to the first airspace scene corrected SAR image, and the area size of the similar part between the UAV inference result of the first airspace scene approved SAR image and the second sequence SAR image corresponding to the first airspace scene approved SAR image.

[0103] Specifically, in step A32, the value of the first cost function is calculated based on the drone inference result of the first airspace scene corrected SAR image and the first sequence SAR image corresponding to the first airspace scene corrected SAR image, that is, the area size of the similar part between the drone inference result of the first airspace scene corrected SAR image and the first sequence SAR image corresponding to the first airspace scene corrected SAR image (that is, the area size of the similar part between the inference result and the gold standard) is measured using the value of the first cost function. Preferably, the first cost function is expressed by the following formula:

[0104]

[0105] Where C represents the category (drone, non-drone), p c(i) represents the recognition accuracy of the fuzzy recognition network unit for the i-th voxel category, g c (i)∈{0,1} represents the sequence category (UAV, non-UAV) of the i-th voxel, and N is the number of voxels.

[0106] That is, the value of the first cost function is a comprehensive result of the size of the similar part area between the UAV inference result and the gold standard and the size of the similar part area between the non-UAV inference result and the gold standard.

[0107] Further, as can be understood by those skilled in the art, the purpose of correcting the identification network unit is to make the drone inference result obtained by the network unit close to the real drone, in other words, the error between the two is reduced to a certain range. Therefore, in step A32, the network unit index of the fuzzy identification network unit is corrected according to the value of the first cost function and the first preliminary correction deviation value to obtain a preliminary corrected fuzzy identification network unit. The correction end threshold of the preliminary corrected fuzzy identification network unit can be the deviation value of the inference result of the first airspace scene corrected SAR image in the first correction object and the corresponding first sequence SAR image regressed to the deviation value, and the deviation value should be set according to the actual situation, and the present invention is not limited to this.

[0108] Furthermore, as one of the preferred implementations, in step A32, the network unit index of the fuzzy identification network unit is corrected according to the value of the first cost function and the first preliminary correction deviation value to obtain a preliminary corrected fuzzy identification network unit, including using a stochastic gradient descent method to optimize the value of the first cost function. This is based on the fact that the correction process of the identification network unit is actually a process of minimizing the cost function, and taking the derivative can quickly and easily achieve this goal, and this method of taking the derivative is the gradient descent method. Therefore, using the stochastic gradient descent method to correct the fuzzy identification network unit can quickly and easily achieve the correction of the fuzzy identification network unit.

[0109] In addition, as those skilled in the art can understand, the correction process of the identification network unit is a multi-cycle correction process, therefore, in step A34, the correction can also be ended by setting the number of corrections. That is, the first correction end threshold can be met when the number of corrections reaches the number of corrections, such as 10,000 times, or the value of the first cost function calculated based on the drone inference result of the first airspace scene approved SAR map and the second sequence SAR map corresponding to the first airspace scene approved SAR map is less than or equal to the first deviation value. In particular, the first deviation value should be set according to the actual situation, and can be the same as the value of the error, or different.

[0110] In summary, the drone identification method for airspace monitoring provided by the present invention, when correcting the fuzzy identification network unit, first uses the first correction object to correct the pre-built fuzzy identification network unit to obtain a preliminary corrected fuzzy identification network unit; and then uses the first approval object to approve the preliminary corrected fuzzy identification network unit until the approval set loss no longer decreases or reaches the maximum correction round, thereby not only improving the correction speed of the fuzzy identification network unit, but also improving the inference accuracy of the fuzzy identification network unit.

