Unmanned aerial vehicle abnormal intelligent inspection, identification and analysis system and method

Abstract

This invention provides a method and system for intelligent inspection, identification, and analysis of anomalies by unmanned aerial vehicles (UAVs). The system includes a discoverer that searches for vessels and determines whether there are any anomalies based on an anomaly identification strategy. If an anomaly is found, the vessel is identified as a target vessel and tracked. During the tracking process, the discoverer acquires the target vessel's location and the status of surrounding predators, selects a corresponding predator, and adjusts the distance between the discoverer and the target vessel and predators using a tracking strategy. The discoverer also adjusts its flight path and transmits the target vessel's characteristics, including the vessel's location, anchor point, and anomaly information, to the selected predator. After acquiring the target characteristics, the predator flies towards the target vessel's location. Once the target vessel enters its acquisition range, the predator locks onto it based on its anchor point and a feature identification strategy. The predator then monitors the target vessel to determine if any anomalies exist.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicles (UAVs), and in particular to an intelligent inspection, identification and analysis system and method for UAV anomalies. Background Technology

[0002] Drones can be used to obtain important ground information, such as images, including still pictures and videos. During aerial ground observation, moving targets on the water surface (such as ships and vessels) are of significant value and are key reconnaissance targets. Drones can be used to detect abnormal discharges and illegal stops by ships, often requiring continuous monitoring during flight. Because moving targets are also in motion, they often become difficult to continuously capture as their movement exceeds the drone's field of view.

[0003] The search, tracking, and localization of moving targets is a hot research topic in the field of machine vision and is widely used in practice. The key challenge lies in ensuring the speed and accuracy of large-scale search and localization. Existing technical solutions mainly fall into two categories: one utilizes monocular or binocular vision for target identification, tracking, and localization. While this approach can accurately identify and localize moving targets, it cannot simultaneously guarantee large-scale search and precise localization of multiple targets; the other involves installing a sufficient number of cameras within the search area to ensure that the entire area is searchable. Although this method addresses the aforementioned problems to some extent, it increases costs and complicates subsequent image processing. In summary, current engineering technologies struggle to simultaneously satisfy the requirements for large-scale search and precise localization of multiple moving targets. For example, application number CN201510684345.8 discloses a device and method for searching and localizing multiple moving targets based on bird visual characteristics. This device, from a biomimetic perspective, utilizes the visual characteristics of animals in predator-prey relationships and incorporates the biconcave structure of birds for biomimetic design. The device consists of a horizontally mounted circular base, four identical 2-DOF monocular mechanisms evenly distributed along the circle, a fixed camera mounted in the center of the base, and corresponding computer and circuit components. This device combines the advantages of a wide field of vision for predators and the well-developed binocular vision of predators, enabling rapid searching and precise positioning of multiple moving targets across a panoramic range. It is suitable for application and promotion in the field of visual surveillance in engineering practice. However, due to the vastness of the water surface, it is impossible to track moving targets on the water surface. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the purpose of this invention is to provide an intelligent inspection, identification and analysis system and method for unmanned aerial vehicles (UAVs) to overcome the above-mentioned defects in the existing technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] Methods for intelligent inspection, identification, and analysis of drone anomalies, including

[0007] In the target search step, the discoverer determines whether there is any abnormality in the search vessel based on the search vessel and the anomaly identification strategy. If there is an anomaly, the vessel is marked as the target vessel and tracked.

[0008] The predator docking process involves the discoverer tracking the target vessel, acquiring the target vessel's location and the status of surrounding predators, selecting the corresponding predator, and adjusting the distance between the discoverer and the target vessel and predators through the tracking strategy. The discoverer also adjusts its flight path and transmits the target vessel's target characteristics to the selected predator. The target characteristics include the vessel's location, anchoring point, and any abnormal situations.

[0009] The target vessel identification process involves the predator acquiring target characteristics and then flying towards the target vessel's location. Once the target vessel enters the predator's collection range, it locks onto the target vessel using a feature recognition strategy based on the anchor points on the target vessel.

[0010] The evidence collection process involves the predator monitoring the target vessel to determine if any anomalies exist. If anomalies are found, evidence frames are acquired. If no anomalies are found, keyframes with target features are acquired by the discoverer, along with detailed images of the target vessel collected by the predator. These are cross-referenced to form a complete chain of evidence.

