An unmanned aerial vehicle image processing method and system based on artificial intelligence

Abstract

The application relates to the field of artificial intelligence, in particular to a UAV image processing method and system based on artificial intelligence, which comprises an image acquisition module, a preprocessing module, an image recognition module, a flight control cooperation module and a user interaction module, the image acquisition module is used for acquiring target images in the flight process of a UAV, the preprocessing module is used for preprocessing the target images and obtaining processed images, the image recognition module is used for recognizing the processed images and acquiring resampling indexes according to the processed images, the flight control cooperation module is used for adjusting the flight state of the UAV, and the user interaction module is used for displaying the acquired processed images and the recognition results of the image recognition module. The application realizes quantitative judgment on the effectiveness of images by setting a resampling index formula. When the image quality is poor or a high-risk area is detected, the system can automatically judge whether it is necessary to take pictures again, thereby avoiding information omission caused by blind shooting of a traditional UAV.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, and in particular to a method and system for processing images of a UAV based on artificial intelligence. BACKGROUND

[0002] With the rapid development of UAV technology, UAVs are widely used in military reconnaissance, agricultural monitoring, power inspection, traffic supervision, disaster rescue and other fields. In particular, in terms of image acquisition and analysis, UAVs have gradually become an important means of aerial data collection due to their high mobility and low cost.

[0003] A method and a UAV for processing images are disclosed in the prior art of WO2020107487A1. The image processing method is applied to the UAV, and the UAV is provided with imaging devices in at least two directions. The image processing method includes: obtaining a to-be-processed image in each direction of the at least two directions; determining a first direction in the at least two directions according to the to-be-processed image in each direction of the at least two directions, and obtaining a reference value of the first direction; the reference value of the first direction is used to determine whether to update a key reference frame corresponding to each direction of the at least two directions; if the reference value of the first direction meets a preset condition, the key reference frame corresponding to each direction of the at least two directions is updated.

[0004] Another typical method for collecting images of a UAV is disclosed in the prior art of WO2019061063A1, which includes: receiving a take-off instruction; obtaining a target shooting object and saving a feature of the target shooting object; tracking the target shooting object according to the feature of the target shooting object to obtain a current image; wherein the current image includes the target shooting object; analyzing the position of the target shooting object in the current image, and if the position of the target shooting object in the current image meets an image collection condition, collecting an image.

[0005] Another image processing method is disclosed in the prior art of WO2020124355A1, which includes: performing n-1 times of downsampling on an original image to determine n images with different resolutions; performing edge-preserving filtering on each image to obtain high-frequency information and a filtering result; fusing the high-frequency information of the i-th image and the high-frequency information of the i-1-th image to obtain an i-1 high-frequency fusion result; and fusing the filtering result of the i-th image and the filtering result of the i-1-th image according to the i-1 high-frequency fusion result to obtain an i-1 filtering fusion result.

[0006] At present, the existing unmanned aerial vehicle usually only shoots images according to a set route in the moving process, and does not adjust the shooting mode according to the image processing result, which is easy to cause omission or large error of the shooting content. SUMMARY

[0007] The present application aims at the existing problems, and provides an unmanned aerial vehicle image processing method and system based on artificial intelligence.

[0008] In order to overcome the shortcomings of the prior art, the present application adopts the following technical solutions:

[0009] An unmanned aerial vehicle image processing system based on artificial intelligence comprises an image acquisition module, a preprocessing module, an image recognition module, a flight control coordination module and a user interaction module, the image acquisition module is used for acquiring target images in the flight process of the unmanned aerial vehicle, the preprocessing module is used for preprocessing the target images and obtaining processed images, the image recognition module is used for recognizing the processed images and obtaining resampling indexes according to the processed images, the flight control coordination module is used for adjusting the flight state of the unmanned aerial vehicle according to the resampling indexes, and the user interaction module is used for displaying the obtained processed images and the recognition results of the image recognition module.

[0010] Further, the image acquisition module comprises a visible light camera, a gimbal stabilizing device and a shooting parameter adjustment unit, the visible light camera is used for shooting visible light images of the target, the gimbal stabilizing device is used for stabilizing the visible light camera, and the shooting parameter adjustment unit is used for automatically focusing the visible light camera.

