Artificial intelligence-based unmanned aerial vehicle swarm target identification and positioning method

Abstract

The application discloses an unmanned aerial vehicle swarm target recognition positioning method based on artificial intelligence, relates to the technical field of target recognition positioning, and comprises the following steps: evaluating the recognition importance of different objects in a flight area, and obtaining important objects in the flight area; determining target unmanned aerial vehicles of the important objects; capturing the uncertainty in the important object recognition process of the target unmanned aerial vehicles, analyzing the conflict conditions of different modal indexes in the target unmanned aerial vehicles on the important object recognition; obtaining historical target recognition records from a platform, combining modal conflict data, analyzing the object categories of the important objects, and determining the categories of the important objects; and according to the categories of the important objects and target setting data, obtaining targets in the flight area, and positioning the targets by using an unmanned aerial vehicle swarm, which not only reduces the waste of resources in the unmanned aerial vehicle swarm, but also greatly improves the target recognition efficiency and accuracy, so that the targets can be accurately positioned.

Description

Technical Field

[0001] This invention relates to the field of target recognition and positioning technology, specifically to a method for target recognition and positioning of drone swarms based on artificial intelligence. Background Technology

[0002] Drone swarms refer to a large number of drones connected into an intelligent and unified group system through artificial intelligence, autonomous collaboration, and high-speed communication technologies. Compared with traditional single drones or a small number of drones, drone swarms do not have an absolute central controller. Even if several or more drones are shot down or malfunction during flight, the overall drone swarm system can still complete the task through self-adaptation. This decentralized feature gives drone swarms extremely strong robustness and damage resistance. Furthermore, drone swarms can share information among themselves, which makes their target identification and positioning capabilities much stronger than those of traditional single drones or a small number of drones.

[0003] Because drones in a swarm are equipped with multiple sensors that can sense modal data such as photoelectric, infrared, and spectral data, in actual operation, the data collected by the drone swarm's sensors may conflict due to the camouflage of the target object. For example, a camouflaged target can cool its surface to make its temperature the same as the infrared temperature, which will cause conflicts in the multimodal data collected by the drone swarm. When multimodal data conflicts occur, the drone swarm cannot accurately identify and locate the target, thus seriously affecting the mission. Summary of the Invention

[0004] The purpose of this invention is to provide an artificial intelligence-based method for target identification and localization of drone swarms, in order to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for target identification and localization of unmanned aerial vehicle (UAV) swarms based on artificial intelligence, the method comprising:

[0006] Step S1: Obtain target setting data from the platform, assess the importance of different objects in the flight area, and obtain important objects in the flight area;

[0007] Step S2: Obtain information collection data from drones in the drone swarm, extract the effective identification time of drones for important objects from the information collection data, and determine the target drones of important objects;

[0008] Step S3: Capture the uncertainty in the process of the target UAV recognizing important objects, analyze the conflict status of different modal indicators of the target UAV in recognizing important objects, and obtain modal conflict data;

[0009] Step S4: Obtain historical target recognition records from the platform and combine them with modal conflict data to analyze the object categories of important objects and determine the categories of important objects;

[0010] Step S5: Based on the category of important objects and target settings, acquire targets in the flight area and use a swarm of drones to locate the targets.

[0011] Furthermore, step S1 includes:

[0012] Step S11: Dispatch the drone swarm to fly in the flight area and acquire sensor data of the drones in the drone swarm in the flight area;

[0013] The region image of the flight area is obtained from the sensor data of the drone. The region where each object is located in the image is obtained and cropped to obtain the object image.

[0014] Step S12: Obtain target setting data from the platform, and obtain target images of each target set by the platform for the drone swarm from the target setting data;

[0015] Step S13: Identify and assess the importance of different objects in the flight area to obtain important objects in the flight area.

[0016] Furthermore, in step S13, the importance of different objects in the flight area is assessed, and the specific process includes:

[0017] The object image is converted to a preset color space, and the histograms of each channel in the object image are obtained and normalized. The average chi-square distance A between the target image and the object image in each channel histogram is calculated, and the similarity between the target image and the object image is calculated as B=1 / (1+A).

[0018] If the maximum similarity between an object image and each target is greater than the preset image similarity threshold B', then the object represented by the object image is determined to be of recognition importance in the drone swarm and is recorded as an important object in the flight area.

