Method and system for identifying flight state of unmanned aerial vehicle

Abstract

The invention discloses an unmanned aerial vehicle flight state identification method and system, and relates to the technical field of identification methods, and the method comprises the steps: determining a panoramic image of a target animal and peripheral images of the target animal according to a plurality of images and a distribution map of a target park; and the current event of the target animal is determined according to the panoramic image of the target animal and the peripheral image of the target animal, so that intelligent identification of the current event of the target animal is realized, and the current event of the target animal is accurately controlled. Therefore, a plurality of sub-events are determined according to division of the current event of the target animal, and the abnormal event of the target animal is determined according to the plurality of sub-events; the abnormal picture is determined based on the abnormal event of the target animal, the panoramic image of the target animal and the peripheral image of the target animal, and the optimized picture is determined based on the abnormal event of the target animal and the optimized mapping picture relationship, so that intelligent analysis of the unmanned aerial vehicle on the abnormal event is realized, and the accuracy of optimization measures is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of identification methods, and in particular to a method and system for identifying a drone in flight. Background Art

[0002] With the development of science and technology, drones fly in the air and fly along patrol routes. The cameras mounted on the drones take dynamic photos as the drones fly. In the existing technology, multiple images taken by the drones are collected, and an image of the target animal is determined based on the screening of multiple images. Then, a single-dimensional recognition is performed on the image, without considering the surrounding images of the target animal. This affects the precise control of the current events of the target animal. Summary of the Invention

[0003] The purpose of the present invention is to overcome the deficiencies of the prior art and to provide a method and system for identifying a UAV in flight.

[0004] An embodiment of the present invention provides a method for identifying a drone in flight state, comprising:

[0005] While the UAV is in flight, the UAV collects multiple images of the target park;

[0006] determining a panoramic image of a target animal and a surrounding image of the target animal based on the plurality of images and a distribution map of the target park;

[0007] determining a current event of the target animal based on a panoramic image of the target animal and an image of the surrounding area of ​​the target animal;

[0008] Determining a plurality of sub-events according to the division of the current event of the target animal, and determining an abnormal event of the target animal according to the plurality of sub-events;

[0009] An abnormal picture is determined based on the abnormal event of the target animal, the panoramic image of the target animal and the surrounding image of the target animal, and an optimized picture is determined based on the abnormal event of the target animal and the optimized mapping picture relationship to output corresponding optimization measures.

[0010] An embodiment of the present invention provides a system for identifying a drone in flight state. The system for identifying a drone in flight state is applied to the above-mentioned method for identifying a drone in flight state. The system for identifying a drone in flight state includes:

[0011] an acquisition module, configured to enable the drone to acquire multiple images of a target park when the drone is in a flight state;

[0012] An image module, configured to determine a panoramic image of a target animal and an image of the surrounding area of ​​the target animal based on the multiple images and a distribution map of the target park;

[0013] a feature module, configured to determine a current event of a target animal based on a panoramic image of the target animal and images of the surrounding areas of the target animal;

[0014] An abnormal event module, configured to determine a plurality of sub-events according to the division of the current event of the target animal, and determine the abnormal event of the target animal according to the plurality of sub-events;

[0015] The optimization module is used to determine abnormal images based on abnormal events of the target animal, panoramic images of the target animal and surrounding images of the target animal, and to determine optimized images based on the abnormal events of the target animal and the relationship between the optimized mapping images to output corresponding optimization measures.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] In an embodiment of the present invention, through the method in the embodiment of the present invention, when the drone is in a flying state, the drone collects multiple images of the target park; determines a panoramic image of the target animal and the surrounding images of the target animal based on the multiple images and the distribution map of the target park; determines the current event of the target animal based on the panoramic image of the target animal and the surrounding images of the target animal, and is compatible with the overall consideration of the panoramic image of the target animal and the surrounding images of the target animal, realizes the intelligent recognition of the current event of the target animal, and accurately controls the current event of the target animal.

[0018] Therefore, multiple sub-events are determined based on the division of the current events of the target animal, and the abnormal events of the target animal are determined based on the multiple sub-events; the abnormal screen is determined based on the abnormal events of the target animal, the panoramic image of the target animal and the surrounding images of the target animal, and the optimized screen is determined based on the abnormal events of the target animal and the optimized mapping screen relationship to output corresponding optimization measures, thereby realizing the intelligent analysis of abnormal events by the drone and ensuring the accuracy of the optimization measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 1 is a flow chart of a method for identifying a UAV in flight state according to an embodiment of the present invention;

[0020] Figure 2 1 is a flow chart of step S11 in the method for identifying the flight status of a UAV in an embodiment of the present invention;

[0021] Figure 3 1 is a flow chart of step S12 in the method for identifying the flight status of a UAV in an embodiment of the present invention;

[0022] Figure 4 1 is a flow chart of step S13 in the method for identifying the flight status of a UAV in an embodiment of the present invention;

[0023] Figure 5 1 is a flow chart of step S14 in the method for identifying the flight status of a UAV in an embodiment of the present invention;

[0024] Figure 6 1 is a flow chart of step S15 in the method for identifying the flight status of a UAV in an embodiment of the present invention;

[0025] Figure 7 Schematic diagram of the structure of the identification system of the UAV in flight state in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention.

[0027] See also Figures 1 to 7 A method for identifying a UAV in flight state is applied to intelligently identify scenarios during flight. The method for identifying a UAV in flight state includes:

[0028] Step S11: When the UAV is in flight, the UAV collects multiple images of the target park;

[0029] Step S12: determining a panoramic image of the target animal and surrounding images of the target animal based on the multiple images and the distribution map of the target park;

[0030] Step S13: determining the current event of the target animal based on the panoramic image of the target animal and the surrounding images of the target animal;

[0031] Step S14: determining a plurality of sub-events according to the division of the current event of the target animal, and determining an abnormal event of the target animal according to the plurality of sub-events;

[0032] Step S15: determining an abnormal image based on the abnormal event of the target animal, the panoramic image of the target animal, and the surrounding image of the target animal, and determining an optimized image based on the abnormal event of the target animal and the relationship between the optimized mapping images, so as to output corresponding optimization measures;

[0033] refer to Figure 2 ,In step S11, when the UAV is in a flying state, the UAV collects multiple images of the target park;

[0034] In the specific implementation process of the present invention, the specific steps are:

[0035] S111: When the drone is above the target park, an inspection route of the drone relative to the target park is determined based on the shape of the target park and the current position of the drone, and the drone flies along the inspection route;

[0036] S112: During the flight of the UAV, the follow-up shooting modes of the multiple cameras are determined according to the relative positions of the UAV and the target park and the orientations of the multiple cameras. Based on the follow-up shooting mode, the multiple cameras are triggered to shoot the target park in different directions to collect multiple images of the target park.

[0037] In an embodiment of the present application, when the drone is above the target park, the inspection route of the drone relative to the target park is determined based on the shape of the target park and the current position of the drone. The drone flies along the inspection route, which takes into account the overall consideration of the shape of the target park and the current position of the drone, thereby ensuring the accuracy of the drone's inspection route relative to the target park.

[0038] At this point, the drone is above the target park. It takes off from the take-off point and rises to a predetermined flight altitude. At this altitude, the drone should be able to clearly see the entire target park. The drone uses the GPS system to determine its precise position and matches it with the pre-acquired target park map, which helps the drone understand its position relative to the target park.