[0011] Further, the preprocessing module comprises an image enhancement unit, an image conversion unit and a synchronization information unit, the image enhancement unit is used for denoising and enhancing the contrast of the target images, the image conversion unit is used for converting the format of the enhanced images into a format conforming to the input of the image recognition module, and the synchronization information unit is used for marking the images converted in the format according to the current positioning and the current time.

[0012] Further, the image recognition module comprises an image quality detection unit, a target detection unit, a storage unit and a calculation unit, the image quality detection unit is used for detecting various parameters of the processed images, the target detection unit is used for image segmentation of the processed images and recognition of the target detection objects in the processed images, the storage unit is used for storing preset parameters required for calculation of the calculation unit, and the calculation unit is used for calculating the resampling indexes of the unmanned aerial vehicle at the current shooting location according to the detection results of the target detection unit.

[0013] Further, the flight control cooperation module comprises a judgment unit and a control command generation unit, the judgment unit is used for judging whether the unmanned aerial vehicle needs to be adjusted according to the resampling index, and the control command generation unit is used for generating a corresponding control command and sending to the unmanned aerial vehicle when the judgment unit judges that adjustment is needed.

[0014] An unmanned aerial vehicle image processing method based on artificial intelligence, applied to an unmanned aerial vehicle image processing system based on artificial intelligence, comprising the following steps:

[0015] S1, the image acquisition module acquires target images in the flight process of the unmanned aerial vehicle;

[0016] S2, the preprocessing module pre-processes the target images and obtains processed images;

[0017] S3, the image recognition module identifies the processed images and obtains the resampling index of the processed images;

[0018] S4, the user interaction module displays the obtained target images and the recognition results of the image recognition module;

[0019] S5, the flight control cooperation module judges whether the flight state of the unmanned aerial vehicle needs to be adjusted according to the resampling index, if yes, generates a control command and sends to the unmanned aerial vehicle, otherwise, ends.

[0020] Further, the image recognition module identifies the processed images and obtains the resampling index of the processed images, comprising the following steps:

[0021] S31, the image quality detection unit detects various parameters of the processed images;

[0022] S32, the target detection unit performs image segmentation on the processed images and identifies the target detection objects in the segmented images;

[0023] S33, the storage unit outputs the preset parameters required for calculation to the calculation unit;

[0024] S34, the calculation unit calculates the resampling index according to the detection results of the target detection unit and the obtained various parameters.

[0025] The beneficial effects obtained by the present application are: 1. By setting the resampling index formula, the effectiveness of the image is quantitatively judged. When the image quality is poor or a high-risk area is detected, the system can automatically judge whether it needs to be re-shot, avoiding the information omission caused by the traditional unmanned aerial vehicle blind shooting.

[0026] 2. The system judges whether to adjust the flight state according to the resampling index, and implements the same area re-shooting of the unmanned aerial vehicle through the around shooting and the like, improves the integrity and pertinence of the image acquisition, and reduces the influence caused by the omission and error of the shooting. BRIEF DESCRIPTION OF DRAWINGS

[0027] The present application can be further understood from the following description in conjunction with the drawings. The components in the drawings are not necessarily drawn to scale, but emphasis is instead placed upon illustrating the principles of the embodiments. Like reference numerals designate like parts throughout the different views.

[0028] Figure 1 It is a structural schematic diagram of the present application.

[0029] Figure 2 It is a workflow diagram of the present application.

[0030] Figure 3 It is a flowchart of the image recognition module of the present application for recognizing and obtaining the resampling index of the processed image.

[0031] Figure 4 It is a relationship diagram of the area risk parameter and the actual number of the target detection object and the ideal number of the target detection object. DETAILED DESCRIPTION

[0032] The following is to illustrate the embodiments of the present application through specific specific embodiments, and the person skilled in the art can understand the advantages and effects of the present application from the disclosed content. The present application can be implemented or applied through other different specific embodiments, and each detail in the specification can be modified and changed based on different viewpoints and applications without departing from the spirit of the present application. In addition, the drawings of the present application are only simple schematic illustrations, not the depiction according to the actual size, and the prior declaration. The following embodiments will further illustrate the related technical content of the present application, but the disclosed content is not used to limit the protection scope of the present application.