[0019] Furthermore, step S2 includes:

[0020] Step S21: Obtain information collection data of important objects by drones in the drone swarm, and obtain the distance set D from the information collection data;

[0021] Step S22: Calculate the average straight-line distance between the drone and the important object in the distance set D over n units of time period T, and use it as the average distance between the drone and the important object;

[0022] Find the minimum average distance d between each drone in the drone swarm and an important object. min Obtain the total number of drones m, set the proportion value M´, and calculate the characteristic distance order between the drone swarm and important objects C=m / M´, where 0 <M´<1;

[0023] Sort the average distances between each drone and the important object, and obtain the minimum value d. min The mean of the distances between the drones and important objects, with the order being C, is denoted as the distance feature value between the drone swarm and the important objects.

[0024] Step S23: Obtain the total duration for which the straight-line distance between each drone in the drone swarm and the important object is less than the distance feature value, and record it as the effective identification time length. Set the target number threshold Q, sort each drone in the drone swarm according to the effective identification time length, and select the first Q drones as the target drones of the important object in order.

[0025] Furthermore, the process of obtaining the distance set D in step S21 includes: obtaining the time period T during which the drone collects important objects from the information collection data, setting a unit duration, extracting the straight-line distance between the drone and the important objects every unit duration within the time period T during which the drone collects important objects, and collecting the data to obtain the distance set D.

[0026] Furthermore, step S3 includes:

[0027] Step S31: Acquire sensor data from the drone swarm, and obtain multimodal data of the target drone when identifying important objects from the sensor data. The multimodal data includes data of various modal indicators.

[0028] Step S32: Based on the multimodal data of important objects, use the pre-deployed intelligent classifier in the target UAV to obtain the object recognition confidence set P of the modal indicators of the target UAV for important objects, and calculate the classification entropy H(p) of the modal indicators of the target UAV for important objects.

[0029] Set a classification entropy threshold H'. If H(p) > H', it is determined that the modal indicators of the target UAV have uncertainty in the identification of important objects; otherwise, it is determined that the modal indicators of the target UAV have deterministic identification of important objects.

[0030] Step S33: Analyze the conflict status of different modal indicators in the target UAV for the identification of important objects, and obtain modal conflict data.

[0031] Furthermore, step S33, which analyzes the conflict between different modal indicators of the target UAV for the identification of important objects, specifically includes the following process:

[0032] Modal indicators that show uncertainty in the identification of important objects in the target UAV are filtered out from all modal indicators. The object identification confidence set F of the modal indicators of the target UAV for important objects is obtained. The object category to which the maximum confidence value in the object identification confidence set F belongs is obtained and used as the feature object category of the modal indicators of the target UAV for the identification of important objects.

[0033] The drone identification set G is obtained by acquiring several target drones from the drone swarm that retain modal indicators during the process of identifying important objects, acquiring the feature object categories of important object identification from the modal indicators of several target drones, and aggregating them.

[0034] When several target drones in the drone identification set G identify important objects in different categories, the mean confidence score of the target drones in the drone identification set G when they identify important objects as any one of the feature object categories is obtained and used as the confidence feature value of any one feature object category.

[0035] Extract the feature object categories whose confidence feature values ​​are greater than the preset confidence feature threshold from the UAV identification set G, and record them as the target object categories identified by the modal index of important objects in the UAV swarm.

[0036] When the confidence feature values ​​of each feature object category extracted from the UAV identification set G are all less than the confidence feature threshold, the feature object categories in the UAV identification set G are sorted according to the values ​​of the confidence feature values. The difference between the confidence feature value of the highest-ranking feature object category and the feature object categories of each rank is calculated. The feature object category with a confidence feature difference greater than the preset feature threshold is obtained and recorded as the target object category.

[0037] The system obtains the target object categories identified by various modal indicators of important objects in the drone swarm. When the target object categories identified by various modal indicators of important objects in the drone swarm are the same, it is determined that there is a conflict in the identification of important objects by various modal indicators of different target drones in the drone swarm.

[0038] The target object categories and confidence feature values ​​identified by various modal indicators of the drone swarm for important objects are obtained and aggregated to obtain modal conflict data of the drone swarm for important objects.

[0039] Furthermore, step S4 includes:

[0040] Step S41: Obtain modal conflict data of important objects by drone swarm, extract the target object categories corresponding to modal indicators from the modal conflict data and aggregate them to obtain the target object category set U of important objects in the drone swarm based on modal indicators.