[0039] The drone uses its camera or pre-loaded map data to analyze the target park's morphology, including its boundaries, major attractions, lakes, and tree distribution. Based on this analysis, the drone uses a path planning algorithm to generate one or more inspection routes that cover the park's main areas while taking into account the drone's flight restrictions (such as obstacle avoidance and altitude restrictions).

[0040] During flight, a drone needs to sense its own position in real time, which is usually achieved through GPS and an inertial navigation system (INS). The drone matches its real-time position with the preset inspection route. If it deviates from the route due to wind, obstacles, or other factors, the drone should be able to automatically adjust its flight trajectory to ensure it flies along the correct route.

[0041] The drone flies along the planned inspection route and makes necessary flight adjustments based on real-time perceived information (such as wind speed, obstacles, etc.); during the flight, the drone also needs to trigger image acquisition based on specific points or areas along the inspection route to obtain detailed image data of the target park.

[0042] Specifically, assuming the target park is a large urban park with lakes, hills, playgrounds, and multiple walking trails, the drone takes off from a takeoff point near the park and rises to an altitude of about 100 meters to clearly see the entire park. The drone first uses GPS to locate its own position and matches it with a pre-loaded park map. By analyzing the map, the drone identifies the park's main attractions (such as lakes and hills) and boundaries. The drone then uses a path planning algorithm to generate a patrol route that starts at the park's entrance, follows the main walking trails through the lakes and hills, and finally reaches the playground.

[0043] During the flight, the drone senses its own position in real time and makes flight adjustments based on the preset inspection route. When the drone flies near a lake, it triggers the camera to collect images to obtain a clear image of the lake. Similarly, when the drone flies to a hill or a playground, it also performs corresponding image collection. Finally, the drone completes the flight and image collection tasks of the entire park along the planned inspection route. These image data are used for subsequent analysis and processing, such as animal behavior monitoring, crowd density analysis, and park facility maintenance.

[0044] Furthermore, during the flight of the UAV, the follow-up shooting mode of multiple cameras is determined according to the relative position of the UAV and the target park and the orientation of multiple cameras. Based on the follow-up shooting mode, multiple cameras are triggered to shoot the target park in different directions to collect multiple images of the target park. This is compatible with the overall consideration of the relative position of the UAV and the target park and the orientation of multiple cameras, ensuring the accuracy of the follow-up shooting mode of multiple cameras.

[0045] At this time, during the flight, the drone uses sensors such as GPS and inertial navigation system (INS) to perceive its own position, altitude, speed and attitude in real time; the drone matches the real-time position information with the pre-loaded target park map to determine its specific position and direction relative to the target park.

[0046] The multiple cameras mounted on the drone have different orientations and viewing angles, including forward, rear, side, and downward views. Based on the drone's flight status and the relative orientation of the target park, the camera's orientation needs to be dynamically adjusted to ensure that key areas of the target park are captured. Furthermore, the tracking shooting mode automatically adjusts shooting parameters (such as angle, focal length, and exposure) to capture the optimal image based on the drone's flight status and the relative orientation of the target park. Based on the drone's real-time position, speed, and attitude, as well as map information of the target park, the drone's control system selects the most appropriate tracking shooting mode and triggers the camera to capture the image.

[0047] Once the follow-up shooting mode is determined, the control system on the drone sends a shooting instruction to the corresponding camera; the camera shoots the target park along the preset or dynamically adjusted direction according to the received instruction, and stores the collected image data in the drone's storage device.

[0048] Specifically, the drone is equipped with four cameras, facing forward, backward, left and right respectively. During flight, the drone uses GPS and INS sensors to perceive its own position, altitude and attitude in real time. At the same time, the drone matches the real-time location information with a pre-loaded large-scale urban park map to determine its specific position and direction relative to the protected area.

[0049] When the drone flew over an open area, it identified a group of rare birds moving below. At this time, the control system on the drone analyzed the orientation of the four cameras and the position of the target bird flock, and determined an optimal follow-up shooting mode: the front-view camera captured a panoramic view of the bird flock at a certain downward angle, and the side-view cameras adjusted their angles to capture the left and right sides of the bird flock. Once the follow-up shooting mode was determined, the control system on the drone immediately sent shooting instructions to the corresponding cameras. The front-view camera and the side-view cameras captured the bird flock along the preset direction according to the received instructions, and stored the collected image data in the drone's storage device.

[0050] In one embodiment of the present application, it is assumed that a drone is flying over a large urban park with a lake, a hill, a recreation area, and an animal sanctuary. The drone is equipped with four cameras, facing forward, backward, left, and right, respectively. A motion-following shooting mode matching table is collected. The motion-following shooting mode matching table is shown in Table 1:

[0051] Table 1 Matching table of follow-up shooting mode

[0052] Drone position / orientation Camera direction Target park characteristics Follow-up shooting mode Over the lake Forward Lake Panorama Panoramic shooting Near the hill Side view (left) Hill side view Side shot Over the amusement area Looking down Crowd Activities Close-up shot (crowd) Edge of the animal sanctuary rearview Animals Tracking (animals)

[0053] When the drone flies over a lake, it checks the matching table and determines that the front-view camera should shoot the lake in panoramic shooting mode; similarly, when the drone flies near a hill, it selects the side-view (left) camera to shoot the hill in profile shooting mode. In this way, the drone automatically selects the best follow-up shooting mode based on different flight positions and target park features.

[0054] refer to Figure 3 , in step S12, determining a panoramic image of the target animal and a surrounding image of the target animal based on the multiple images and the distribution map of the target park;

[0055] In the specific implementation process of the present invention, the specific steps are:

[0056] S121: The drone determines a distribution map of the target park based on the name of the target park and a regional database, matches the distribution map of the target park with the multiple images, and adjusts the alignment positions of the multiple images. The drone determines an overall image based on the synthesis of the multiple images and marks the target animal.

[0057] S122: capturing the target animal and determining a panoramic image of the target animal based on the morphology and overall image of the target animal. In the panoramic image of the target animal, multiple sides of the target animal are presented along different directions, and the outline of each side can be clearly presented.

[0058] S123: The drone determines the influence range of the target animal according to the type of the target animal and the movement route of the target animal, and determines the surrounding image of the target animal according to the influence range of the target animal, the position of the target animal, and the overall image.

[0059] In an embodiment of the present application, the drone determines the distribution map of the target park based on the name of the target park and the regional database, and matches the distribution map of the target park with multiple images to adjust the alignment positions of the multiple images. The drone determines the overall image based on the synthesis of multiple images, marks the target animal, and introduces the target animal.

[0060] At this time, before performing a mission, the drone usually receives the name of the target park as part of the instruction; a regional database is stored inside the drone or in the cloud, which contains detailed information on multiple parks, including the park's name, boundaries, main attractions, roads, etc.; by matching the name of the target park, the drone retrieves the corresponding park distribution map in the regional database. This distribution map is a vector diagram that contains the park's precise boundaries and internal features.

[0061] The multiple images collected by the drone need to be preprocessed, such as denoising, contrast enhancement, and distortion correction, to improve matching accuracy. The drone uses image processing algorithms to extract feature points or feature areas from the distribution map and each image. These features are edges, corners, textures, etc. The drone matches the feature points on the distribution map with the feature points in the image, which usually involves complex algorithms such as scale-invariant feature transform (SIFT) and speeded up robust features (SURF). Once the feature points are successfully matched, the drone will calculate a transformation matrix that describes the geometric transformation of the image relative to the distribution map (such as translation, rotation, scaling, etc.).