[0111] See also Figure 3, which schematically shows a correction flow diagram of an accurate identification network unit (local deep learning network unit) provided by an embodiment of the present invention. Figure 3 It can be seen that, similar to the correction process of the fuzzy identification network unit, the precise identification network unit is corrected by the following steps:

[0112] Step B10: obtaining a second original corrected object, wherein the second original corrected object includes a second original airspace scene corrected SAR image and a third sequence SAR image corresponding to the second original airspace scene corrected SAR image; and importing the second original corrected object into a fuzzy recognition network unit to obtain a UAV inferred SAR image corresponding to the second original airspace scene corrected SAR image; the third sequence SAR image is a third airspace scene SAR image in which the UAV has been marked;

[0113] Step B20: respectively obtaining the second original airspace scene corrected SAR image and an axial sub-image of the UAV inferred SAR image corresponding to the second original airspace scene corrected SAR image in the Z-axis direction, to obtain a second airspace scene corrected SAR image;

[0114] Step B30: preprocessing the second spatial scene corrected SAR image to obtain a second corrected object, wherein the second corrected object includes the preprocessed second spatial scene corrected SAR image and a third sequence SAR image corresponding to the preprocessed second spatial scene corrected SAR image;

[0115] Step B40: setting an initial value of a network unit indicator for accurately identifying the network unit, and correcting the pre-built accurate identification network unit according to the second correction object and the initial value of the network unit indicator for accurately identifying the network unit until a second correction end threshold is met;

[0116] The third sequence SAR images corresponding to the preprocessed second spatial scene SAR image and the preprocessed second spatial scene corrected SAR image are both local SAR images.

[0117] First of all, it should be noted that, as a preference, the structure of the precise identification network unit (local deep learning network unit) is similar to that of the fuzzy identification network unit (global deep learning network unit). The difference is that, first, the three-dimensional convolution, three-dimensional group normalization, and three-dimensional deconvolution in all modules in the fuzzy identification network unit are all replaced by two-dimensional. Secondly, the local SAR map and the axial position sub-map of the inference result of the fuzzy identification network unit are spliced ​​and imported into the precise identification network unit in the channel dimension (Z-axis direction). Based on this, the basic principle of the correction method of the precise identification network unit is similar to that of the correction method of the fuzzy identification network unit. In order to avoid redundancy, only the difference between the correction of the precise identification network unit and the correction of the fuzzy identification network unit is explained. For the part not explained, technicians in this field can refer to the correction of the fuzzy identification network unit.

[0118] Specifically, the preprocessing of the second spatial scene corrected SAR map described in step B30 is similar to the correction of the fuzzy identification network unit. In order to improve the accuracy of the correction of the precise identification network unit, preferably, after obtaining the second spatial scene corrected SAR map, before the precise identification network unit correction, the second spatial scene corrected SAR map is preprocessed, and the preprocessing includes but is not limited to SAR map cropping, SAR map resampling, SAR map CT value truncation, and CT value normalization. Furthermore, unlike the correction of the fuzzy identification network unit, the preprocessing of the second spatial scene corrected SAR map also includes size transformation. This is because the SAR map size of the POI area is not fixed, and the import required by the precise identification network unit is a fixed size, so the second spatial scene corrected SAR map (local matrix) needs to be resized to be unified to the same size.

[0119] Preferably, in one preferred embodiment, the pre-built accurate identification network unit is corrected according to the second correction object and the initial value of the network unit indicator of the accurate identification network unit until the second correction end threshold is met, including:

[0120] Step B41: importing the second correction object into the precise identification network unit, and obtaining the drone inference result of the second airspace scene corrected SAR image according to the initial value of the network unit indicator of the precise identification network unit;

[0121] Step B42: Calculate the value of the second cost function according to the drone inference result of the second airspace scene corrected SAR image and the third sequence SAR image corresponding to the second airspace scene corrected SAR image, and correct the network unit index of the accurately identified network unit according to the value of the second cost function and the second preliminary corrected deviation value to obtain a preliminary corrected accurately identified network unit;

[0122] Step B43: obtaining a second approval object, wherein the second approval object includes a second airspace scene approval SAR map and a fourth sequence SAR map corresponding to the second airspace scene approval SAR map, wherein the fourth sequence SAR map is a fourth airspace scene SAR map in which the drone is marked; and importing the second approval object into the preliminary corrected precise identification network unit to obtain a drone inference result of the second airspace scene approval SAR map;

[0123] Step B44: Calculate the value of the second cost function based on the UAV inference result of the second airspace scenario approved SAR map and the fourth sequence SAR map corresponding to the second airspace scenario approved SAR map; if the value of the second cost function regresses to the second deviation value or the number of corrections of the accurately identified network unit is greater than or equal to the second number of corrections, the correction ends; otherwise, correct the network unit indicator of the accurately identified network unit, and update the initial value of the network unit indicator of the accurately identified network unit to the corrected network unit indicator, and return to execute step B41.