[0041] The historical target category set ζ is obtained by retrieving the categories of historical targets identified in each historical target recognition record from the platform and aggregating them.

[0042] Step S42: When the intersection between the historical target category set ζ and the target object category set U is empty, obtain the target object category corresponding to the maximum confidence feature value in the target object category set U and mark it, obtain the marked object category of important objects in the modal index of the UAV swarm, and determine the marked object category as the category identified by the modal index of important objects in the UAV swarm.

[0043] If the intersection between the historical target category set ζ and the target object category set U is not empty, then obtain the target object category in the intersection of the historical target category set ζ and the target object category set U, and record it as the category to be analyzed in the modal index of the UAV swarm. Calculate the determination value L for identifying important objects as the category to be analyzed in the modal index of the UAV swarm.

[0044] Step S43: Identify important objects as categories to be analyzed based on various modal indicators in the drone swarm, analyze the object categories of important objects, and determine the categories of important objects;

[0045] The above steps analyze the object category of important objects, thereby solving the problem of conflicting modal indicators when drone swarms identify important objects. This enables drone swarms to identify target camouflage, improves the efficiency and accuracy of target identification, and also facilitates task coordination and planning of drone swarms.

[0046] Furthermore, the specific process of analyzing the object category of important objects in step S43 includes:

[0047] Obtain the risk category corresponding to the maximum value of the modal index in each category to be analyzed in the drone swarm, obtain the marked object category of important objects by the modal index in the drone swarm, and determine the marked object category as the category identified by the modal index of important objects in the drone swarm.

[0048] Obtain the object categories identified by various modal indicators of important objects in the drone swarm;

[0049] When all modal indicators in the drone swarm identify the same object category for an important object, the same object category is taken as the category of the important object. Otherwise, a definite value is obtained for each modal indicator in the drone swarm to identify the important object as the object category. The definite value is specifically between 0 and 1.

[0050] Obtain environmental data of the flight area within the current period, obtain environmental modal weight comparison data from the platform, and, based on the environmental data, determine the weights of various modal indicators in the drone swarm within the current period from the environmental modal weight comparison data.

[0051] Obtain the average value L´ of the determined values ​​of various modal indicators in the drone swarm that identify important objects as the marked object category, and calculate the determined value ΔL=1-L´ of a certain modal indicator in the drone swarm that identifies important objects as not being the marked object category.

[0052] The modal determination value γ is obtained by multiplying the weights of various modal indicators that identify important objects as object categories in the drone swarm and summing the determined values.

[0053] The modal uncertainty value w is obtained by multiplying the weights of several modal indicators in the drone swarm that identify important objects that are not classified as labeled objects by the sum of their determined values. The fusion determination value V = γ - w is then calculated to determine whether the important objects are classified as labeled objects.

[0054] Obtain the category corresponding to the maximum value of the fusion determination value of each labeled object category for important objects, and record it as the category of important objects.

[0055] Furthermore, step S5 includes:

[0056] Step S51: Obtain the categories of important objects in the flight area, and obtain the categories of each target set for the current period of the UAV swarm from the target setting data;

[0057] Step S52: When the category of an important object in the flight area is the same as the category of a target set by the drone swarm in the current cycle, the important object is identified as the target of the drone swarm in the flight area. The drone swarm is then used to locate the target in the flight area, obtain the target's coordinates, and control the drone swarm to upload the target's category and coordinates in the flight area to the platform.

[0058] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention realizes intelligent identification and positioning of targets by drone swarms. First, it assesses the importance of each object within the flight area of ​​the drone swarm, thereby saving resources of the drone swarm. Furthermore, considering that the effective data collection time for different objects varies among the numerous drones in the swarm, it identifies key target drones. In addition, it analyzes the conflicts in multimodal data collected by different drones in the swarm, which not only reduces the waste of resources in the drone swarm but also greatly improves the efficiency and accuracy of target identification, enabling the target to be accurately located. Attached Figure Description

[0059] Figure 1 This is a flowchart of the method for target identification and localization of drone swarms based on artificial intelligence according to the present invention;

[0060] Figure 2 This is a flowchart illustrating the target category determination process of the artificial intelligence-based drone swarm target identification and positioning method of the present invention. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] Example: Figures 1-2 As shown, this invention provides a technical solution: a method for target identification and localization of drone swarms based on artificial intelligence. The method includes:

[0063] Step S1: Obtain target setting data from the platform, assess the importance of different objects in the flight area, and obtain important objects in the flight area;

[0064] Step S2: Obtain information collection data from drones in the drone swarm, extract the effective identification time of drones for important objects from the information collection data, and determine the target drones of important objects;

[0065] Step S3: Capture the uncertainty in the process of the target UAV recognizing important objects, analyze the conflict status of different modal indicators of the target UAV in recognizing important objects, and obtain modal conflict data;

[0066] Step S4: Obtain historical target recognition records from the platform and combine them with modal conflict data to analyze the object categories of important objects and determine the categories of important objects;

[0067] Step S5: Based on the category of important objects and target settings, acquire targets in the flight area and use a swarm of drones to locate the targets.