[0062] The drone uses the calculated transformation matrix to transform each image to align it with the distribution map;

[0063] Sometimes, the drone also needs to further fine-tune the alignment of the images through iterative optimization algorithms to ensure the best alignment effect; the drone will stitch the aligned images together to form an overall image covering most or all of the target park area; in order to ensure the color consistency of the overall image, the drone also needs to perform color correction on the stitched image; the drone uses target detection algorithms (such as deep learning models) to detect and mark the location of the target animal in the overall image; if necessary, the drone will further identify the species of the animal and add the corresponding label to the image.

[0064] Specifically, suppose a drone is sent to a target park called "Green Park" for an animal monitoring mission; after receiving the name "Green Park", the drone accesses the regional database and retrieves the distribution map of the park; the distribution map shows information such as the park's boundaries, main roads, lakes, and vegetation distribution; the drone collects multiple images of the park during flight and preprocesses and extracts features from these images; then, the drone matches the feature points on the distribution map with the feature points in the image and calculates the transformation matrix of each image relative to the distribution map.

[0065] The drone applied a transformation matrix to each image, aligning it with the distribution map. Using an iterative optimization algorithm, the drone further fine-tuned the image alignment to ensure optimal alignment. The drone stitched the aligned images together to form a single image covering most of Green Park. The drone then performed color correction on the stitched image to ensure color consistency. Using a target detection algorithm, the drone detected and marked the location of a deer foraging near a lake within the overall image. Using a deep learning model, the drone also identified the deer species and added a corresponding label to the image. Through these steps, the drone successfully determined Green Park's distribution map, matched and aligned multiple images, and ultimately synthesized a single image covering most of the park, marking the location and species of the target animal.

[0066] Furthermore, the target animal is collected, and a panoramic image of the target animal is determined based on the morphology and overall image of the target animal. In the panoramic image of the target animal, multiple sides of the target animal are presented along different directions, and the outline of each side can be clearly presented, which is compatible with the overall consideration of the morphology of the target animal and the overall image, thereby ensuring the accuracy of the panoramic image of the target animal.

[0067] At this point, the drone first needs to identify the target animal in the overall image. This is usually achieved through a target detection algorithm, which can automatically locate the object of interest (in this case, the animal) in the image; once the target animal is identified, the drone will use a tracking algorithm to continuously track the animal to ensure that its position can be accurately captured when collecting panoramic images.

[0068] Drones determine the target animal's morphology by analyzing its shape, size, color and other characteristics. This helps the drone understand the animal's body shape and structure, and decide how to capture panoramic images to best show all sides of the animal. The drone also assesses the animal's posture, including whether it is still, walking, running or performing other activities, which helps the drone choose the best time to shoot to avoid blurry images when the animal moves.

[0069] Based on the target animal's morphology and overall image information, the drone will plan a shooting path to ensure that it can capture multiple sides of the animal in different directions. This path needs to take into account the animal's movement direction, the obstruction of the surrounding environment, and the drone's flight restrictions. The drone will set the camera's focal length, exposure time, shutter speed and other parameters to ensure clear images in different light and distances.

[0070] The drone flies along a planned path and captures images of the animal at each preset position. These images will cover multiple sides of the animal, forming a panoramic image. During the shooting process, the drone will check the image quality in real time to ensure that the outline of each side is clearly presented. If the image is found to be blurred or missing, the drone will adjust the shooting parameters or reshoot. After shooting is completed, the drone will use image processing algorithms to stitch multiple images together to form a panoramic image. During the stitching process, the drone will ensure that the outline of each side is seamlessly connected and the color and brightness of the overall image are consistent. After stitching and processing, the drone will generate a high-quality panoramic image in which multiple sides of the animal are clearly presented. This image is used for further analysis, research or display.

[0071] Specifically, suppose a drone is performing an animal monitoring mission in the "Green Park" park and successfully identifies a deer walking in the woods as the target animal; the drone uses a target detection algorithm to identify the deer's position in the overall image and uses a tracking algorithm to continuously track its movement; the drone analyzes the deer's morphological characteristics, including its tall body, slender limbs and unique horn shape; at the same time, the drone evaluates the deer's posture and finds that it is strolling leisurely and shows no signs of escape or attack.

[0072] Based on the deer's morphology and overall image information, the drone planned a shooting path that would revolve around the deer's walking trajectory to ensure that multiple sides of the deer, including the front, side, and back, could be captured. At the same time, the drone set appropriate shooting parameters to ensure clear images under different lighting conditions. The drone flew along the planned path and captured images of the deer at each preset position. These images covered multiple sides of the deer, from the front to the side and then to the back, and the outline of each side was clearly presented. After the shooting was completed, the drone used image processing algorithms to stitch the multiple images together to form a panoramic image of the deer. During the stitching process, the drone ensured that the outline of each side was seamlessly connected and that the color and brightness of the overall image remained consistent. In the final panoramic image, the deer's multiple sides, including the front, side, and back, were clearly presented, providing strong support for subsequent analysis, research, or display.

[0073] Therefore, the drone determines the influence range of the target animal based on the type of the target animal and the movement route of the target animal, and determines the surrounding image of the target animal based on the influence range of the target animal, the position of the target animal and the overall image, which is compatible with the overall consideration of the influence range of the target animal, the position of the target animal and the overall image, and ensures the accuracy of the surrounding image of the target animal.

[0074] At this time, the drone first needs to make a preliminary judgment on the animal's activity range and impact area based on the type of target animal and combined with knowledge of animal behavior; different types of animals have different activity ranges due to their different living habits and ecological needs; the drone also needs to consider the animal's adaptability to the current environment, such as water sources, food sources, shelter and other factors, which will affect the animal's activity range.

[0075] By tracking the movement of target animals and analyzing their activity patterns and habitual paths over a period of time, drones can more accurately determine their actual range of influence, including areas the animals frequently visit, length of stay, and interactions with other animals. If the animal's activity route changes, the drone needs to be able to adjust its range of influence in real time to reflect the animal's latest behavioral patterns.

[0076] Once the target animal's sphere of influence is determined, the drone will define a surrounding area based on this range and the target animal's current location. This area should be large enough to include all key areas of animal activity, while being precise enough to avoid unnecessary redundant information. The drone will collect images within this defined surrounding area. These images should be able to fully reflect the relationship between the animal and its surrounding environment, including terrain, vegetation, water sources, other potential food sources or shelters, etc. The collected images need to be screened and synthesized to ensure that the final surrounding image is both accurate and beautiful. The drone will use image processing technology to enhance the clarity, contrast and color saturation of the image to highlight key information.

[0077] Specifically, suppose a drone is performing a monitoring mission in the "Tranquil Lake" Nature Reserve, and the target animal is an adult male deer. After the drone identifies the target animal as a deer, it preliminarily determines that its activity range is near lakes, grasslands, and the edges of woods based on the deer's living habits and ecological needs. Deer usually need water for drinking, grasslands as a source of food, and woods provide shelter and a place to rest.