[0124] Preferably, in one exemplary embodiment, the second cost function is used to measure the size of the similar part area and the size of the boundary area between the UAV inference result of the second airspace scene corrected SAR image and the third sequence SAR image corresponding to the second airspace scene corrected SAR image, and the size of the similar part area and the size of the boundary area between the UAV inference result of the second airspace scene approved SAR image and the fourth sequence SAR image corresponding to the second airspace scene approved SAR image.

[0125] Preferably, specifically, in step B42, the value of the second cost function is calculated according to the UAV inference result of the second airspace scene corrected SAR image and the third sequence SAR image corresponding to the second airspace scene corrected SAR image, that is, the value of the second cost function is used to measure the area size of the similar part between the UAV inference result of the second airspace scene corrected SAR image and the third sequence SAR image corresponding to the second airspace scene corrected SAR image (that is, the area size of the similar part between the inference result and the standard) and the area size of the boundary area (that is, the area size of the similar part between the inference result and the gold standard).

[0126] Preferably, the size of the similar part area between the drone inference result of the second airspace scene corrected SAR image and the third sequence SAR image corresponding to the second airspace scene corrected SAR image is still measured by the first cost function, i.e., the above-mentioned formula (1). For details, please refer to the relevant description above. In order to avoid redundancy, it will not be described in detail here.

[0127] The area of ​​the boundary region between the drone inference result of the second airspace scene corrected SAR image and the third sequence SAR image corresponding to the second airspace scene corrected SAR image can be measured by the following formula (2):

[0128]

[0129] in, It is the binary SAR image where the contour is located. is the level set representation of the gold standard (the third sequence SAR graph), S θ (q) is the probability distribution of the inference result. The cost function of each second airspace scene corrected SAR image (screening task) is given by the following formula:

[0130] L=(1-l dice )+αl boundary (3)

[0131] In formula (3), α is the coefficient of the size of the boundary area, 0<α<1. The second cost function is the weighted average of the two categories of drones and non-drones:

[0132]

[0133] Where C represents the category (UAV, non-UAV).

[0134] It should be noted that, according to the second cost function, which is a weighted average of the two categories of drones and non-drone, it is not difficult for those skilled in the art to understand that formula (2) is used to explain the size of the boundary area between the drone inference result (or non-drone inference result) and the gold standard. Correspondingly, the size of the similar part area between the drone inference result (or non-drone inference result) and the standard can be obtained by taking drone (or non-drone) c using formula (1). That is, the value of the cost function of each second airspace scene corrected SAR map (screening task) calculated by formula (3) measures the drone inference result (or non-drone inference result), rather than the comprehensive result of drone inference.

[0135] Further, as can be understood by those skilled in the art, the purpose of the correction of the identification network unit is to make the drone inference result obtained by the network unit close to the real drone, in other words, the error between the two is reduced to a certain range, therefore, the value of the second cost function is calculated according to the drone inference result of the second airspace scene corrected SAR map and the third sequence SAR map corresponding to the second airspace scene corrected SAR map in step B42. The correction end threshold of the preliminary corrected precise identification network unit can be the deviation value of the inference result of the second airspace scene corrected SAR map in the second correction object and the corresponding third sequence SAR map regressed to the deviation value.

[0136] Furthermore, as one of the preferred implementations, the step B42 corrects the network unit index of the accurately identified network unit according to the value of the second cost function and the second preliminary correction deviation value to obtain the initially corrected accurately identified network unit, including using the stochastic gradient descent method to optimize the value of the second cost function. This is based on the fact that the correction process of the identified network unit is actually the process of minimizing the cost function, and the derivative can quickly and easily achieve this goal, and this method of finding the derivative is the gradient descent method. Therefore, the use of the stochastic gradient descent method to correct the accurately identified network unit can quickly and easily achieve the correction of the accurately identified network unit.