[0068] Furthermore, step S1 includes:

[0069] Step S11: Dispatch the drone swarm to fly in the flight area and acquire sensor data of the drones in the drone swarm in the flight area;

[0070] The region image of the flight area is obtained from the sensor data of the drone. The region where each object is located in the image is obtained and cropped to obtain the object image.

[0071] Step S12: Obtain target setting data from the platform, and obtain target images of each target set by the platform for the drone swarm from the target setting data;

[0072] For example, YOLO's detection tools can be used to quickly crop the area containing an object from an image;

[0073] Step S13: Identify and assess the importance of different objects in the flight area to obtain the important objects in the flight area;

[0074] In step S13, the importance of different objects in the flight area is assessed, and the specific process includes:

[0075] The object image is converted to a preset color space, and the histograms of each channel in the object image are obtained and normalized. The average chi-square distance A between the target image and the object image in each channel histogram is calculated, and the similarity between the target image and the object image is calculated as B=1 / (1+A).

[0076] If the maximum similarity between an object image and each target is greater than the preset image similarity threshold B', then the object represented by the object image is determined to be of recognition importance in the drone swarm and is recorded as an important object in the flight area.

[0077] Step S2 includes:

[0078] Step S21: Obtain information collection data of important objects by drones in the drone swarm, and obtain the distance set D from the information collection data;

[0079] The process of obtaining the distance set D in step S21 includes: obtaining the time period T for the drone to collect important objects from the information collection data, setting the unit time period, extracting the straight-line distance between the drone and the important objects every unit time period within the time period T for collecting important objects, and collecting them to obtain the distance set D.

[0080] Step S22: Calculate the average straight-line distance between the drone and the important object in the distance set D over n units of time period T, and use it as the average distance between the drone and the important object;

[0081] Find the minimum average distance d between each drone in the drone swarm and an important object. min Obtain the total number of drones m, set the proportion value M´, and calculate the characteristic distance order between the drone swarm and important objects C=m / M´, where 0 <M´<1;

[0082] Sort the average distances between each drone and the important object, and obtain the minimum value d. min The mean of the distances between the drones and important objects, with the order being C, is denoted as the distance feature value between the drone swarm and the important objects.

[0083] Step S23: Obtain the total duration for which the straight-line distance between each drone in the drone swarm and the important object is less than the distance feature value, and record it as the effective identification time length. Set the target number threshold Q, sort each drone in the drone swarm according to the effective identification time length, and select the first Q drones as the target drones of the important object in order.

[0084] Step S3 includes:

[0085] Step S31: Acquire sensor data from the drone swarm, and obtain multimodal data of the target drone when identifying important objects from the sensor data. The multimodal data includes data of various modal indicators.

[0086] For example, various modal indicators include photoelectric, infrared, and spectral parameters;

[0087] Step S32: Based on the multimodal data of important objects, use the pre-deployed intelligent classifier in the target UAV to obtain the object recognition confidence set P of the modal indicators of the target UAV for important objects, and calculate the classification entropy H(p) of the modal indicators of the target UAV for important objects.

[0088] For example, the specific formula for calculating the classification entropy H(p) is:

[0089] ,

[0090] Where j is the total number of elements in the object recognition confidence set P; p i The intelligent classifier outputs the confidence score of the i-th object category based on the modal index data; p j The confidence level for identifying the j-th object category in the confidence set P;

[0091] For example, it is necessary to extract relevant features for each modality and use a pre-trained intelligent separator (such as SVM or neural network) to perform object recognition.

[0092] For example, for each modality, the relevant feature extraction is as follows:

[0093] Photoelectric images: shape, texture, and color features;

[0094] Infrared images: thermal features, temperature distribution;

[0095] Spectral data: spectral signature, reflectance characteristics;

[0096] Set a classification entropy threshold H'. If H(p) > H', it is determined that the modal indicators of the target UAV have uncertainty in the identification of important objects; otherwise, it is determined that the modal indicators of the target UAV have deterministic identification of important objects.