[0078] The drone tracked the deer's movements and found that it often started from the edge of the woods in the morning, crossed the grassland to the lake to drink water, and then returned to the woods to rest in the afternoon. Based on this information, the drone more accurately determined the deer's impact range, including the grassland areas it often visited, the edge of the lake, and the nearby woods. Based on the deer's impact range and current location, the drone defined a surrounding area, which covered the grassland, lake edge, and woods frequented by the deer. The drone then collected a series of images within this area, including scenes of the deer foraging on the grass, drinking water by the lake, and resting postures in the woods. These images not only show the deer's behavioral patterns, but also reflect its close relationship with the surrounding environment. Finally, the drone used image processing technology to screen and synthesize these images, generating a clear, beautiful, and information-rich surrounding image for subsequent analysis and research.

[0079] In one embodiment of the present application, an influence range matching table is collected, and the influence range matching table matches the influence range according to the animal type and the action route; the influence range matching table is shown in Table 2:

[0080] Table 2 Impact range matching table

[0081]

[0082] In this example, when the drone identifies the target animal as a deer and observes its movement route to the lake to drink water in the morning, it will determine the deer's influence range as a circular area with a radius of approximately 500 meters centered on the lake based on the influence range matching table.

[0083] refer to Figure 4 , in step S13, determining the current event of the target animal based on the panoramic image of the target animal and the surrounding image of the target animal;

[0084] In the specific implementation process of the present invention, the specific steps are:

[0085] S131: When the drone is in flight, the drone determines a plurality of side images based on segmentation of the panoramic image of the target animal, determines a plurality of first sub-features based on the plurality of side images, the morphology of the target animal, and the species of the target animal, and determines an action event of the target animal based on a combination of the plurality of first sub-features;

[0086] S132: The drone determines a plurality of orientation images based on the surrounding image of the target animal and the relative orientation of the target animal, determines a plurality of second sub-features based on recognition of the plurality of orientation images, and determines a plurality of surrounding events of the target animal based on the plurality of second sub-features and the corresponding orientations;

[0087] S133: Determine multiple event combinations based on the target animal's action event and multiple surrounding events of the target animal. The drone determines the current event of the target animal based on the multiple event combinations and time.

[0088] In an embodiment of the present application, when the drone is in flight, the drone determines multiple side images based on the division of the panoramic image of the target animal, determines multiple first sub-features based on the multiple side images, the morphology of the target animal, and the species of the target animal, and determines the action event of the target animal based on the combination of the multiple first sub-features, which is compatible with the overall consideration of multiple side images, the morphology of the target animal, and the species of the target animal, ensuring the accuracy of the multiple first sub-features.

[0089] At this time, the drone needs to collect and process images in a stable flight state to ensure that the panoramic images obtained are clear and without shaking; the drone should maintain an appropriate position and height so that it can fully capture the panoramic image of the target animal while avoiding getting too close and disturbing the animal; at the same time, the drone will first divide the collected panoramic image of the target animal, usually through an image processing algorithm to divide the image into multiple areas, each area corresponding to a side or perspective of the animal; the images of these areas are extracted from the panoramic image to form multiple side images, which should be able to clearly show the different side features of the animal.

[0090] The drone uses image processing technology and machine learning algorithms to analyze the animal's morphology in each side image, extracting features such as body shape, color, texture, and posture; combined with the species information of the target animal (obtained through the previous identification step), the drone can more accurately identify and understand the animal's specific characteristics and behavioral patterns; based on the above analysis, the drone determines a series of first sub-features related to the animal's movements, such as "extended limbs indicate running" and "lowered head indicates foraging".

[0091] The drone combines multiple identified first sub-features to form a complete feature set to describe the animal's current motion state; through pattern matching, machine learning models or rule reasoning, the drone determines the animal's specific motion events based on the feature set, such as "a deer is running" or "a bird is flying."

[0092] Specifically, suppose a drone is flying over a forest and monitors a moving deer. The drone maintains a stable flight at a moderate altitude to ensure that it can fully capture a panoramic image of the deer. The drone divides the collected panoramic image of the deer into four areas: front, back, left, and right, and extracts the corresponding side images, which clearly show the different side features of the deer.

[0093] The drone analyzes the deer in each side image, identifying morphological features such as tall stature, slender limbs, and brown color. Combined with the deer species information, it determines the first sub-features such as "rapid alternating movement of limbs" and "head remaining alert." At the same time, the drone combines the identified first sub-features to form a complete feature set. Through pattern matching and machine learning model analysis, the drone determines that the deer's current action event is "running fast, avoiding predators or looking for food."

[0094] Furthermore, the drone determines multiple orientation images based on the surrounding images of the target animal and the relative orientation of the target animal, determines multiple second sub-features based on the recognition of the multiple orientation images, and determines multiple surrounding events of the target animal based on the multiple second sub-features and the corresponding orientations, which is compatible with the overall consideration of the multiple second sub-features and the corresponding orientations, ensuring the accuracy of the multiple surrounding events of the target animal.

[0095] At this point, the drone will first obtain a wide range of images around the target animal, which is usually achieved through a wide-angle lens or panoramic shooting technology to ensure that a comprehensive view of the animal's surrounding environment is captured; the drone uses image processing technology and geographic location information (such as GPS data) to determine the relative position of the target animal in the surrounding images, as well as the orientation of various key environmental elements (such as other animals, vegetation, water sources, etc.) relative to the target animal; based on the relative orientation information, the drone extracts images of specific orientations from the surrounding images, and these images focus on specific areas or objects around the animal.

[0096] The drone uses advanced image recognition technology (such as deep learning algorithms) to analyze each azimuth image and identify key elements and features; the drone determines a series of second sub-features related to the animal's surrounding environment, including the presence of other animals, vegetation type, water source location, terrain features, etc.

[0097] The drone associates each second sub-feature with its position information in the surrounding image to form a comprehensive description of the environment; based on the feature-position association information, the drone uses pattern matching, logical reasoning or machine learning models to determine multiple events around the animal, such as "other deer are drinking water nearby" and "potential predators are nearby".

[0098] Specifically, suppose a drone monitors a group of deer foraging; the drone captures a panoramic image of the deer herd and determines the center position of the deer herd in the image; then, the drone extracts directional images of the front, back, left and right sides of the deer herd from the panoramic image, and these images focus on different areas around the deer herd.

[0099] The drone uses a deep learning algorithm to analyze each azimuth image and identify multiple secondary sub-features; for example, it identifies an open grassland in the front image, a small hill in the rear image, a group of approaching lions in the left image, and a river in the right image.

[0100] The drone associates the identified second sub-feature with its positional information and determines multiple events around the deer herd; for example, "there is an open grassland ahead, suitable for running quickly to escape danger"; "there is a small hill behind, which can be used as an emergency shelter"; "there is a lion approaching on the left, posing a potential threat"; "there is a river on the right, which can be used as a water source or escape route."

[0101] Therefore, multiple event combinations are determined based on the action events of the target animal and multiple surrounding events of the target animal. The drone determines the current event of the target animal based on the multiple event combinations and time, and is compatible with the overall consideration of the panoramic image of the target animal and the surrounding image of the target animal, thereby realizing the intelligent recognition of the current event of the target animal and precise control of the current event of the target animal.

[0102] At this point, the drone needs to integrate the previously identified action events of the target animal (such as running, foraging, etc.) with multiple events around it (such as other animals approaching, environmental changes, etc.); through logical analysis or algorithm processing, the drone generates multiple event combinations; each combination represents a hypothetical target animal's current situation or ongoing activity.