[0011] Preferably, the tracking strategy includes acquiring the target vessel's speed, trajectory, and environmental visibility; obtaining the docking weight and positioning weight based on the positional relationship between the target vessel, the predator, and the discoverer, and calculating the tracking value according to a data index table; and presetting a tracking threshold. When the tracking value is less than the tracking threshold, an adjustment strategy is generated to adjust the docking weight and positioning weight so that the tracking value is greater than the tracking threshold.

[0012] Preferably, the tracking strategy further includes an interface step, the interface step including...

[0013] The data acquisition sub-step is used to dynamically adjust the communication frequency and transmission frequency based on the distance between the discoverer and the predator, environmental conditions, and their flight speeds, in order to obtain the real-time communication quality between the discoverer and the predator.

[0014] The docking distance calculation sub-step is used to obtain the docking weight and real-time communication quality, and to obtain the docking distance through a distance adjustment algorithm. Based on the docking distance, the flight speed and flight path between the discoverer and the predator are dynamically adjusted.

[0015] Preferably, the tracking strategy further includes a positioning step, which is used to obtain a positioning weight, obtain a positioning distance based on the positioning weight, and dynamically adjust the flight speed and flight path of the discoverer based on the positioning distance.

[0016] Preferably, the predator docking strategy includes an anchor point acquisition step. This step involves acquiring several frames of images collected by the discoverer as target images, acquiring image information of the target vessel and other vessels in the target images, comparing the target vessel with the other vessels, acquiring several feature images of the target vessel as anchor points, and acquiring anchor point weights based on the frequency of the anchor points appearing on the other vessels.

[0017] Preferably, the feature recognition strategy includes obtaining anchor points and obtaining images collected by predators as images to be analyzed, calculating the similarity between the ship images and anchor point images in the images to be analyzed, calculating the similarity based on the anchor point weights, prioritizing the similarity based on the similarity, and selecting the ship with the highest similarity as the target ship.

[0018] Preferably, the predator docking step also includes a relay matching sub-step. When the predator does not enter the predation range during the current discoverer's tracking process, and the target vessel is about to leave the current discoverer's monitoring range, the current discoverer searches for other discoverers and passes the obtained target features to the next discoverer. The next discoverer docks with the next predator, and the current discoverer continues to track the target vessel until the next discoverer or predator tracks the target vessel.

[0019] Preferably, the relay matching sub-step also includes a predator value assessment strategy, which obtains the distance and movement trajectory between the current predator and the target predator, and the current predator has a preset cross-regional parameter. The predator tracking cost is calculated, the tracking costs of the current predator and the target predator are compared, and the predator with the lower tracking cost is selected to track the target vessel.

[0020] A drone-based intelligent inspection, identification, and analysis system for anomalies, including

[0021] The target search module allows the discoverer to determine whether a vessel has any abnormalities based on the searched vessels and the anomaly identification strategy. If an anomaly is found, the vessel is marked as the target vessel and tracked.

[0022] The predator docking module, during the process of the discoverer tracking the target vessel, obtains the position of the target vessel and the status of the surrounding predators, selects the corresponding predator, and adjusts the distance between the discoverer and the target vessel and the predators through the tracking strategy, adjusts the flight path of the discoverer, and transmits the target characteristics of the target vessel to the selected predator. The target characteristics include the vessel's position, anchor point, and abnormal situations.

[0023] The evidence collection process involves the predator monitoring the target vessel to determine if any anomalies exist. If anomalies are found, evidence frames are acquired. If no anomalies are found, keyframes with target features are acquired by the discoverer, along with detailed images of the target vessel collected by the predator. These are cross-referenced to form a complete chain of evidence.