[0097] Step S33: Analyze the conflict status of different modal indicators in the target UAV for the identification of important objects, and obtain modal conflict data;

[0098] The specific process of analyzing the conflict between different modal indicators of the target UAV for the identification of important objects in step S33 includes:

[0099] Modal indicators that show uncertainty in the identification of important objects in the target UAV are filtered out from all modal indicators. The object identification confidence set F of the modal indicators of the target UAV for important objects is obtained. The object category to which the maximum confidence value in the object identification confidence set F belongs is obtained and used as the feature object category of the modal indicators of the target UAV for the identification of important objects.

[0100] The drone identification set G is obtained by acquiring several target drones from the drone swarm that retain modal indicators during the process of identifying important objects, acquiring the feature object categories of important object identification from the modal indicators of several target drones, and aggregating them.

[0101] When several target drones in the drone identification set G identify important objects in different categories, the mean confidence score of the target drones in the drone identification set G when they identify important objects as any one of the feature object categories is obtained and used as the confidence feature value of any one feature object category.

[0102] Extract the feature object categories whose confidence feature values ​​are greater than the preset confidence feature threshold from the UAV identification set G, and record them as the target object categories identified by the modal index of important objects in the UAV swarm.

[0103] When the confidence feature values ​​of each feature object category extracted from the UAV identification set G are all less than the confidence feature threshold, the feature object categories in the UAV identification set G are sorted according to the values ​​of the confidence feature values. The difference between the confidence feature value of the highest-ranking feature object category and the feature object categories of each rank is calculated. The feature object category with a confidence feature difference greater than the preset feature threshold is obtained and recorded as the target object category.

[0104] The system obtains the target object categories identified by various modal indicators of important objects in the drone swarm. When the target object categories identified by various modal indicators of important objects in the drone swarm are the same, it is determined that there is a conflict in the identification of important objects by various modal indicators of different target drones in the drone swarm.

[0105] The target object categories and confidence feature values ​​identified by various modal indicators of the drone swarm for important objects are obtained and aggregated to obtain modal conflict data of the drone swarm for important objects.

[0106] Step S4 includes:

[0107] Step S41: Obtain modal conflict data of important objects by drone swarm, extract the target object categories corresponding to modal indicators from the modal conflict data and aggregate them to obtain the target object category set U of important objects in the drone swarm based on modal indicators.

[0108] The historical target category set ζ is obtained by retrieving the categories of historical targets identified in each historical target recognition record from the platform and aggregating them.

[0109] Step S42: When the intersection between the historical target category set ζ and the target object category set U is empty, obtain the target object category corresponding to the maximum confidence feature value in the target object category set U and mark it, obtain the marked object category of important objects in the modal index of the UAV swarm, and determine the marked object category as the category identified by the modal index of important objects in the UAV swarm.

[0110] If the intersection between the historical target category set ζ and the target object category set U is not empty, then obtain the target object category in the intersection of the historical target category set ζ and the target object category set U, and record it as the category to be analyzed in the modal index of the UAV swarm. Calculate the determination value L for identifying important objects as the category to be analyzed in the modal index of the UAV swarm.

[0111] For example, the specific calculation process for determining the value L is as follows:

[0112] Obtain several target identification records with historical targets as the category to be analyzed, and obtain the confidence level of the category to be analyzed in the modal index from several historical target identification records;

[0113] Calculate the confidence similarity value Y=1-[|pp´| / max(p,p´)] of the category to be analyzed in the historical target recognition record, where p is the confidence feature value of the category to be analyzed in the target object category set U, and p´ is the confidence level of the category to be analyzed in the historical target recognition record;

[0114] Calculate the mean of the confidence values ​​of the categories to be analyzed in several historical target identification records, and record it as the historical confidence value of the modal index of the drone swarm on the category to be analyzed;

[0115] The confidence feature values ​​of the modal indicators in the drone swarm for each category to be analyzed are obtained and normalized. The determination value L = Y × η for identifying important objects as the category to be analyzed in the modal indicators of the drone swarm is then calculated.Y +p Δ ×η p , where η Y and η p These represent the historical confidence values ​​and confidence feature values ​​of modal indicators in the drone swarm for the category to be analyzed;

[0116] Step S43: Identify important objects as categories to be analyzed based on various modal indicators in the drone swarm, analyze the object categories of the important objects, and determine the categories of the important objects.