[0103] When analyzing event combinations, drones pay special attention to the time factor; time information helps drones understand the sequence, duration, and causal relationships of events; combining time information with the logical relationship of event combinations, drones can more accurately determine the current event of the target animal. This determination result is usually the most logical and realistic event description.

[0104] Specifically, suppose a drone detects a deer moving in the woods; action event: the drone first recognizes that the deer is walking slowly, which is a typical action event for deer; surrounding events: then, the drone identifies multiple events around the deer; for example, there is a berry bush on the left, and the deer is looking for food; not far to the right, there is a group of tourists taking pictures, which interferes with the deer; at the same time, the roar of airplanes comes from the sky, which does not directly affect the deer, but increases the noise level of the environment; event combination generation: based on this information, the drone generates multiple event combinations; for example, "the deer is looking for berries as food, and at the same time notices the tourists on the right and the aircraft noise in the sky"; "the deer is walking slowly, alert to the surrounding environment, and preparing to avoid tourists and aircraft noise."

[0105] Considering the temporal factor: The drone noticed that the deer's behavior was continuous, and there was a certain temporal sequence and logical relationship between each event. For example, the deer first showed interest in the berry bush, then noticed the presence of tourists, and finally responded to the roar of the airplane. Current event determination: Combining temporal information and the logical relationship between event combinations, the drone determined the deer's current event as follows: "The deer was walking slowly in the woods, its main goal being to find berries for food. At the same time, it remained alert to the tourists on its right and noticed the noise of the airplane in the sky, but had not yet shown obvious escape or attack behavior." This example demonstrates the workflow and effect of step S133 in actual application. By integrating the target animal's motion events and surrounding events and considering the temporal factor, the drone can more accurately determine the target animal's current event, providing valuable information for subsequent monitoring, research, and conservation actions.

[0106] In one embodiment of the present application, an event matching table is collected, and the event matching table is shown in Table 3:

[0107] Table 3 Event matching table

[0108]

[0109] The drone identifies that the target animal is "walking" and "near a food source (such as a berry bush)", and "there is no other obvious interference"; the drone searches the event matching table for a row that matches these event combinations; after finding the matching row, the drone determines that the current event description is "animal foraging"; at this time, the current event: animal foraging; event combination: walking, near a food source, no interference.

[0110] refer to Figure 5 , in step S14, a plurality of sub-events are determined according to the division of the current event of the target animal, and an abnormal event of the target animal is determined according to the plurality of sub-events;

[0111] In the specific implementation process of the present invention, the specific steps are:

[0112] S141: The drone collects the target animal's motion signal, divides the work according to the target animal's motion signal and the target animal's current event, and outputs multiple sub-events with the work as the main line;

[0113] S142: determining a first matching coefficient based on the matching between the plurality of sub-events, and determining a second matching coefficient based on the matching between the plurality of sub-events and the target animal;

[0114] S143: In the drone, an abnormal event of the target animal is determined based on the first matching coefficient, the second matching coefficient, and the abnormal mapping relationship.

[0115] In an embodiment of the present application, the drone collects the motion signals of the target animal, divides the work according to the motion signals of the target animal and the current event of the target animal, and outputs multiple sub-events with the work as the main line;

[0116] At this time, drones are usually equipped with various sensors, such as high-definition cameras, infrared thermal imagers, sound collectors, etc., to capture the motion signals of target animals; motion signals include the target animal's location information (GPS coordinates), movement trajectory, speed, acceleration, posture (such as standing, running, jumping, etc.), sound (such as calling, breathing, etc.) and other biological signals (such as heart rate, body temperature, etc., if they can be measured non-contact); in order to ensure the accuracy and continuity of the data, drones need to collect these motion signals at a certain frequency, and this frequency depends on the complexity of the animal's behavior and the drone's ability to process data.

[0117] Before or simultaneously with collecting the action signal, the drone has determined the current event of the target animal through other steps (such as S133), such as foraging, escaping from predators, resting, etc.; based on the current event and the action signal, the drone divides the target animal's behavior into different work stages or tasks, which should be able to reflect the specific behavior or activity process of the animal when performing the current event; the basis for the work division includes the animal's behavioral pattern, ecological habits, and behavioral responses in specific situations.

[0118] Each work phase or task is defined as a sub-event; sub-events should have clear time boundaries and identifiable behavioral characteristics; drones output these sub-events in the form of time series, including each sub-event's timestamp, behavioral description, and related action signal data. These sub-event data are used for further analysis and research, such as animal behavior pattern recognition and abnormal event detection.

[0119] Specifically, suppose a drone is monitoring a deer that is foraging; the drone captures the deer's movement trajectory, speed, and posture changes through a high-definition camera, and at the same time records the sound of the deer occasionally lowering its head to nibble on the grass through a sound collector.

[0120] The drone has determined that the deer's current activity is "foraging"; based on the collected motion signals, the drone divides the deer's foraging behavior into the following work stages: searching for food sources (the deer walks around on the grass at a slow speed and in a more relaxed posture), approaching the food source (the deer speeds up and moves towards a specific area), starting to eat (the deer lowers its head to nibble on the grass, and the sound collector records the nibbling sound), and resting while eating (the deer occasionally raises its head to observe the surrounding environment and slows down its speed).

[0121] The drone outputs these sub-events in the form of a time series, including the timestamp of each sub-event, behavior description (such as "searching for food source", "starting to eat", etc.) and related action signal data (such as location information, speed, sound, etc.). These data are used for subsequent research such as animal behavior pattern recognition and foraging efficiency analysis. Through this example, we can see the workflow and effect of step S141 in actual application. The drone collects the action signals of the target animal and divides the work according to these signals and the current event. Finally, it outputs multiple sub-events with the work as the main line, providing valuable data support for subsequent analysis and research.

[0122] Furthermore, the first matching coefficient is determined based on the matching between multiple sub-events, and the second matching coefficient is determined based on the matching between multiple sub-events and the target animal, which is compatible with the overall consideration of the matching between multiple sub-events and the target animal and ensures the accuracy of the second matching coefficient.

[0123] At this time, the first matching coefficient measures the logical consistency and coherence between multiple sub-events, which usually involves the order in which the sub-events occur, the time intervals between them, and the causal relationship between them; the drone has a set of rules or models about the behavior patterns of the target animals pre-set inside to determine the reasonable matching between sub-events. These rules are based on the results of animal behavior research and are also learned from historical data through machine learning; based on the actual matching between sub-events and the degree of compliance with the preset standards, the drone calculates a first matching coefficient between 0 and 1; the closer the coefficient is to 1, the higher the matching degree between the sub-events.

[0124] The second matching coefficient measures the overall degree of conformity of multiple sub-events with the typical behavioral characteristics of the target animal. This involves whether the behavioral patterns reflected by the sub-events are consistent with the known behavioral characteristics of the target animal. The drone maintains a database of the behavioral characteristics of target animals, including typical behavioral patterns in different situations, the temporal patterns of behavior, etc. The drone compares the sub-events with the information in the behavioral characteristic database and calculates a second matching coefficient based on the accuracy and completeness of the match. Similarly, the closer the coefficient is to 1, the higher the degree of match between the sub-event and the behavioral characteristics of the target animal.

[0125] Specifically, assume that the drone is monitoring a deer that is foraging and has output multiple sub-events according to step S141, including "searching for food source", "approaching food source", "starting to eat" and "resting while eating".