[0033] Example one: according to Figure 1 , Figure 2 , Figure 3 and Figure 4The embodiment provides an unmanned aerial vehicle image processing system based on artificial intelligence, which comprises an image acquisition module, a preprocessing module, an image recognition module, a flight control coordination module and a user interaction module, the image acquisition module is used for acquiring target images in the flight process of an unmanned aerial vehicle, the preprocessing module is used for preprocessing the target images and obtaining processed images, the image recognition module is used for recognizing the processed images and obtaining resampling indexes according to the processed images, the flight control coordination module is used for adjusting the flight state of the unmanned aerial vehicle according to the resampling indexes, and the user interaction module is used for displaying the obtained processed images and the recognition results of the image recognition module.

[0034] Preferably, the processed images comprise positioning and time information of the images, and the image recognition module transmits the positioning and time information corresponding to the resampling indexes to the flight control coordination module when the resampling indexes are transmitted to the flight control coordination module, so that the flight control coordination module has data reference when the flight state of the unmanned aerial vehicle is adjusted according to the resampling indexes, for example, the flight control coordination module can be enabled to judge whether the work task in the range of the area corresponding to the resampling indexes has been completed, so as to ensure that the flight adjustment has accurate geographical reference basis; or the flight control coordination module can be enabled to judge whether the work task has been completed within the ideal working time.

[0035] Specifically, the user interaction module comprises a display unit and a user instruction receiving unit, the display unit is used for displaying the target images and the recognition results of the image recognition module, and the user instruction receiving unit is used for receiving user instructions and sending the user instructions to the unmanned aerial vehicle to manually control the unmanned aerial vehicle.

[0036] Furthermore, the image acquisition module comprises a visible light camera, a gimbal stabilizing device and a shooting parameter adjusting unit, the visible light camera is used for shooting visible light images of a target, the gimbal stabilizing device is used for stabilizing the visible light camera, and the shooting parameter adjusting unit is used for automatically focusing the visible light camera.

[0037] Furthermore, the preprocessing module comprises an image enhancement unit, an image conversion unit and a synchronization information unit, the image enhancement unit is used for denoising and enhancing the contrast of the target images, the image conversion unit is used for converting the format of the enhanced images into a format conforming to the input of the image recognition module, and the synchronization information unit is used for marking the images converted in the format according to the current positioning and the current time.

[0038] Furthermore, the image recognition module includes an image quality detection unit, a target detection unit, a storage unit, and a computing unit. The image quality detection unit is used to detect various parameters of the processed image. The target detection unit is used to segment the processed image and identify target objects in the processed image. The storage unit is used to store preset parameters required for the computing unit to perform calculations. The computing unit is used to calculate the resampling index of the UAV at the current shooting location based on the detection results of the target detection unit.

[0039] Specifically, the target detection object is set by those skilled in the art according to the task to be performed by the drone. For example, if the drone is to detect pests, the target detection object is pests; if the drone is to detect pedestrian traffic, the target detection object is humans. The target detection unit detects the target detection object in the image through an artificial intelligence recognition model (such as YOLO series or Faster R-CNN) and outputs corresponding parameters (average pest density per leaf or human density per road, etc., the specific parameter type is set by those skilled in the art according to the task). The artificial intelligence recognition model is trained by those skilled in the art based on existing technology and existing datasets.

[0040] Furthermore, the flight control coordination module includes a judgment unit and a control command generation unit. The judgment unit is used to determine whether the UAV needs to be adjusted based on the resampling index. The control command generation unit is used to generate a corresponding control command and send it to the UAV when the judgment unit determines that adjustment is needed.