[0137] In addition, as those skilled in the art can understand, the correction process of identifying the network unit is a multi-cycle correction process. Therefore, in step B44, the correction can be ended by setting the number of corrections. The correction end threshold for accurately identifying the network unit can also be when the number of corrections reaches the number of corrections.

[0138] In summary, the drone identification method for airspace monitoring provided by the present invention, when correcting the precise identification network unit, first uses the second correction object to correct the pre-built precise identification network unit to obtain a preliminary corrected precise identification network unit; then uses the second approval object to approve the preliminary corrected precise identification network unit until the approval set loss no longer decreases or reaches the maximum correction round, thereby not only improving the correction speed of the precise identification network unit, but also improving the inference accuracy of the precise identification network unit; further, since the second cost function of the precise identification network unit not only considers the area size of the similar part between the inference result and the gold standard, but also considers the area size of the boundary area between the inference result and the gold standard, therefore, the drone identification method for airspace monitoring provided by the present invention has both the high efficiency of the fuzzy identification network unit inference and the high precision of the precise identification network unit inference, and can achieve a balance between inference accuracy and inference time efficiency when screening drones in panoramic SAR images.

[0139] Another embodiment of the present invention provides a drone identification system for airspace monitoring. Specifically, see Figure 4 , which schematically shows a block diagram of a drone identification system for airspace monitoring provided by an embodiment of the present invention. Figure 4 It can be seen that the drone identification system for airspace monitoring provided in this embodiment includes: a global SAR image screening module 110 , a local SAR image screening module 120 and a global SAR image reconstruction module 130 .

[0140] Specifically, the global SAR image screening module 110 is configured to import the SAR image of the airspace scene to be identified into the fuzzy recognition network unit to obtain the candidate drone identification result. The local SAR image screening module 120 is configured to import the candidate drone identification result and the airspace scene SAR image to be identified into the precise identification network unit layer by layer to obtain multiple intermediate drone identification results. The global SAR image reconstruction module 130 is configured to filter the intermediate drone identification results layer by layer to obtain accurate drone identification results. Among them, the airspace scene SAR image to be identified, the candidate drone identification results and the accurate drone identification results are all global SAR images; the intermediate drone identification result is a local SAR image.

[0141] Since the basic principle of the drone identification system for airspace monitoring provided by the present invention is the same as that of the drone identification method for airspace monitoring provided by the present invention, please refer to the above description of the drone identification method for airspace monitoring for details. Furthermore, since the drone identification system for airspace monitoring and the drone identification method for airspace monitoring belong to the same inventive concept, they have at least the same beneficial effects as those, and will not be described one by one here.

[0142] Those skilled in the art will appreciate that the contents disclosed in this disclosure may be subject to various variations and improvements. For example, the various devices or components described above may be implemented by hardware, or by software, firmware, or a combination of some or all of the three.

[0143] Flowcharts are used in this disclosure to illustrate the steps of the method according to the embodiments of the present disclosure. It should be understood that the preceding or following steps are not necessarily performed precisely in order. On the contrary, various steps may be processed in reverse order or simultaneously. At the same time, other operations may also be added to these processes.

[0144] Those skilled in the art will appreciate that all or part of the steps in the above method can be completed by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk. Optionally, all or part of the steps in the above embodiment can also be implemented using one or more integrated circuits. Accordingly, each module / unit in the above embodiment can be implemented in the form of hardware or in the form of a software functional module. The present disclosure is not limited to any particular form of combination of hardware and software.

[0145] Unless otherwise defined, all terms used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the present disclosure belongs. It should also be understood that terms such as those defined in common dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an idealized or highly formal sense, unless explicitly defined as such herein.

[0146] The above is an explanation of the present disclosure and should not be considered as a limitation thereof. Although several exemplary embodiments of the present disclosure are described, it will be readily understood by those skilled in the art that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present disclosure. Therefore, all such modifications are intended to be included within the scope of the present disclosure as defined in the claims. It should be understood that the above is an explanation of the present disclosure and should not be considered to be limited to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The present disclosure is defined by the claims and their equivalents.