[0126] The first matching coefficient was determined based on the matches between multiple sub-events: the drone examined the order in which these sub-events occurred and found that they occurred in a logical sequence: first searching for a food source, then approaching it, then starting to eat, and finally resting while eating; the drone also noticed that the time intervals between each sub-event were reasonable, with no abrupt jumps or repetitions; based on these observations and the drone's pre-set rules for deer foraging behavior, the drone calculated a first matching coefficient of 0.95, indicating a high degree of match between the sub-events.

[0127] A second matching coefficient was determined based on the matching of multiple sub-events with the target animal. The drone compared the sub-events with the typical characteristics of deer foraging behavior and found that the behavioral patterns reflected by these sub-events were very consistent with the typical behavior of deer when foraging. For example, "searching for a food source" and "approaching a food source" are the preparatory behaviors of deer before foraging, "starting to eat" is the core behavior during foraging, and "resting while eating" is a common way for deer to rest during long foraging periods. Based on these comparison results and the database of deer behavioral characteristics maintained by the drone, the drone calculated a second matching coefficient of 0.98, indicating that the sub-events highly matched the behavioral characteristics of the target animal.

[0128] Therefore, in the drone, the abnormal events of the target animal are determined based on the first matching coefficient, the second matching coefficient and the abnormal mapping relationship, which is compatible with the overall consideration of the first matching coefficient, the second matching coefficient and the abnormal mapping relationship, and ensures the accuracy of the abnormal events of the target animal.

[0129] At this point, the drone will first integrate the first matching coefficient and the second matching coefficient calculated previously. These two coefficients reflect the logical consistency between sub-events and the degree of conformity between sub-events and the behavioral characteristics of the target animal respectively; the drone will take these two coefficients into consideration to form a comprehensive assessment of the current behavioral pattern of the target animal. This assessment involves weighted averaging, multiplication, or other forms of combination of the coefficients, depending on the design of the drone's internal algorithm.

[0130] An abnormal mapping relationship refers to a set of rules or models preset inside the drone that are used to identify which behavioral patterns or sub-event combinations indicate an abnormal state of the target animal. These rules are based on animal behavior research, statistical analysis of historical data, or expert experience. The abnormal mapping relationship defines which combinations of first matching coefficients and second matching coefficients indicate the occurrence of an abnormal event. For example, if both coefficients are below a certain threshold, or if their combination does not conform to a known normal behavioral pattern, the judgment of an abnormal event is triggered.

[0131] The drone determines abnormal events based on the integrated coefficients and abnormal mapping relationships; if the current behavior pattern matches the rules in the abnormal mapping relationship, the drone will determine that an abnormal event has occurred in the target animal; once the abnormal event is determined, the drone will output relevant abnormal information, including the type of abnormal event, time of occurrence, cause, etc., which will be used for subsequent monitoring, alarm or intervention actions.

[0132] Specifically, suppose the drone is monitoring a foraging deer and has calculated the first matching coefficient (0.95) and the second matching coefficient (0.98) based on the previous steps. The drone takes these two coefficients into consideration. Since both coefficients are close to 1, the logical consistency between the sub-events and the degree of conformity with the behavioral characteristics of the target animal are both high. In this example, the drone uses a weighted average or product method to integrate the two coefficients to obtain a comprehensive evaluation value. Since both coefficients are high, the comprehensive evaluation value will also be a high value.

[0133] The drone has a pre-set set of abnormal mapping relationships for deer foraging behavior. For example, if both the first and second matching coefficients are less than 0.8, or their product is less than 0.7, this indicates that the deer's behavior pattern is abnormal. In this example, because the comprehensive evaluation value is very high, far exceeding the threshold in the abnormal mapping relationship, the drone will not determine that the deer has experienced an abnormal event. However, to illustrate the process of abnormal event determination, assume a different scenario: suppose the drone monitors a change in the deer's behavior pattern in another time period, and the calculated first matching coefficient drops to 0.6 and the second matching coefficient drops to 0.75. In this case, the comprehensive evaluation value obtained by the drone after comprehensively considering these two coefficients will be lower than a certain threshold in the abnormal mapping relationship (for example, 0.8). According to the rules of the abnormal mapping relationship, the drone will determine that the deer has experienced an abnormal event and output relevant abnormal information, such as "The deer's behavior pattern is abnormal and has encountered danger or health problems."

[0134] In one embodiment of the present application, an abnormal event type matching table is collected, and the abnormal event type matching table is shown in Table 4:

[0135] Table 4 Abnormal event type matching table

[0136] The first matching coefficient The second matching coefficient Abnormal event type ≥0.9 ≥0.9 normal <0.9 and ≥0.7 ≥0.9 Slight anomaly (logical inconsistency) ≥0.9 <0.9 and ≥0.7 Slight abnormality (distinct features) <0.7 Any value Severe abnormalities Any value <0.7 Severe abnormalities

[0137] In this abnormal event type matching table, the drone compares the calculated first matching coefficient and second matching coefficient with the thresholds in the table to determine the abnormal event type of the target animal.

[0138] refer to Figure 6In step S15, an abnormal picture is determined based on the abnormal event of the target animal, the panoramic image of the target animal, and the surrounding image of the target animal, and an optimized picture is determined based on the abnormal event of the target animal and the relationship between the optimized mapping pictures to output corresponding optimization measures;

[0139] In the specific implementation process of the present invention, the specific steps are:

[0140] S151: The drone determines an abnormal area based on analysis of an abnormal event of the target animal, determines a first sub-image based on the abnormal area and a panoramic image of the target animal, determines a second sub-image based on the abnormal area and surrounding images of the target animal, and determines an abnormal image based on a synthesis of the first sub-image and the second sub-image;

[0141] S152: Determine an optimized image of the target animal based on the abnormal event of the target animal and the relationship between the optimized mapping images, and determine an area to be optimized based on the optimized image and the abnormal image;

[0142] S153: In the drone, corresponding optimization measures are determined according to the area to be optimized and the target animal.

[0143] In an embodiment of the present application, the drone determines the abnormal area based on the analysis of abnormal events of the target animal, and determines the first sub-screen based on the abnormal area and the panoramic image of the target animal, determines the second sub-screen based on the abnormal area and the surrounding image of the target animal, and determines the abnormal screen based on the synthesis of the first sub-screen and the second sub-screen, which is compatible with the overall consideration of the synthesis of the first sub-screen and the second sub-screen to ensure the accuracy of the abnormal screen.

[0144] At this time, the drone will first receive a report or data about an abnormal event involving the target animal. This abnormal event is detected through steps such as S143 and includes information such as the specific type, time of occurrence, and location of the abnormal event. Based on the report of the abnormal event, the drone uses built-in map data, image recognition algorithms, etc. to determine the specific location where the abnormal event occurred, and marks this area as an abnormal area in the panoramic image.

[0145] During the monitoring process, the drone will continuously capture panoramic images of the target animal and its surrounding environment. These images contain rich information for subsequent analysis and processing. Once the abnormal area is determined, the drone will crop a sub-image from the panoramic image that contains the abnormal area and the core part of the target animal. This sub-image is called the first sub-image; it is mainly used to display the core information of the abnormal event.