[0117] The specific process of analyzing the object category of important objects in step S43 includes:

[0118] Obtain the risk category corresponding to the maximum value of the modal index in each category to be analyzed in the drone swarm, obtain the marked object category of important objects by the modal index in the drone swarm, and determine the marked object category as the category identified by the modal index of important objects in the drone swarm.

[0119] Obtain the marked object categories identified by various modal indicators of important objects in the drone swarm;

[0120] When all modal indicators in the drone swarm identify the same object category for an important object, the same object category is taken as the category of the important object. Otherwise, a definite value is obtained for each modal indicator in the drone swarm to identify the important object as the object category. The definite value is specifically between 0 and 1.

[0121] The specific weights corresponding to each modal index are obtained from the environmental modal weight comparison data as follows:

[0122] The environmental data of the flight area in the current period includes the values ​​of various environmental indicators. The environmental modal weight comparison data includes various modal weight reassemblies. Each modal weight reassembly has a corresponding range in each environmental indicator. When the environmental data of the flight area in the current period includes the values ​​of various environmental indicators in a certain modal weight reassembly, then the weight of each modal indicator in that modal weight reassembly is the weight of each modal indicator in the drone swarm in the current period.

[0123] Obtain environmental data of the flight area within the current period, obtain environmental modal weight comparison data from the platform, and, based on the environmental data, determine the weights of various modal indicators in the drone swarm within the current period from the environmental modal weight comparison data.

[0124] For example, the specific weights corresponding to each modal index obtained from the environmental modal weight comparison data are as follows:

[0125] The environmental data of the flight area in the current period includes the values ​​of various environmental indicators. The environmental modal weight comparison data includes various modal weight reassemblies. Each modal weight reassembly has a corresponding range in each environmental indicator. When the environmental data of the flight area in the current period includes the values ​​of various environmental indicators in a certain modal weight reassembly, then the weight of each modal indicator in that modal weight reassembly is the weight of each modal indicator in the drone swarm in the current period.

[0126] For example, various environmental indicators include temperature, light intensity, etc.

[0127] Obtain the average value L´ of the determined values ​​of various modal indicators in the drone swarm that identify important objects as the marked object category, and calculate the determined value ΔL=1-L´ of a certain modal indicator in the drone swarm that identifies important objects as not being the marked object category.

[0128] The modal determination value γ is obtained by multiplying the weights of various modal indicators that identify important objects as object categories in the drone swarm and summing the determined values.

[0129] The modal uncertainty value w is obtained by multiplying the weights of several modal indicators in the drone swarm that identify important objects that are not classified as labeled objects by the sum of their determined values. The fusion determination value V = γ - w is then calculated to determine whether the important objects are classified as labeled objects.

[0130] Obtain the category corresponding to the maximum value of the fusion determination value of each labeled object category for important objects, and record it as the category of important objects.

[0131] Step S5 includes:

[0132] Step S51: Obtain the categories of important objects in the flight area, and obtain the categories of each target set for the current period of the UAV swarm from the target setting data;

[0133] Step S52: When the category of an important object in the flight area is the same as the category of a target set by the drone swarm in the current cycle, the important object is identified as the target of the drone swarm in the flight area. The drone swarm is then used to locate the target in the flight area, obtain the target's coordinates, and control the drone swarm to upload the target's category and coordinates in the flight area to the platform.

[0134] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for target identification and localization of unmanned aerial vehicle (UAV) swarms based on artificial intelligence, characterized in that, The method includes: Step S1: Obtain target setting data from the platform, assess the importance of different objects in the flight area, and obtain important objects in the flight area; Step S2: Obtain information collection data from drones in the drone swarm, extract the effective identification time of drones for important objects from the information collection data, and determine the target drones of important objects; Step S3: Capture the uncertainty in the process of the target UAV recognizing important objects, analyze the conflict status of different modal indicators of the target UAV in recognizing important objects, and obtain modal conflict data; Step S4: Obtain historical target recognition records from the platform and combine them with modal conflict data to analyze the object categories of important objects and determine the categories of important objects; Step S5: Based on the category and target settings of important objects, acquire targets in the flight area and use a drone swarm to locate the targets; Step S2 includes: Step S21: Obtain information collection data of important objects by drones in the drone swarm, and obtain the distance set D from the information collection data; Step S22: Calculate the average straight-line distance between the drone and the important object in the distance set D over n units of time period T, and use it as the average distance between the drone and the important object; Find the minimum average distance d between each drone in the drone swarm and an important object. min Obtain the total number of drones m, set the proportion value M´, and calculate the characteristic distance order between the drone swarm and important objects C=m / M´, where 0 <M´<1; Sort the average distances between each drone and the important object, and obtain the minimum value d. min The mean of the distances between the drones and important objects, with the order being C, is denoted as the distance feature value between the drone swarm and the important objects. Step S23: Obtain the total duration for which the straight-line distance between each drone in the drone swarm and the important object is less than the distance feature value, and record it as the effective identification time length. Set the target number threshold Q, sort each drone in the drone swarm according to the effective identification time length, and select the first Q drones as the target drones of the important object in order.