[0024] The beneficial effects of this invention are as follows: In the target search step, the discoverer makes an initial judgment on the vessel based on an anomaly identification strategy, marking target vessels that may exhibit anomalies. Subsequently, in the evidence collection step, the predator monitors and judges the target vessel again. This multi-stage anomaly judgment mechanism greatly reduces the possibility of misjudgment and improves the accuracy of anomaly detection. For example, in complex marine environments, some interfering factors may cause deviations in the initial judgment, but these interferences can be effectively eliminated after the predator's reconfirmation. In the target vessel identification step, the predator locks onto the target based on the anchor points on the target vessel using a feature recognition strategy. These anchor points are carefully selected and determined in the target search step, possessing high uniqueness and recognizability. In this way, it can be ensured that the predator accurately finds the target vessel, avoiding target misjudgment in complex scenarios. In the predator docking step, the discoverer is responsible for tracking the target vessel, obtaining the target vessel's position and the status of surrounding predators, and selecting a suitable predator for docking. At the same time, the discoverer adjusts its own flight path to transmit the target vessel's target features to the selected predator. This collaborative model allows both the discoverer and the predator to fully leverage their respective strengths, achieving efficient task allocation and execution. For example, the discoverer possesses broader search capabilities, while the predator boasts more precise target locking and evidence collection capabilities. In the evidence collection phase, the predator not only monitors the target vessel in real time, acquiring potentially anomaly evidence frames, but also obtains keyframes with target features collected by the discoverer, cross-referencing them with its own detailed images of the target vessel. This multi-source evidence collection and cross-comparison method forms a complete chain of evidence, providing strong support for subsequent anomaly analysis and processing. For instance, when handling vessel violations, a complete chain of evidence can accurately prove the time, location, and specific circumstances of the violation. Furthermore, in the predator docking phase, the discoverer adjusts its flight path according to the tracking strategy, optimizing the distance between itself and both the target vessel and the predator. This intelligent path adjustment mechanism improves inspection efficiency and reduces unnecessary flight time and energy consumption. Attached Figure Description

[0025] Figure 1 This is an overall flowchart of the present invention;

[0026] Figure 2 This is a step diagram of the present invention;

[0027] Figure 3 This is a schematic diagram of the tracking strategy of the present invention;

[0028] Figure 4 This is a module connection diagram of the present invention;

[0029] Figure 5 This is the target vessel identification diagram of the present invention. Detailed Implementation

[0030] 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.

[0031] It should be noted that when a component is described as "fixed to" another component, it can be directly on the other component or may have a component in between. When a component is considered "connected to" another component, it can be directly connected to the other component or may have a component in between. When a component is considered "set on" another component, it can be directly set on the other component or may have a component in between. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0032] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0033] The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings:

[0034] Methods for intelligent inspection, identification, and analysis of drone anomalies, including

[0035] The target search process involves the discoverer identifying vessels and using anomaly detection strategies to determine if any abnormalities exist. If anomalies are found, the vessel is designated as the target and tracked. The discoverer, equipped with a multi-sensor algorithm, searches the patrol area using cameras, infrared sensors, and radar. It performs a comprehensive scan of the area, capturing images at a specific frame rate, detecting heat signatures and distance from the vessels using infrared sensors, and monitoring their position and operational status in real-time using radar. A pre-trained convolutional neural network model identifies vessel appearance features and detects abnormal emissions or oil leaks, comparing these features with normal data. The system detects the image features of ships. If the deviation between the detected ship features and normal features exceeds a preset threshold, it is judged as an abnormal situation. The system also compares the temperature data of the ship and the surrounding water detected by the infrared sensor with normal temperature data. If the temperature of the water around the ship is significantly abnormal compared with the temperature of other areas, the ship is suspected of having abnormal emissions. When an abnormal situation is detected, the system marks the target ship and marks its outline in the image, assigns it a unique identification number, and predicts its future position based on the target ship's current position and direction of movement. The system adjusts its flight path to continuously maintain the detection of the target ship and continuously updates the target ship's position information during the tracking process, storing relevant data such as the abnormal situation.

[0036] The predator docking process involves the discoverer tracking the target vessel, acquiring its position and the status of surrounding predators, selecting a suitable predator, and adjusting its flight path by varying the distance between itself, the target vessel, and the predators. The discoverer transmits the target vessel's characteristics, including its position, anchorage, and any unusual activity, to the selected predator. The discoverer uses its positioning system and continuous tracking to acquire the target vessel's latitude and longitude coordinates, speed, and heading. It establishes communication links with surrounding predators to obtain their status, including position, remaining battery power, device operational status (camera functionality, data storage capacity, etc.), and whether they are busy. Based on this predator status, an optimized U-shaped Anze algorithm is used to select a suitable predator, prioritizing those closer to the target vessel. Shorten response time; select predators with sufficient power, fault-free equipment, and in an idle state; the discoverer dynamically adjusts the distance between the target vessel, the selected non-direct-light location, and current environmental factors through a tracking strategy to ensure that the discoverer can continuously track the target vessel and successfully dock with the predator; the discoverer transmits the target vessel's characteristics to the selected predator, including the vessel's position, to guide the predator to quickly approach the target; anchor points, such as unique stardust or colored structural parts on the vessel, serve as key references for the predator to subsequently lock onto the target, and detailed descriptions of anomalies, such as the type of anomaly and its location, help the predator understand the target situation in advance and conduct targeted monitoring.