[0041] Specifically, the judgment unit determines whether adjustment is needed by judging whether the resampling index is greater than the resampling index threshold. When the resampling index is greater than the resampling index threshold, the UAV is adjusted. The resampling index threshold is preferably between 0 and 2*ln(3). For example, when the task type is precision detection (such as pest identification), the requirements for image quality and target quantity accuracy are high, and the threshold can be set to 0.4. When the task type is general detection (such as crowd detection), the tolerance for error is larger, and the threshold can be set to 0.6. The specific value is set by those skilled in the art based on experience and the required accuracy. The drone is adjusted by maintaining the pitch angle (the preferred initial pitch angle selection rules are: 90° (vertically downward) when detecting pests, 75°-90° when detecting large crowds, and 45°-60° (biased towards vertical downward) when detecting local density, with the initial pitch angle being smaller as the flight altitude decreases) and shooting distance. The drone then circles along the vertical line between the center of the shooting location and the ground to retake multiple target images and automatically uses obstacle avoidance algorithms to avoid obstacles during the circle.

[0042] The application discloses an unmanned aerial vehicle image processing method based on artificial intelligence, which is applied to an unmanned aerial vehicle image processing system based on artificial intelligence and comprises the following steps.

[0043] S1, an image acquisition module acquires target images in the flight process of an unmanned aerial vehicle;

[0044] S2, a preprocessing module pre-processes the target images and obtains processed images;

[0045] S3, an image recognition module recognizes the processed images and obtains resampling indexes of the processed images;

[0046] S4, a user interaction module displays the obtained target images and the recognition results of the image recognition module;

[0047] S5, a flight control coordination module judges whether the flight state of the unmanned aerial vehicle needs to be adjusted according to the resampling indexes, generates a control command and sends the control command to the unmanned aerial vehicle if yes, and ends otherwise.

[0048] Further, the image recognition module recognizes the processed images and obtains resampling indexes of the processed images, and the method comprises the following steps.

[0049] S31, an image quality detection unit detects various parameters of the processed images;

[0050] S32, a target detection unit performs image segmentation on the processed images and recognizes target detection objects in the segmented images;

[0051] S33, a storage unit outputs preset parameters required for calculation to a calculation unit;

[0052] Specifically, the preset parameters include various threshold values and ideal values set by technicians in the field.

[0053] S34, the calculation unit calculates resampling indexes according to the detection results of the target detection unit and the obtained various parameters. Specifically, the resampling indexes can be calculated according to the following formula:

[0054] CCY=k1*TX+k2*FX

[0055]

[0056]

[0057] Wherein, CCY is a resampling index, used to measure the necessity of additional shooting target image, the larger the index, the greater the necessity, k1 is an image quality weight, TX is an image quality parameter, k2 is a regional risk weight, FX is a regional risk parameter; the value range of k1 and k2 is between 0 and 1, and satisfies k1+k2=1. The specific value is set by the person skilled in the art according to the purpose of the image to be shot by the unmanned aerial vehicle, such as the purpose of judging that too many target detection objects such as pests will cause adverse effects, then the value of k2 can be appropriately increased, one of the optional values is k2 equal to 0.7 and k1 equal to 0.3, such as the purpose of monitoring the flow of people and other target detection objects has less adverse effects, then the value of k1 can be appropriately increased, one of the optional values is k1 equal to 0.6 and k2 equal to 0.4;

[0058] e is a natural constant, L is the standard deviation of the Laplacian gradient of the processed image, which is obtained by the image quality detection unit through OpenCV, l is the lower threshold of the gradient difference, which is set by the person skilled in the art according to the required image accuracy, the higher the required accuracy, the larger the threshold. M is the number of pixels per row of the processed image, N is the number of pixels per column of the processed image, XS a is the gray value of the a-th pixel, xs a is the pixel value of the a-th pixel after convolution smoothing.