2. The method for target identification and localization of unmanned aerial vehicle (UAV) swarms based on artificial intelligence according to claim 1, characterized in that, Step S1 includes: Step S11: Dispatch the drone swarm to fly in the flight area and acquire sensor data of the drones in the drone swarm in the flight area; The region image of the flight area is obtained from the sensor data of the drone. The region where each object is located in the image is obtained and cropped to obtain the object image. Step S12: Obtain target setting data from the platform, and obtain target images of each target set by the platform for the drone swarm from the target setting data; Step S13: Identify and assess the importance of different objects in the flight area to obtain important objects in the flight area.

3. The method for target identification and localization of unmanned aerial vehicle (UAV) swarms based on artificial intelligence according to claim 2, characterized in that, In step S13, the importance of different objects in the flight area is assessed, and the specific process includes: The object image is converted to a preset color space, and the histograms of each channel in the object image are obtained and normalized. The average chi-square distance A between the target image and the object image in each channel histogram is calculated, and the similarity between the target image and the object image is calculated as B=1 / (1+A). If the maximum similarity between the object image and each of the targets is greater than the preset image similarity threshold B', then the object represented by the object image is determined to be of recognition importance in the drone swarm and is recorded as an important object in the flight area.

4. The artificial intelligence-based drone swarm target identification and localization method according to claim 3, characterized in that, The process of obtaining the distance set D in step S21 includes: obtaining the time period T for the drone to collect important objects from the information collection data, setting a unit time period, extracting the straight-line distance between the drone and the important objects every unit time period within the time period T for collecting important objects, and collecting them to obtain the distance set D.

5. The artificial intelligence-based drone swarm target identification and localization method according to claim 4, characterized in that, Step S3 includes: Step S31: Acquire sensor data from the drone swarm, and obtain multimodal data of the target drone when identifying important objects from the sensor data. The multimodal data includes data of various modal indicators. Step S32: Based on the multimodal data of important objects, use the pre-deployed intelligent classifier in the target UAV to obtain the object recognition confidence set P of the modal indicators of the target UAV for important objects, and calculate the classification entropy H(p) of the modal indicators of the target UAV for important objects. Set a classification entropy threshold H'. If H(p) > H', it is determined that the modal indicators of the target UAV have uncertainty in the identification of important objects; otherwise, it is determined that the modal indicators of the target UAV have deterministic identification of important objects. Step S33: Analyze the conflict status of different modal indicators in the target UAV for the identification of important objects, and obtain modal conflict data.

6. The artificial intelligence-based drone swarm target identification and localization method according to claim 5, characterized in that, The specific process of analyzing the conflict situation of different modal indicators in the target UAV for the identification of important objects in step S33 includes: Modal indicators that show uncertainty in the identification of important objects in the target UAV are filtered out from all modal indicators. The object identification confidence set F of the modal indicators of the target UAV for important objects is obtained. The object category to which the maximum confidence value in the object identification confidence set F belongs is obtained and used as the feature object category of the modal indicators of the target UAV for the identification of important objects. The drone identification set G is obtained by acquiring several target drones from the drone swarm that retain modal indicators during the process of identifying important objects, acquiring the feature object categories of important object identification from the modal indicators of several target drones, and aggregating them. When several target drones in the drone identification set G identify important objects in different categories, the mean confidence score of the target drones in the drone identification set G when they identify important objects as any one of the feature object categories is obtained and used as the confidence feature value of any one feature object category. Extract the feature object categories whose confidence feature values ​​are greater than the preset confidence feature threshold from the UAV identification set G, and record them as the target object categories identified by the modal index of important objects in the UAV swarm. When the confidence feature values ​​of each feature object category extracted from the UAV identification set G are all less than the confidence feature threshold, the feature object categories in the UAV identification set G are sorted according to the values ​​of the confidence feature values. The difference between the confidence feature value of the highest-ranking feature object category and the feature object categories of each rank is calculated. The feature object category with a confidence feature difference greater than the preset feature threshold is obtained and recorded as the target object category. The system obtains the target object categories identified by various modal indicators of important objects in the drone swarm. When the target object categories identified by various modal indicators of important objects in the drone swarm are the same, it is determined that there is a conflict in the identification of important objects by various modal indicators of different target drones in the drone swarm. The target object categories and confidence feature values ​​identified by various modal indicators of the drone swarm for important objects are obtained and aggregated to obtain modal conflict data of the drone swarm for important objects.