[0037] The target vessel identification process involves the following steps: After acquiring target features, the predator flies towards the target vessel's location. Once the target vessel enters the predator's acquisition range, it locks onto the vessel using a feature recognition strategy based on the anchor points on the target. Upon receiving the target features, the predator uses its built-in navigation algorithm and positioning system to fly towards the target vessel's location. During flight, it adjusts its flight path in real-time based on the target vessel's updated position information from the discoverer to ensure accurate approach. Once the target vessel enters the predator's image acquisition range, the predator uses a high-definition camera to capture images of the target vessel. Using image recognition algorithms, it identifies the anchor points provided by the discoverer. A feature point matching algorithm is employed to find areas in the acquired images that match the anchor points, thus accurately pinpointing the target vessel's location in the image. After identifying the anchor points, the predator further compares and verifies the target vessel's overall appearance features, navigation attitude, and other information with the target features provided by the discoverer. If the matching degree of each feature reaches a set threshold, the vessel is confirmed as the target, and the target is locked.

[0038] The evidence collection process involves the predator monitoring the target vessel to determine if any anomalies exist. If an anomaly is found, evidence frames are acquired. If no anomalies are found, keyframes containing target features captured by the discoverer are acquired, along with detailed images of the target vessel collected by the predator. These are cross-referenced to form a complete chain of evidence. The predator utilizes its onboard specialized monitoring equipment, such as high-resolution cameras, multispectral sensors, and gas detectors, to comprehensively monitor the target vessel. Image analysis algorithms and sensor data analysis are then used to determine if the target vessel exhibits any anomalies. For example, multispectral sensors are used to detect the spectral characteristics of oil spills on the sea surface, and gas detectors detect abnormal pollutant emissions. If anomalies are detected, the predator immediately acquires evidence frames. For image-based evidence, high-resolution images containing clear details of the anomalous areas are captured to ensure the anomaly is clearly identifiable. For other types of evidence, such as gas detection data and pollutant sample data, accurate recording and storage are performed. If no anomalies are detected, the predator acquires keyframes containing target features captured by the discoverer, along with detailed images of the target vessel collected by itself. By cross-referencing the two sets of data, any anomalies missed due to differences in monitoring equipment accuracy or monitoring angles are checked. By comparing and analyzing the vessel's external details and structural integrity, it is ensured that no anomalies are overlooked. Evidence frames obtained by the predator, keyframes from the discoverer, and related sensor data, monitoring time, and location information are integrated to form a complete chain of evidence. The evidence is numbered, timestamped, and accompanied by relevant descriptions to establish an evidence database for subsequent retrieval, analysis, and processing. The formation of this chain of evidence ensures the accuracy and traceability of monitoring anomalies on target vessels, while preventing situations where vessels cease abnormal behavior after the discoverer or predator approaches, thus preventing the inability to promptly punish the vessel.

[0039] The tracking strategy includes acquiring the target vessel's speed, trajectory, and environmental visibility. It also calculates the tracking value by determining the positional relationships between the target vessel, the predator, and the discoverer, and using a data index table to obtain corresponding docking and positioning weights. A preset tracking threshold is established; if the tracking value is less than the threshold, an adjustment strategy is generated to adjust the docking and positioning weights to ensure the tracking value exceeds the threshold. The system acquires the target vessel's speed and trajectory, and environmental visibility, considering its impact on image analysis accuracy. Based on the acquired data, the system queries the data index table. This pre-established index table contains docking and positioning weights for different UAV performance levels. Using these weights, the system calculates the tracking value. The preset tracking threshold serves as a standard for evaluating tracking effectiveness. When the calculated tracking value is less than the threshold, the current tracking strategy is ineffective and requires adjustment. By employing a dynamic tracking strategy, it is ensured that the predator remains within the observer's communication range during the tracking process, preventing the observer from losing contact with the predator due to the target vessel's excessive speed. At the same time, it accurately locks onto the target vessel, preventing the observer from being unable to effectively guide the predator due to limited field of vision when too close to the target vessel, and also preventing the target vessel from being lost due to decreased positioning accuracy caused by distance from the orchard.