[0059] K is an amplitude parameter, which is set by the person skilled in the art according to the conventional number of target detection objects of the image to be shot by the unmanned aerial vehicle between 0.5 and 1.5, the larger the number, the smaller the amplitude parameter, the purpose of designing the amplitude parameter is to control the influence of the difference between AU and SET on the final result through the slope, the larger the number, the larger the numerical value of the difference, and the influence of large numerical value on the final result is reduced by reducing K, and the error is reduced. Preferably, the amplitude parameter is reduced by 0.05 for each additional target detection object, and when there are more than 30 target detection objects, the amplitude parameter is constant at the minimum value of 0.5. AU is the actual number of target detection objects, and SET is the ideal number of target detection objects, which is set by the person skilled in the art according to the actual situation, For the tolerance threshold, which is set by the skilled person in the art according to the allowable error range (ideal value and actual value of the target detection object). If K is too small (<0.5), no matter how much the difference between AU and SET is, the change of FX is very slow; in high-risk tasks where the pest density is too high, this sluggish response is inappropriate; when K is very large (>1.5), even if the difference between AU and SET is small, FX is already close to 1; the system may frequently misjudge as high risk, leading to excessive resampling or false triggering of the bypass camera; it will seriously affect the system efficiency and the utilization rate of UAV resources. The range of 0.5~1.5 is the best compromise in practical tasks. In common applications, the typical difference between AU and SET is in the range of 0~20; setting K=0.5~1.5 in this range can ensure that the response is not sluggish: not too aggressive.

[0060] The following is the program needed to calculate the resampling index:

[0061] import torch

[0062] import cv2

[0063] import numpy as np

[0064] from PILimportImage

[0065] defload_image(image_path:str)->np.ndarray:

[0066] """Load image as OpenCV format grayscale image"""

[0067] image=cv2.imread(image_path)

[0068] gray_image=cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)

[0069] return gray_image

[0070] def detect_targets_yolov5(image_path:str,target_class_names=['person'])->int:

[0071] """

[0072] Use YOLOv5 model to detect targets in image, count the number of target detection objects

[0073] target_class_names: Specifies the target class, such as ['person'], ['insect'], etc.

[0074] """

[0075] #Load the pre-trained model, which can be replaced with custom weights.

[0076] model=torch.hub.load('ultralytics / yolov5','yolov5s',pretrained=True)

[0077] model.classes = None # Check all classes

[0078] results = model(image_path)

[0079] detected=results.pandas().xyxy[0]

[0080] #Filter by target category

[0081] count = 0

[0082] for_,row in detected.iterrows():

[0083] if row['name']in target_class_names:

[0084] count+=1

[0085] return countdef calculate_CCY_updated(image:np.ndarray,AU:float,SET:float,

[0086] k1:float=0.7,k2:float=0.3,K:float=1.0,

[0087] L:float=1.0,delta:float=5.0,e:float=0.01)->float:

[0088] """

[0089] Calculate the image resampling index CCY.

[0090] image: Input grayscale image

[0091] AU: Actual quantity of the detected target

[0092] SET: target detection physical quantity

[0093] The rest of the parameters refer to the patent document

[0094] """

[0095] M, N = image.shape

[0096] total_pixels = M * N

[0097] laplacian = cv2.Laplacian(image, cv2.CV_64F)

[0098] XS = np.abs(laplacian)

[0099] XS_flat = XS.flatten()

[0100] XS_avg = np.mean(XS_flat)

[0101] # TX part

[0102] diff_sum = np.sum(np.abs(XS_flat - XS_avg))

[0103] norm_diff = diff_sum / (total_pixels * 255)

[0104] TX_part = 1 / (1 + np.exp(1 / L))

[0105] TX_inner = TX_part + norm_diff + e

[0106] TX = np.log(TX_inner) - 1 if TX_inner > 0 else -1.0

[0107] # FX part

[0108] FX_exp = -K * abs(AU - SET) / delta

[0109] FX = 1 / (1 + np.exp(FX_exp))

[0110] # Comprehensive index

[0111] CCY = k1 * TX + k2 * FX

[0112] return CCY

[0113] # Main flow call example

[0114] if __name__ == '__main__':

[0115] image_path = 'drone_image.jpg' # replace with actual drone image path

[0116] ideal_count = 10 # ideal detection number (e.g. 10 pest targets per frame)

[0117] # Step 1: Image loading

[0118] image_gray = load_image(image_path)