[0147] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does 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.

[0148] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A method for identifying drones for airspace monitoring, characterized in that: include: Import the SAR image of the airspace scene to be identified into the fuzzy recognition network unit to obtain the candidate drone identification result; Importing the candidate drone identification results and the SAR image of the airspace scene to be identified into the precise identification network unit layer by layer to obtain multiple intermediate drone identification results; Filtering the intermediate drone identification results layer by layer to obtain accurate drone identification results; The SAR map of the airspace scene to be identified, the candidate drone identification result and the accurate drone identification result are all global SAR maps; the intermediate drone identification result is a local SAR map; Before the candidate UAV identification results and the to-be-identified airspace scene SAR image are imported layer by layer into the accurate identification network unit to obtain a plurality of intermediate UAV identification results, the method further includes: The first screening process is performed on the candidate identification SAR image of the UAV identified by the fuzzy identification network unit to obtain the candidate UAV identification result; the first screening process includes: For each POI in the candidate drone identification SAR map, obtain the drone environment identification accuracy and non-drone environment identification accuracy of the POI; determine whether the drone environment identification accuracy is greater than or equal to the non-drone environment identification accuracy, and if so, set the POI value of the POI to the drone POI value; if not, set the POI value of the POI to the non-drone POI value to obtain the candidate drone identification result; The candidate identification SAR image of the UAV is a SAR image containing a flying object; Each POI in the candidate identification SAR image of the drone is a flying object in the corresponding SAR image; The filtering of the intermediate drone identification results layer by layer to obtain accurate drone identification results includes: Filtering the intermediate drone identification results layer by layer to obtain a third drone identification result; wherein the third drone identification result is a global SAR map; Reconstructing the third UAV identification result to obtain the accurate UAV identification result with the same scale as the SAR image of the airspace scene to be identified; The fuzzy recognition network unit is a global convolution recognition network unit, and the precise recognition network unit is a local convolution recognition network unit; The fuzzy recognition network unit is modified by the following steps: Step A10: obtaining a first original correction object, wherein the first original correction object includes a first original airspace scene correction SAR image and a first sequence SAR image corresponding to the first original airspace scene correction SAR image, wherein the first sequence SAR image is a first airspace scene SAR image in which the UAV is marked; Step A20: Expanding the first original corrected object to obtain a first corrected object, wherein the first corrected object includes a first spatial scene corrected SAR image and a first sequence SAR image corresponding to the first spatial scene corrected SAR image; Step A30: setting an initial value of a network element indicator of the fuzzy identification network element; and Correcting the pre-built fuzzy identification network unit according to the first correction object and the initial value of the network unit indicator of the fuzzy identification network unit until a first correction end threshold is met; The first original spatial scene corrected SAR map, the first sequence SAR map corresponding to the first original spatial scene corrected SAR map, the first spatial scene corrected SAR map, and the first sequence SAR map corresponding to the first spatial scene corrected SAR map are all global SAR maps; In step A30, the pre-built fuzzy identification network unit is corrected according to the first correction object and the initial value of the network unit indicator of the fuzzy identification network unit until a first correction end threshold is met, including: Step A31: importing the first correction object into the fuzzy identification network unit, and obtaining the drone inference result of the first airspace scene correction SAR image according to the initial value of the network unit indicator of the fuzzy identification network unit; Step A32: Calculate the value of a first cost function according to the drone inference result of the first airspace scene corrected SAR image and the first sequence SAR image corresponding to the first airspace scene corrected SAR image; and correct the network unit index of the fuzzy identification network unit according to the value of the first cost function and the first preliminary corrected deviation value to obtain a preliminary corrected fuzzy identification network unit; Step A33: obtaining a first approval object, the first approval object comprising a first airspace scene approval SAR map and a second sequence SAR map corresponding to the first airspace scene approval SAR map, the second sequence SAR map being a second airspace scene SAR map with the drone marked; and importing the first approval object into the preliminarily corrected fuzzy recognition network unit to obtain a drone inference result of the first airspace scene approval SAR map; Step A34: Calculate the value of the first cost function according to the drone inference