[0040] Formula 1

[0041] ;

[0042] in, This is the spatiotemporal visibility coupling value. For time Normalized environmental visibility at any given time. For the maximum visibility of the discoverer's theory, The instantaneous speed of the target ship. For reference speed, For signal attenuation rate, The time oscillation period;

[0043] Specific calculation formula

[0044] =0.9, =20 sections =0.12 , =8s;

[0045] , =18 sections;

[0046] ;

[0047] ;

[0048] ;

[0049] Formula 2

[0050] ;

[0051] For the two-distance cooperative entropy, For complementary error functions, , This represents the real-time distance between the discoverer and the target vessel. The standard deviation of the observation error. It is a hyperbolic secant function. The real-time distance between the discoverer and the predator. This is the distance scaling factor. For smoothing parameters, To align with weights,

[0052] Specific calculation formula

[0053] =1.5km, =0.6, =1.2km;

[0054] =4km, =3km, =0.8;

[0055] ;

[0056] ;

[0057] ;

[0058] ;

[0059] right After correction, the result is 0.68;

[0060] Formula 3

[0061] ;

[0062] in, For trajectory locking factor, To determine the weight, Let be the prediction error vector for the k-th trajectory point. The error tolerance radius is... Azimuth The probability density function, It is just a function. At the effective observation angle The value is 1 if it is within the specified time, and 0 otherwise. The maximum observation angle is 30 degrees; data exceeding this angle are invalid.

[0063] Specific calculation formula

[0064] =0.9, ;

[0065] =0.4km, ;

[0066] ;

[0067] ;

[0068] Revised to 0.0966;

[0069] Revised to 1.18;

[0070] Formula 4

[0071] ;

[0072] In order to track value, This is the theoretical maximum value. , As a sensitivity index, To prevent zero factor, For emergency correction factors,

[0073] Specific calculation formula

[0074] =1.2, =1.2, =0.8, =0.01, =0.05;

[0075] ;

[0076] ;

[0077] ;

[0078] ;

[0079] ;

[0080] like Then it needs to be corrected.

[0081] The tracking strategy also includes a docking step, which comprises a data acquisition sub-step. This sub-step dynamically adjusts the communication and transmission frequencies based on the distance between the discoverer and predator, environmental conditions, and their flight speeds to obtain real-time communication quality between them. The distance between the discoverer and predator is a crucial factor affecting communication and signal transmission. Generally, the greater the distance, the more severe the signal attenuation, and the worse the communication quality may be. Environmental conditions, such as weather (rain, fog, strong winds, etc.) and electromagnetic interference, also significantly impact communication and signal transmission. The flight speed of both birds also affects communication; high-speed flight can lead to Doppler shift and other issues, further impacting communication quality. The system dynamically adjusts the communication and transmission frequencies. For example, when the distance is large or environmental interference is significant, the transmission frequency is appropriately increased to enhance signal strength, while the communication frequency is adjusted to avoid interference bands, thereby improving communication stability and reliability. By continuously adjusting the communication and transmission frequencies, the system can monitor the communication quality between the discoverer and predator in real time.

[0082] The docking distance calculation sub-step is used to obtain the docking weight and real-time communication quality, and to obtain the docking distance through a distance adjustment algorithm. Based on the docking distance, the system dynamically adjusts the flight speed and flight path between the discoverer and the predator. Combining the docking weight and real-time communication quality, the system uses a distance adjustment algorithm to calculate the docking distance. Based on the calculated docking distance, the system dynamically adjusts the flight speed and flight path between the discoverer and the predator. If the docking distance needs to be shortened, the discoverer and the predator may reduce their flight speed and adjust their flight path to gradually approach each other. If the docking distance can be increased, they may appropriately increase their flight speed and optimize their flight path to improve docking efficiency.

[0083] The distance adjustment algorithm is as follows:

[0084] Formula 1

[0085] ;

[0086] Here, Q is the communication quality degradation factor, and Q is the current communication quality. This is a communication quality threshold; if the value is below this, the departure distance will be adjusted. To align with weights, Here, represents the weight saturation parameter, and erfc represents the complementary error function. It is a hyperbolic secant function;

[0087] Specific calculation formula

[0088] Q = -85dB, =-80dB;

[0089] ;

[0090] ;

[0091] ;

[0092] ;

[0093] 1 - 0.486 = 0.514;

[0094] ;

[0095] Formula 2

[0096] ;

[0097] K(D) is the dynamic distance adjustment parameter. For the theoretically optimal distance, Distance sensitivity index This is the current actual distance. For distance parameter coefficients, As a reference distance, The Hessian matrix for communication quality. Frobenius norm, This represents the mutation suppression coefficient.