[0119] # Step 2: Target detection

[0120] actual_count = detect_targets_yolov5(image_path, target_class_names=['person']) # target classes can be customized

[0121] # Step 3: Indicator calculation

[0122] ccy_value = calculate_CCY_updated(image_gray, AU=actual_count, SET=ideal_count)

[0123] print(f'Detected target count: {actual_count}')

[0124] print(f'Calculated resampling indicator CCY: {ccy_value:.4f}')

[0125] Specifically, as shown in the following figure, when K is 1 and N is 5, the relationship between the regional risk parameter FX and the actual number of target detections AU and the ideal number of target detections SET is shown. Figure 4 Figure 4

[0126] The beneficial effects of the present scheme are: 1. By setting the resampling indicator formula, the effectiveness of the image can be quantitatively judged. When the image quality is poor or a high-risk area is detected, the system can automatically determine whether it needs to be re-shot, avoiding the information omission caused by traditional blind shooting of unmanned aerial vehicles.

[0127] ​​​2. The system judges whether to adjust the flight state according to the resampling index, and implements the same area re-shooting of the unmanned aerial vehicle through the around shooting and the like, thereby improving the integrity and pertinence of the image collection, and reducing the influence caused by the omission and errors in the shooting.

[0128] Embodiment two: this embodiment should be understood as containing all the features of any one of the preceding embodiments, and further improving on the basis thereof, and further comprising a method for adjusting the flight state of the unmanned aerial vehicle according to the resampling index. When the current shooting distance is greater than the recommended shooting distance (the recommended shooting distance is set by the person skilled in the art according to the camera performance), it is considered that shortening the shooting distance can effectively improve the accuracy of the shooting image, and the shooting distance is adjusted by using the method. When the absolute value of the difference between the actual number and the ideal number of the target detection object is still greater than the tolerance threshold (tolerance threshold) after the adjustment of the resampling index and the distance, it is considered that there may be a shooting number omission caused by the shooting pitch angle, and the pitch angle is adjusted by using the method, otherwise the adjustment according to the resampling index and the distance is continued.

[0129] The shooting distance is adjusted according to the following formula:

[0130]

[0131] Wherein, D is the adjusted shooting distance, D0 is the shooting distance before adjustment, e is a natural constant, D set is the recommended shooting distance, CCY is the resampling index, and CCY set is the resampling index threshold.

[0132] The shooting pitch angle is adjusted according to the following formula:

[0133]

[0134] Wherein, C1 is the first adjusted pitch angle, C2 is the second adjusted pitch angle, C0 is the current pitch angle, AU is the actual number of the target detection object, SET is the ideal number of the target detection object, is the tolerance threshold, CCY is the resampling index, and CCY set is the resampling index threshold, ver is the average function, and 90 and 45 in the formula respectively correspond to 90 degree pitch angle and 45 degree pitch angle.

[0135] ​After the adjustment, the UAV performs two rounds of shooting at the adjusted shooting distance and at the first and second pitch angles respectively. If the current pitch angle of the UAV is in the range of [0°, 45°), it is considered that the pitch angle is high, and the first pitch angle is used for shooting first (lower first and then raise). If it is in the range of [45°, 90°], it is considered that the pitch angle is low, and the second pitch angle is used for shooting first (raise first and then lower). In each round, the UAV re-shoots multiple target images along the vertical line of the center of the shooting location and the ground.

[0136] The beneficial effects of the embodiment are as follows: when the resampling index is high, the system automatically reduces the shooting distance, increases the target area in the imaging, and increases the shooting images from different pitch angles, so as to obtain more accurate actual number of target detection objects, improve the image clarity and the perception resolution of the recognition model, and ensure the accuracy of the subsequent recognition task.

[0137] The above disclosed content is only the preferred feasible embodiment of the present application, and does not limit the protection scope of the present application, so any equivalent technical changes made according to the content of the specification and drawings of the present application are included in the protection scope of the present application. In addition, as technology develops, the elements can be updated. The above units are only examples, and those skilled in the art can design different units according to actual needs when implementing the scheme.