[0146] In addition to panoramic images, the drone also captures local images of the target animal's surroundings, which provide contextual information about the abnormal event. The drone cuts out a sub-image from the surrounding images that includes the abnormal area and its surrounding environment. This sub-image is called the second sub-image, which is mainly used to display the environmental information of the target animal when the abnormal event occurred.

[0147] The drone synthesizes the first and second sub-pictures, involving operations such as image stitching, transparency adjustment, and color correction, to generate an abnormal picture that contains both the core information of the abnormal event and the surrounding environment. The synthesized picture is the abnormal picture, which integrates the core information and contextual information of the abnormal event, providing an important basis for subsequent analysis and processing.

[0148] Specifically, suppose a drone detects an abnormal event in which a deer suddenly collapses. The drone receives an abnormal event report and learns that a deer collapsed at a specific location in the forest. The drone uses built-in map data and image recognition algorithms to mark the location where the deer collapsed as the abnormal area in the panoramic image.

[0149] The first sub-image is determined based on the panoramic image of the abnormal area and the target animal: the drone cuts out a sub-image containing the fallen deer and its surroundings from the panoramic image. This sub-image shows the specific location and posture of the fallen deer.

[0150] The second sub-image is determined based on the abnormal area and the surrounding images of the target animal: the drone crops a sub-image from the surrounding images that contains environmental information such as the fallen deer and its surrounding trees and vegetation. This sub-image provides contextual information about the deer when it fell.

[0151] The drone synthesizes the first and second sub-images, adjusts parameters such as image stitching and color correction, and generates an abnormal image that contains both specific information about the fallen deer and the surrounding environment. This abnormal image is used for subsequent analysis and processing, such as determining the deer's health and notifying relevant personnel to provide rescue services. This example shows the workflow and effectiveness of step S151 in actual application. The drone can determine the abnormal area based on the analysis of the abnormal event, crop the first and second sub-images based on the panoramic image and surrounding images, and finally synthesize an abnormal image that contains the core information and context of the abnormal event, providing important support for subsequent processing.

[0152] Furthermore, the optimized picture of the target animal is determined based on the abnormal events of the target animal and the relationship between the optimized mapping pictures, and the area to be optimized is determined based on the optimized picture and the abnormal picture, which is compatible with the overall consideration of the abnormal events of the target animal and the relationship between the optimized mapping pictures, and ensures the accuracy of the optimized picture of the target animal.

[0153] At this time, the drone needs to conduct an in-depth analysis of the abnormal events of the target animals, including the type, time of occurrence, location and cause of the abnormal events; the drone stores a set of optimized mapping picture relationships, which is a rule or model for mapping abnormal event types with optimized picture templates; different abnormal event types correspond to different optimized picture templates, which contain parameters such as the best observation angle, picture composition, color adjustment, etc. for the abnormal event type; based on the analysis results of the abnormal event and the optimized mapping picture relationship, the drone selects the most appropriate optimized picture template from the internal storage, and fine-tunes the template according to the actual situation (such as the position and posture of the target animal), and finally generates an optimized picture of the target animal. This optimized picture is designed to display the abnormal event and its impact in the best way, facilitating subsequent analysis and processing.

[0154] The drone compares and analyzes the optimized image with the abnormal image, including comparison of the image content, color, brightness, contrast, etc. Through comparative analysis, the drone identifies the difference areas between the optimized image and the abnormal image. These difference areas contain information that requires further attention or optimization, such as abnormal behavior of the target animal, changes in physical condition, abnormalities in the surrounding environment, etc. These difference areas are identified as areas to be optimized; the drone marks the areas to be optimized in the optimized image or abnormal image for subsequent processing and analysis.

[0155] Specifically, suppose a drone detects an abnormal event in which a deer suddenly falls to the ground. The drone analyzes the abnormal event and learns that the deer has fallen due to injury or illness. The drone selects an optimized image template for the "animal fall" abnormal event from the optimized mapping image relationships stored internally. This template includes a vertical shooting angle from above, a clear image composition, and appropriate color adjustments to best show the deer's fallen state. The drone fine-tunes the optimized image template based on the deer's actual position and posture to generate an optimized image of the target animal.

[0156] The drone compared and analyzed the optimized image with the previously generated abnormal image. Through comparison and analysis, the drone identified that the body parts of the deer in the optimized image were different from those in the abnormal image, especially the deer's legs, which seemed to have signs of injury. The drone marked the deer's legs and the surrounding area as areas to be optimized for further observation and analysis.

[0157] Therefore, in the drone, the corresponding optimization measures are determined according to the area to be optimized and the target animal, which is compatible with the overall consideration of the area to be optimized and the target animal, ensuring the accuracy of the corresponding optimization measures, realizing the drone's intelligent analysis of abnormal events, and further ensuring the accuracy of the optimization measures.

[0158] At this point, the drone first conducts a detailed assessment of the area to be optimized, including the size, shape, location, and relative relationship of the area to the target animal; at the same time, the drone conducts an in-depth analysis of the status of the target animal, including its health status, behavior patterns, threats faced, etc. Based on the above analysis, the drone determines the specific relationship between the area to be optimized and the target animal, such as whether the area to be optimized is directly related to the abnormal behavior or health status of the target animal, as well as the degree and importance of this relationship.

[0159] The drone stores a library of optimization measures designed for different types of relationships and abnormal events, aiming to improve the status or environment of the target animal. Based on the analysis results of the relationship between the area to be optimized and the target animal, the drone selects the most appropriate type of measure from the library. These measures include changing the flight trajectory to monitor the target animal more closely, adjusting image acquisition parameters to obtain clearer images, triggering alarms to notify relevant personnel, releasing rescue supplies, etc.

[0160] The drone will develop a specific optimization measures plan based on the selected measure type and the actual situation (such as the location of the target animal, weather conditions, surrounding environment, etc.); once the measures plan is completed, the drone will immediately execute these measures, which involves adjusting the flight path, changing image acquisition settings, sending alarm information, releasing rescue supplies, etc.; while executing the optimization measures, the drone will also continuously monitor the status of the target animals and changes in the area to be optimized to evaluate the effectiveness of the measures and adjust the measures plan as needed.

[0161] Specifically, suppose the drone detects that a deer's leg appears to be injured, and the area is marked as an area to be optimized; the drone evaluates the area to be optimized (the deer's leg and its surroundings) and finds obvious swelling and blood in the area; at the same time, the drone analyzes the deer's condition and finds that its movements are slow, seemingly affected by the leg injury; based on the above analysis, the drone determines that there is a direct correlation between the area to be optimized and the target animal (the injured deer), and this correlation poses a threat to the deer's health.

[0162] The drone selected "trigger an alarm and notify relevant personnel" and "continuously monitor the status of the target animal" as optimization measure types from the optimization measure library; the reason for selecting these measures is that they can quickly attract the attention of relevant personnel and provide timely assistance or rescue to the injured deer.

[0163] The drone developed a specific action plan, including sending an alert message to nearby wildlife protection agencies and providing the deer's exact location and injury condition; at the same time, the drone adjusted its flight path to more closely monitor the deer's condition and ensure its safety before rescuers arrived; during the implementation of the measures, the drone continued to send updated information to the wildlife protection agency, including the deer's current condition, changes in the surrounding environment, etc.