[0098] Specific calculation formula

[0099] =600m, =3, =1500m, =0.8, =1000m, =0.5;

[0100] ;

[0101] ;

[0102] ;

[0103] ;

[0104] ;

[0105] ;

[0106] Formula 3

[0107] ;

[0108] For docking distance, For the minimum allowable distance, For the maximum allowable distance, The threshold response exponent, To adjust the smoothing parameters, ;

[0109] Specific calculation formula

[0110] =400m, =800m, =0.2, =1.2;

[0111] ;

[0112] ;

[0113] ;

[0114] .

[0115] The tracking strategy also includes a positioning step, which acquires positioning weights, calculates the positioning distance based on these weights, and dynamically adjusts the discoverer's flight speed and route accordingly. The positioning distance is calculated based on the acquired positioning weights and the characteristics of different positioning technologies. For GPS positioning, its accuracy can ideally reach meter-level, but in practical applications it is affected by various factors. When acquiring the positioning distance, a relatively accurate positioning distance range is calculated based on the current GPS signal strength, the number of satellites, and the positioning error model. Based on the acquired positioning distance, the discoverer dynamically adjusts its flight speed and route in real time. If the positioning distance indicates that the distance between the discoverer and the target vessel is too close, approaching a range that may limit visibility or pose other risks (e.g., less than 100 meters), the discoverer will appropriately increase its flight speed to increase the distance from the target vessel, thereby expanding the monitoring field of view, while simultaneously adjusting its flight route to maintain an effective monitoring angle of the target vessel.

[0116] The predator docking strategy includes an anchor point acquisition step. This step acquires several frames of images collected by the discoverer as target images, and obtains image information of the target vessel and other vessels within these target images. By comparing the target vessel with the other vessels, several feature images from the target vessel are acquired as anchor points. Anchor point weights are then determined based on the frequency of these anchor points appearing on the other vessels. The discoverer continuously acquires images, selecting several frames as target images. These target images contain information about several vessels in the current scene. Using these frames, vessel features are captured from multiple angles and time points, avoiding inaccurate feature extraction due to vessel movement or limitations of single-frame images. The target images are then processed using a target detection algorithm to quickly identify the target vessel's position within the image and obtain image information of the target vessel and other vessels. The target vessel is compared with the other vessels to find unique images within it as anchor points. These features can be the vessel's outline, color distribution, texture, etc. If a ship has a uniquely shaped smokestack that other ships do not, this smokestack feature can be used as an anchor point. A feature extraction algorithm is used to extract feature points from the image, and based on the comparison results, features unique to the target ship are selected as anchor points. The frequency of the anchor point appearing on other ships is counted. If an anchor point appears frequently on other ships, it indicates low uniqueness and poor distinguishability in subsequent feature recognition, thus it is assigned a lower weight. Conversely, if an anchor point rarely appears on other ships, it indicates strong uniqueness and can effectively distinguish the target ship from other ships, thus it is assigned a higher weight.

[0117] The feature recognition strategy includes acquiring anchor points and using images collected by the predator as images to be analyzed. It calculates the similarity between the ship images and anchor point images in the images to be analyzed, calculates similarity scores based on anchor point weights, prioritizes similarity scores, and selects the ship with the highest similarity score as the target ship. The strategy also involves acquiring anchor points and their weights obtained in the anchor point acquisition step; images collected by the predator device are used as images to be analyzed, containing ship information in the current scene. Since the predator device may be in different positions and at different angles, the images it collects may differ from the target images collected by the discoverer; therefore, feature recognition is needed on these images to determine the target ship. The similarity between the ship images and anchor point images in the images to be analyzed can be calculated using image matching algorithms. For each anchor point, a similarity score is calculated between it and the ship images in the images to be analyzed. Then, these similarity scores are weighted and summed according to the weights of the anchor points to obtain the comprehensive similarity score between each ship and the target ship. For example, if there are three anchor points A, B, and C with weights of 0.2, 0.3, and 0.5 respectively, and the similarity scores of a ship in the image to be analyzed with these three anchor points are 0.8, 0.6, and 0.7 respectively, then the comprehensive similarity score between this ship and the target ship is 0.2 × 0.8 + 0.3 × 0.6 + 0.5 × 0.7 = 0.69. Ships in the image to be analyzed are prioritized according to their comprehensive similarity scores; ships with higher scores are more likely to be the target ship. Finally, the ship with the highest similarity score is selected as the target ship.