Claims

1. An artificial intelligence-based unmanned aerial vehicle (UAV) image processing system, characterized in that, The system includes an image acquisition module, a preprocessing module, an image recognition module, a flight control coordination module, and a user interaction module. The image acquisition module is used to acquire target images during the flight of the UAV. The preprocessing module is used to preprocess the target images to obtain processed images. The image recognition module is used to recognize the processed images and obtain resampling indicators based on the processed images. The flight control coordination module is used to adjust the flight status of the UAV based on the resampling indicators. The user interaction module is used to display the acquired processed images and the recognition results of the image recognition module. The resampling index is calculated using the following formula: ； ； ； This is a resampling metric used to measure the necessity of taking additional images of the target; the higher the metric, the greater the necessity. Image quality weights, For image quality parameters, For regional risk weights, For regional risk parameters; and The values ​​of are all between 0 and 1, and satisfy . + =1; e is the natural constant. To handle the standard deviation of the image's Laplacian gradient, this standard deviation is obtained by the image quality detection unit using OpenCV. is the lower bound threshold for the gradient difference; M is the number of pixels in each row of the processed image, and N is the number of pixels in each column of the processed image. Let be the grayscale value of the a-th pixel. Let K be the convolutionally smoothed pixel value corresponding to the a-th pixel; K is the amplitude parameter. The actual quantity of the target object to be detected. The ideal number of the target detection object, This is the tolerance threshold.

2. The artificial intelligence-based UAV image processing system according to claim 1, characterized in that, The image acquisition module includes a visible light camera, a gimbal stabilization device, and a shooting parameter adjustment unit. The visible light camera is used to capture visible light images of the target, the gimbal stabilization device is used to stabilize the visible light camera, and the shooting parameter adjustment unit is used to automatically focus the visible light camera.

3. The artificial intelligence-based UAV image processing system according to claim 1, characterized in that, The preprocessing module includes an image enhancement unit, an image conversion unit, and a synchronization information unit. The image enhancement unit is used to denoise and enhance the contrast of the target image. The image conversion unit is used to convert the format of the enhanced image into a format that conforms to the input format of the image recognition module. The synchronization information unit is used to mark the format-converted image according to the current location and the current time.

4. The artificial intelligence-based UAV image processing system according to claim 1, characterized in that, The image recognition module includes an image quality detection unit, a target detection unit, a storage unit, and a computing unit. The image quality detection unit is used to detect various parameters of the processed image. The target detection unit is used to segment the processed image and identify target objects in the processed image. The storage unit is used to store preset parameters required for the computing unit to perform calculations. The computing unit is used to calculate the resampling index of the UAV at the current shooting location based on the detection results of the target detection unit.

5. The artificial intelligence-based UAV image processing system according to claim 1, characterized in that, The flight control coordination module includes a judgment unit and a control command generation unit. The judgment unit is used to determine whether the UAV needs to be adjusted based on the resampling index. The control command generation unit is used to generate a corresponding control command and send it to the UAV when the judgment unit determines that adjustment is needed.

6. An artificial intelligence-based drone image processing method, applied to an artificial intelligence-based drone image processing system as described in claim 1, characterized in that, The method includes the following steps: S1, the image acquisition module acquires target images during the flight of the UAV; S2, the preprocessing module preprocesses the target image and obtains the processed image; S3, the image recognition module recognizes the processed image and obtains the resampling index of the processed image; S4, the user interaction module displays the acquired target image and the recognition results from the image recognition module; S5, the flight control coordination module determines whether the UAV's flight status needs to be adjusted based on the resampling index. If so, it generates control commands and sends them to the UAV; otherwise, it terminates the process.

7. The UAV image processing method based on artificial intelligence according to claim 6, characterized in that, The image recognition module identifies the processed image and obtains the resampling index of the processed image, including the following steps: S31, the image quality detection unit detects various parameters of the processed image; S32, the target detection unit performs image segmentation on the processed image and identifies the target objects in the segmented image; S33, the storage unit outputs the preset parameters required for calculation to the calculation unit; S34, the calculation unit calculates the resampling index based on the detection results of the target detection unit and the acquired parameters.

Images