7. The method for target identification and localization of unmanned aerial vehicle swarms based on artificial intelligence according to claim 5, characterized in that, Step S4 includes: Step S41: Obtain modal conflict data of important objects by drone swarm, extract the target object categories corresponding to modal indicators from the modal conflict data and aggregate them to obtain the target object category set U of important objects in the drone swarm based on modal indicators. The historical target category set ζ is obtained by retrieving the categories of historical targets identified in each historical target recognition record from the platform and aggregating them. Step S42: When the intersection between the historical target category set ζ and the target object category set U is empty, obtain the target object category corresponding to the maximum confidence feature value in the target object category set U and mark it, obtain the marked object category of important objects in the modal index of the UAV swarm, and determine the marked object category as the category identified by the modal index of important objects in the UAV swarm. If the intersection between the historical target category set ζ and the target object category set U is not empty, then obtain the target object category in the intersection of the historical target category set ζ and the target object category set U, and record it as the category to be analyzed in the modal index of the UAV swarm. Calculate the determination value L for identifying important objects as the category to be analyzed in the modal index of the UAV swarm. Step S43: Identify important objects as categories to be analyzed based on various modal indicators in the drone swarm, analyze the object categories of the important objects, and determine the categories of the important objects.

8. The artificial intelligence-based drone swarm target identification and localization method according to claim 7, characterized in that, The specific process of analyzing the object category of important objects in step S43 includes: Obtain the risk category corresponding to the maximum value of the modal index in each category to be analyzed in the drone swarm, obtain the marked object category of important objects by the modal index in the drone swarm, and determine the marked object category as the category identified by the modal index of important objects in the drone swarm. Obtain the object categories identified by various modal indicators of important objects in the drone swarm; When all modal indicators in the drone swarm identify the same object category for an important object, the same object category is taken as the category of the important object. Otherwise, a definite value is obtained for each modal indicator in the drone swarm to identify the important object as the object category. The definite value is specifically between 0 and 1. Obtain environmental data of the flight area within the current period, obtain environmental modal weight comparison data from the platform, and, based on the environmental data, determine the weights of various modal indicators in the drone swarm within the current period from the environmental modal weight comparison data. Obtain the average value L´ of the determined values ​​of various modal indicators in the drone swarm that identify important objects as the marked object category, and calculate the determined value ΔL=1-L´ of a certain modal indicator in the drone swarm that identifies important objects as not being the marked object category. The modal determination value γ is obtained by multiplying the weights of various modal indicators that identify important objects as object categories in the drone swarm and summing the determined values. The modal uncertainty value w is obtained by multiplying the weights of several modal indicators in the drone swarm that identify important objects that are not classified as labeled objects by the sum of their determined values. The fusion determination value V = γ - w is then calculated to determine whether the important objects are classified as labeled objects. Obtain the category corresponding to the maximum value of the fusion determination of each labeled object category for important objects, and denote it as the category of important objects.

9. The method for target identification and localization of unmanned aerial vehicle swarms based on artificial intelligence according to claim 7, characterized in that, Step S5 includes: Step S51: Obtain the categories of important objects in the flight area, and obtain the categories of each target set for the current period of the UAV swarm from the target setting data; Step S52: When the category of an important object in the flight area is the same as the category of a target set by the drone swarm in the current cycle, the important object is identified as the target of the drone swarm in the flight area. The drone swarm is then used to locate the target in the flight area, obtain the target's coordinates, and control the drone swarm to upload the target's category and coordinates in the flight area to the platform.

Images