[0118] The predator docking process also includes a relay matching sub-step. When the current discoverer is tracking the target vessel and the predator has not entered its predation range, and the target vessel is about to leave the current discoverer's monitoring range, the current discoverer searches for other discoverers and passes the obtained target characteristics to the next discoverer. The next discoverer then docks with the next predator, and the current discoverer continues to track the target vessel until the next discoverer or predator tracks it. During the current discoverer's tracking of the target vessel, the positional status of the predator and the target vessel is continuously monitored. When the predator has not entered its predation range and the target vessel is about to leave the current discoverer's monitoring range, the relay matching mechanism is triggered. This situation may be due to factors such as the target vessel's high speed or the current discoverer's limited monitoring range. The current discoverer will begin searching for other available discoverers; through a pre-established communication network or location information system, the current discoverer can quickly obtain the location and status information of other discoverers; the current discoverer will transmit the target vessel's characteristic information obtained in the anchor point acquisition step, such as the anchor point and anchor point weight, to the selected next discoverer to help the next discoverer quickly and accurately locate the target vessel; after receiving the target characteristic information, the next discoverer will connect with the next predator and guide it to the target vessel's location; after completing the information transmission and connection, the current discoverer will not immediately stop tracking the target vessel, but will continue tracking until the next discoverer or predator successfully takes over the tracking of the target vessel, ensuring that the target vessel is not lost during the relay.

[0119] The relay matching sub-step also includes a predator value assessment strategy. This involves acquiring the distance and trajectory between the current predator and the target predator, and preserving the current predator's cross-regional parameters. The tracking cost of each predator is calculated, compared to the tracking costs of the current predator and the next predator, and the predator with the lower tracking cost is selected to track the target vessel. Based on the acquired distance, trajectory, and cross-regional parameters, the tracking costs for both the current and target predators are calculated. This cost calculation considers multiple factors, such as the greater the distance, the more energy and time are consumed during the tracking process, resulting in a higher cost. The complex movement trajectory requires more turning and acceleration operations, which also increases the cost. The cross-regional parameters will be adjusted according to the actual area crossed. If the current predator needs the current detection area to track the target vessel, the cross-regional cost will be obtained, and the cross-regional cost will be adjusted accordingly based on factors such as the signal strength of the next area. If the signal in the next detection area is weak, the cross-regional cost will increase accordingly. By comparing the tracking costs of the current predator and the next predator, the predator with the lower tracking cost is selected to track the target vessel, ensuring that continuous tracking of the target vessel can be achieved at the lowest cost during the relay matching process.

[0120] A drone-based intelligent inspection, identification, and analysis system for anomalies, including

[0121] The target search module allows the discoverer to determine whether a vessel has any abnormalities based on the searched vessels and the anomaly identification strategy. If an anomaly is found, the vessel is marked as the target vessel and tracked.

[0122] The predator docking module, during the process of the discoverer tracking the target vessel, obtains the position of the target vessel and the status of the surrounding predators, selects the corresponding predator, and adjusts the distance between the discoverer and the target vessel and the predators through the tracking strategy, adjusts the flight path of the discoverer, and transmits the target characteristics of the target vessel to the selected predator. The target characteristics include the vessel's position, anchor point, and abnormal situations.

[0123] The evidence collection process involves the predator monitoring the target vessel to determine if any anomalies exist. If anomalies are found, evidence frames are acquired. If no anomalies are found, keyframes with target features are acquired by the discoverer, along with detailed images of the target vessel collected by the predator. These are cross-referenced to form a complete chain of evidence.