result of the first airspace scene approved SAR map and the second sequence SAR map corresponding to the first airspace scene approved SAR map; if the value of the first cost function regresses to the first deviation value or the number of corrections of the fuzzy identification network unit is greater than or equal to the first correction number, the correction ends; otherwise, correct the network unit index of the fuzzy identification network unit, and update the initial value of the network unit index of the fuzzy identification network unit to the corrected network unit index, and return to execute the step A31; The accurate identification network unit is corrected by the following steps: Step B10: obtaining a second original corrected object, wherein the second original corrected object includes a second original airspace scene corrected SAR image and a third sequence SAR image corresponding to the second original airspace scene corrected SAR image; and importing the second original corrected object into a fuzzy recognition network unit to obtain a UAV inferred SAR image corresponding to the second original airspace scene corrected SAR image; the third sequence SAR image is a third airspace scene SAR image in which the UAV has been marked; Step B20: respectively obtaining the second original airspace scene corrected SAR image and an axial sub-image of the UAV inferred SAR image corresponding to the second original airspace scene corrected SAR image in the Z-axis direction, to obtain a second airspace scene corrected SAR image; Step B30: preprocessing the second spatial scene corrected SAR image to obtain a second corrected object, wherein the second corrected object includes the preprocessed second spatial scene corrected SAR image and a third sequence SAR image corresponding to the preprocessed second spatial scene corrected SAR image; Step B40: Setting an initial value of a network element indicator that accurately identifies the network element; and Modifying the pre-built accurate identification network unit according to the second modification object and the initial value of the network unit indicator of the accurate identification network unit until a second modification end threshold is met; Wherein, the third sequence SAR images corresponding to the preprocessed second airspace scene SAR image and the preprocessed second airspace scene corrected SAR image are both local SAR images; The step of correcting the pre-built accurate identification network unit according to the second correction object and the initial value of the network unit indicator of the accurate identification network unit until a second correction end threshold is met includes: Step B41: importing the second correction object into the precise identification network unit, and obtaining the drone inference result of the second airspace scene corrected SAR image according to the initial value of the network unit indicator of the precise identification network unit; Step B42: Calculate the value of the second cost function according to the drone inference result of the second airspace scene corrected SAR image and the third sequence SAR image corresponding to the second airspace scene corrected SAR image, and correct the network unit index of the accurately identified network unit according to the value of the second cost function and the second preliminary corrected deviation value to obtain a preliminary corrected accurately identified network unit; Step B43: obtaining a second approval object, wherein the second approval object includes a second airspace scene approval SAR map and a fourth sequence SAR map corresponding to the second airspace scene approval SAR map, wherein the fourth sequence SAR map is a fourth airspace scene SAR map in which the drone is marked; and importing the second approval object into the preliminary corrected precise identification network unit to obtain a drone inference result of the second airspace scene approval SAR map; Step B44: Calculate the value of the second cost function based on the UAV inference result of the second airspace scenario approved SAR map and the fourth sequence SAR map corresponding to the second airspace scenario approved SAR map; if the value of the second cost function regresses to the second deviation value or the number of corrections of the accurately identified network unit is greater than or equal to the second number of corrections, the correction ends; otherwise, correct the network unit indicator of the accurately identified network unit, and update the initial value of the network unit indicator of the accurately identified network unit to the corrected network unit indicator, and return to execute step B41.

2. A drone identification system for airspace monitoring, characterized in that: include: The global SAR image screening module is configured to import the SAR image of the airspace scene to be identified into the fuzzy recognition network unit to obtain the candidate UAV identification result; A local SAR image screening module is configured to import the candidate drone identification results and the airspace scene SAR image to be identified into the accurate identification network unit layer by layer to obtain multiple intermediate drone identification results; A global SAR image reconstruction module is configured to filter the intermediate drone identification results layer by layer to obtain accurate drone identification results; The SAR map of the airspace scene to be identified, the candidate drone identification result and the accurate drone identification result are all global SAR maps; the intermediate drone identification result is a local SAR map; The drone identification system for airspace monitoring is used to execute the drone identification method for airspace monitoring as claimed in claim 1.