[0164] In one embodiment of the present application, an optimization measure matching table is collected, and the optimization measure matching table is shown in Table 5: Table 5 Optimization measure matching table

[0165]

[0166] Suppose a drone discovers a deer with a leg injury and slow movements during monitoring. According to the optimization measure matching table, the drone will take the following optimization measures: trigger an alarm to notify nearby wildlife rescuers, provide the deer's exact location and injury status; and continuously monitor the deer's condition to ensure its safety before rescuers arrive.

[0167] See also Figure 7 , Figure 7 : is a schematic diagram of the structural composition of a system for identifying a drone in flight state in an embodiment of the present invention; the system for identifying a drone in flight state includes:

[0168] The acquisition module 21 is configured to enable the drone to acquire multiple images of the target park when the drone is in flight;

[0169] An image module 22 is configured to determine a panoramic image of a target animal and an image of the surrounding area of ​​the target animal based on the multiple images and a distribution map of the target park;

[0170] a feature module 23, configured to determine a current event of the target animal based on the panoramic image of the target animal and the surrounding image of the target animal;

[0171] an abnormal event module 24 for determining a plurality of sub-events according to the division of the current event of the target animal, and determining an abnormal event of the target animal according to the plurality of sub-events;

[0172] The optimization module 25 is used to determine abnormal images based on abnormal events of the target animal, the panoramic image of the target animal and the surrounding image of the target animal, and to determine optimized images based on the abnormal events of the target animal and the relationship between the optimized mapping images to output corresponding optimization measures.

[0173] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for identifying a drone in flight state, characterized in that: include: While the UAV is in flight, the UAV collects multiple images of the target park; determining a panoramic image of a target animal and a surrounding image of the target animal based on the plurality of images and a distribution map of the target park; determining a current event of the target animal based on a panoramic image of the target animal and an image of the surrounding area of ​​the target animal; Determining a plurality of sub-events according to the division of the current event of the target animal, and determining an abnormal event of the target animal according to the plurality of sub-events; An abnormal picture is determined based on the abnormal event of the target animal, the panoramic image of the target animal and the surrounding image of the target animal, and an optimized picture is determined based on the abnormal event of the target animal and the optimized mapping picture relationship to output corresponding optimization measures.

2. The method for identifying a drone in flight state according to claim 1, characterized in that: When the drone is in flight, the drone collects multiple images of the target park, including: When the UAV is above the target park, the inspection route of the UAV relative to the target park is determined according to the shape of the target park and the current position of the UAV, and the UAV flies along the inspection route; During the flight of the drone, the follow-up shooting mode of multiple cameras is determined according to the relative position of the drone and the target park and the orientation of multiple cameras. Based on the follow-up shooting mode, multiple cameras are triggered to shoot the target park in different directions to collect multiple images of the target park.

3. The method for identifying a UAV in flight state according to claim 1, characterized in that: The determining of a panoramic image of the target animal and a surrounding image of the target animal based on the plurality of images and the distribution map of the target park includes: The drone determines the distribution map of the target park based on the name of the target park and a regional database, matches the distribution map of the target park with multiple images to adjust the alignment of the multiple images, and determines the overall image based on the synthesis of the multiple images and marks the target animals. Capture the target animal and determine a panoramic image of the target animal based on the morphology and overall image of the target animal. In the panoramic image of the target animal, multiple sides of the target animal are presented along different directions, and the outline of each side can be clearly presented; The drone determines the target animal's impact range based on the target animal's species and its movement route, and determines the target animal's surrounding image based on the target animal's impact range, the target animal's location, and the overall image.

4. The method for identifying a UAV in flight state according to claim 1, wherein: The determining of the current event of the target animal based on the panoramic image of the target animal and the surrounding image of the target animal includes: When the drone is in flight, the drone determines a plurality of side images based on the division of the panoramic image of the target animal, determines a plurality of first sub-features based on the plurality of side images, the morphology of the target animal, and the species of the target animal, and determines an action event of the target animal based on a combination of the plurality of first sub-features; The drone determines multiple orientation images based on the surrounding images of the target animal and the relative orientation of the target animal, determines multiple second sub-features based on the recognition of the multiple orientation images, and determines multiple surrounding events of the target animal based on the multiple second sub-features and the corresponding orientations.

5. The method for identifying a UAV in flight state according to claim 4, characterized in that: The method further includes determining a current event of a target animal based on a panoramic image of the target animal and images of the surrounding areas of the target animal. A plurality of event combinations are determined according to the action event of the target animal and a plurality of surrounding events of the target animal, and the drone determines the current event of the target animal according to the plurality of event combinations and time.

6. The method for identifying a UAV in flight state according to claim 1, characterized in that: The step of determining a plurality of sub-events according to the division of the current event of the target animal, and determining an abnormal event of the target animal according to the plurality of sub-events, includes: The drone collects the motion signals of the target animal, divides the work according to the motion signals of the target animal and the current events of the target animal, and outputs multiple sub-events with the work as the main line.

7. The method for identifying a UAV in flight state according to claim 6, characterized in that: The method of determining a plurality of sub-events according to the division of the current event of the target animal, and determining the abnormal event of the target animal according to the plurality of sub-events, further includes: Determining a first matching coefficient based on the matching between the plurality of sub-events, and determining a second matching coefficient based on the matching between the plurality of sub-events and the target animal; In the drone, abnormal events of the target animal are determined based on the first matching coefficient, the second matching coefficient, and the abnormal mapping relationship.

8. The method for identifying a UAV in flight state according to claim 1, characterized in that: The abnormal image is determined based on the abnormal event of the target animal, the panoramic image of the target animal, and the surrounding image of the target animal, and the optimized image is determined based on the abnormal event of the target animal and the relationship between the optimized mapping images to output corresponding optimization measures, including: The drone determines the abnormal area based on the analysis of abnormal events of the target animal, and determines the first sub-screen based on the abnormal area and the panoramic image of the target animal, determines the second sub-screen based on the abnormal area and the surrounding image of the target animal, and determines the abnormal screen based on the synthesis of the first sub-screen and the second sub-screen.

9. The method for identifying a UAV in flight state according to claim 8, characterized in that: The method further includes determining an abnormal image based on the abnormal event of the target animal, the panoramic image of the target animal, and the surrounding image of the target animal, and determining an optimized image based on the abnormal event of the target animal and the relationship between the optimized mapping images to output corresponding optimization measures. Determine the optimized image of the target animal based on the abnormal events of the target animal and the relationship between the optimized mapping images, and determine the area to be optimized based on the optimized image and the abnormal image; In the drone, corresponding optimization measures are determined based on the area to be optimized and the target animals.

10. A system for identifying a drone in flight, characterized in that: The system for identifying a drone in flight state is applied to the method for identifying a drone in flight state as claimed in any one of claims 1 to 9, and the system for identifying a drone in flight state includes: an acquisition module, configured to enable the drone to acquire multiple images of a target park when the drone is in a flight state; An image module, configured to determine a panoramic image of a target animal and an image of the surrounding area of ​​the target animal based on the multiple images and a distribution map of the target park; a feature module, configured to determine a current event of a target animal based on a panoramic image of the target animal and images of the surrounding areas of the target animal; An abnormal event module, configured to determine a plurality of sub-events according to the division of the current event of the target animal, and determine the abnormal event of the target animal according to the plurality of sub-events; The optimization module is used to determine abnormal images based on abnormal events of the target animal, panoramic images of the target animal and surrounding images of the target animal, and to determine optimized images based on the abnormal events of the target animal and the relationship between the optimized mapping images to output corresponding optimization measures.

Images