[0124] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for abnormal intelligent inspection, identification and analysis of unmanned aerial vehicles, characterized in that, include In the target search step, the discoverer determines whether there is any abnormality in the search vessel based on the search vessel and the anomaly identification strategy. If there is an anomaly, the vessel is marked as the target vessel and tracked. In the predator docking process, during the tracking of the target vessel, the discoverer acquires the target vessel's position and the status of surrounding predators, selects the corresponding predator, and adjusts the distance between the discoverer, the target vessel, and the predators using a tracking strategy. The discoverer also adjusts its flight path and transmits the target vessel's characteristics, including its position, anchoring point, and any abnormal situations, to the selected predator. The tracking strategy includes acquiring the target vessel's speed, trajectory, and environmental visibility. Based on the positional relationships between the target vessel, predators, and discoverer, and according to a data index table, it obtains the corresponding docking weight and positioning weight, calculates the tracking value, and presets a tracking threshold. When the tracking value is less than the tracking threshold, an adjustment strategy is generated to adjust the docking weight and positioning weight so that the tracking value is greater than the tracking threshold. The predator docking step also includes a relay matching sub-step. When the predator does not enter the predation range during the current discoverer's tracking process, and the target vessel is about to leave the current discoverer's monitoring range, the current discoverer searches for other discoverers and passes the obtained target characteristics to the next discoverer. The next discoverer docks with the next predator, and the current discoverer continues to track the target vessel until the next discoverer or predator tracks the target vessel. The relay matching sub-step also includes a predator value assessment strategy, which obtains the distance and movement trajectory between the current predator and the target predator, and the current predator has a preset cross-region parameter. The predator tracking cost is calculated, the tracking costs of the current predator and the target predator are compared, and the predator with the lower tracking cost is selected to track the target vessel. The target vessel identification process involves the predator acquiring target characteristics and then flying towards the target vessel's location. Once the target vessel enters the predator's collection range, it locks onto the target vessel using a feature recognition strategy based on the anchor points on the target vessel. The evidence collection process involves the predator monitoring the target vessel to determine if any anomalies exist. If anomalies are found, evidence frames are acquired. If no anomalies are found, keyframes with target features are acquired by the discoverer, along with detailed images of the target vessel collected by the predator. These are cross-referenced to form a complete chain of evidence. 2.The UAV abnormal intelligent patrol, identification and analysis method of claim 1, wherein, The tracking strategy also includes an integration step, which includes: The data acquisition sub-step is used to dynamically adjust the communication frequency and transmission frequency based on the distance between the discoverer and the predator, environmental conditions, and their flight speeds, in order to obtain the real-time communication quality between the discoverer and the predator. The docking distance calculation sub-step is used to obtain the docking weight and real-time communication quality, and to obtain the docking distance through a distance adjustment algorithm. Based on the docking distance, the flight speed and flight path between the discoverer and the predator are dynamically adjusted. 3.The UAV abnormal intelligent patrol, identification and analysis method of claim 1, wherein, The tracking strategy also includes a positioning step, which is used to obtain a positioning weight, obtain a positioning distance based on the positioning weight, and dynamically adjust the flight speed and flight path of the discoverer based on the positioning distance. 4.The UAV abnormal intelligent patrol, identification and analysis method of claim 1, wherein, The predator docking strategy includes an anchor point acquisition step. This step involves acquiring several frames of images collected by the discoverer as target images, acquiring the target vessel and other vessel image information in the target images, comparing the target vessel with the other vessels, acquiring several feature images of the target vessel as anchor points, and acquiring anchor point weights based on the frequency of the anchor points appearing on the other vessels.

5. The method of claim 4, wherein, The feature recognition strategy includes obtaining anchor points and acquiring images collected by predators as images to be analyzed, calculating the similarity between the ship images and anchor point images in the images to be analyzed, calculating the similarity based on the anchor point weights, prioritizing the similarity based on the similarity, and selecting the ship with the highest similarity as the target ship.

6. The unmanned aerial vehicle abnormal intelligent inspection, identification and analysis system, used to realize the unmanned aerial vehicle abnormal intelligent inspection, identification and analysis method of any one of claims 1-5, characterized in that, include The target search module allows the discoverer to determine whether a vessel has any abnormalities based on the searched vessels and the anomaly identification strategy. If an anomaly is found, the vessel is marked as the target vessel and tracked. The predator docking module, during the process of the discoverer tracking the target vessel, obtains the position of the target vessel and the status of the surrounding predators, selects the corresponding predator, and adjusts the distance between the discoverer and the target vessel and the predators through the tracking strategy, adjusts the flight path of the discoverer, and transmits the target characteristics of the target vessel to the selected predator. The target characteristics include the vessel's position, anchor point, and abnormal situations. The evidence collection process involves the predator monitoring the target vessel to determine if any anomalies exist. If anomalies are found, evidence frames are acquired. If no anomalies are found, keyframes with target features are acquired by the discoverer, along with detailed images of the target vessel collected by the predator. These are cross-referenced to form a complete chain of evidence.

Images