Image acquisition and data processing system for vertical take-off and landing unmanned aerial vehicle

Abstract

The invention discloses an image acquisition and data processing system for a vertical take-off and landing unmanned aerial vehicle, and relates to the technical field of data processing, and the system comprises a region division unit, an image acquisition unit, an image preprocessing unit, an image feature matching unit and an image geometric transformation unit. The method comprises the following steps that: a plurality of sub-regions are divided into a plurality of sub-regions, each sub-region is numbered, the occurrence of region shooting omission in the acquisition process of an unmanned aerial vehicle is prevented, and meanwhile, an image preprocessing unit performs distortion correction on acquired image information, so that pixel points at offset positions in an image can be adjusted to actual positions, errors existing in the image information are reduced, and the accuracy of the image information is improved. The adjacent image information is smoothly spliced through the image feature matching unit and the image geometric transformation unit, and the existing feature matching points are adjusted by using the genetic algorithm and the RANSAC algorithm in the image feature matching unit, so that the mismatching points can be found and deleted in time.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to an image acquisition and data processing system for a vertical take-off and landing unmanned aerial vehicle. Background Art

[0002] UAV image acquisition technology is to carry an aerial survey camera on a drone to achieve high-precision, high-resolution image acquisition of a specific area, and stitch together several overlapping images collected into a seamless panorama or high-resolution image, which can expand the field of view and achieve the acquisition of panoramic images. The invention patent with application number CN202311629194.7 discloses "an image recognition system and image recognition method based on AI+UAV, which includes: a network interface layer, which is used to receive image recognition requests and scene images to be recognized; a dynamic library layer, which is connected to the network interface layer and is used to store several different image recognition models; And a chip, which is used to carry the network interface layer and the dynamic library layer; the image recognition request and the scene image to be recognized are sent to the dynamic library layer through the network interface layer, and the dynamic library layer receives the image recognition request sent by the network interface layer, and loads the corresponding image recognition model according to the image recognition request, and uses the image recognition model to recognize and detect the scene image to be recognized, and generates an image detection result. This application carries an independent network interface layer, a dynamic library layer, and a decoder on a high-computing power chip, which can complete efficient image detection in real time on the drone body, and at the same time realize the expansion capabilities of the AI ​​framework and image recognition model. ";

[0003] The above-mentioned existing technology solves the problem of being unable to identify and detect image information in the scene, but when the system is running and the shooting area is divided, the drone may miss some areas during the collection process, and no distortion correction is performed on the collected image information, resulting in the position of some pixels in the image information being offset, which will cause large errors in the subsequent analysis of multiple image information. At the same time, the system cannot splice the image information, making it impossible for the operator to obtain more comprehensive image information. Summary of the invention

[0004] The purpose of the present invention is to provide an image acquisition and data processing system for a vertical take-off and landing unmanned aerial vehicle to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above-mentioned object, the present invention provides the following technical solutions: an image acquisition and data processing system for a vertical take-off and landing UAV, comprising an image geometric transformation unit;

[0006] A region division unit, wherein the region division unit divides the photographed region into a plurality of sub-regions, numbers each sub-region one by one, and constructs a corresponding data set according to the number;

[0007] An image acquisition unit, wherein the image acquisition unit uses a HY-T5 aerial survey camera carried by a drone to take pictures of each sub-area. The resolution of the HY-T5 aerial survey camera is 6252×4168 pixels. After setting the shooting height of the drone, 200 images are collected for each sub-area, and these images are stored as initial image information in a corresponding data set;

[0008] An image preprocessing unit, wherein the image preprocessing unit extracts the initial image information in the data set corresponding to each sub-region, calculates the theoretical imaging coordinates of each pixel in the image information using an image distortion correction model, adjusts the position of the pixel in the image information, thereby obtaining the image information after distortion correction, sets the number of pixels corresponding to the grayscale of the current image according to the grayscale distribution probability density function, calculates the image information through a pixel analysis algorithm, obtains the peak signal-to-noise ratio corresponding to each image information, deletes the noise points contained in the image information through a mean filtering method, calculates the pixel mean of the current image information according to the pixel values ​​corresponding to the processed image information using a pixel analysis algorithm, determines the peak signal-to-noise ratio corresponding to the current image information, and compares the peak signal-to-noise ratio value corresponding to the original image information with the peak signal-to-noise ratio value corresponding to the current image information;

[0009] The image feature matching unit uses Gaussian filtering to construct an image scale space, subtracts pixel points of adjacent images to determine local extreme points and their specific positions, constructs feature points and feature vectors corresponding to each image according to the local gradient of the image, and uses genetic algorithm and RANSAC algorithm to adjust the matching points generated from the feature points of adjacent images and delete the mismatched points.

[0010] Preferably, the image preprocessing unit includes a distortion correction module and a grayscale analysis module. After the distortion correction module extracts the initial image information in the data set corresponding to each sub-region, it uses the image distortion correction model to calculate the theoretical imaging coordinates of each pixel in the image information according to the radial distortion coefficient and the pixel distance between each pixel and the image center, and then uses the theoretical imaging coordinates to adjust the pixel position in the image information to obtain the image information after distortion correction, which is stored in the corresponding data set, and the corresponding initial image information is deleted. After the grayscale analysis module processes the grayscale in the image information through a transformation function, it obtains the corrected image grayscale information, and uses the probability density analysis algorithm to calculate according to the corrected image grayscale information to obtain the grayscale distribution probability density function corresponding to the current image, and sets the number of pixels corresponding to the grayscale of the current image according to the grayscale distribution probability density function to obtain new image information, which is stored in the corresponding data set, and the original image information is deleted. The image distortion correction model is specifically:

[0011]

[0012] Among them, x corrected ,y corrected represents the theoretical imaging point of the image point (x, y) in the plane coordinate system, r represents the pixel distance between the image point (x, y) and the image center (x0, y0), and kn represents the radial distortion coefficient.

[0013] Preferably, the image preprocessing unit further includes a noise point removal module and a signal-to-noise ratio determination module. The noise point removal module calculates the image information through a pixel analysis algorithm, obtains the peak signal-to-noise ratio corresponding to each image information, determines the number of noise points in the image information, and deletes the noise points contained in the image information through a mean filtering method. The signal-to-noise ratio determination module determines the pixel value corresponding to each pixel point in the current image information, and calculates the pixel mean of the current image information according to these pixel values ​​using a pixel analysis algorithm, thereby obtaining the peak signal-to-noise ratio corresponding to the current image information. The pixel analysis algorithm is specifically:

[0014]

[0015]

[0016] Where (x, y) represents the pixel coordinates, m represents the number of pixels in the neighborhood S, M represents the number of rows of image pixels, N represents the number of columns of image pixels, and P max Represents the maximum pixel value in the image, R p represents the peak signal-to-noise ratio, f(x,y) represents the pixel value of any point (x,y), and g(x,y) represents the pixel mean.

[0017] Preferably, the image preprocessing unit also includes a signal-to-noise ratio analysis module, which compares the peak signal-to-noise ratio value corresponding to the original image information with the peak signal-to-noise ratio value corresponding to the current image information. If the peak signal-to-noise ratio value corresponding to the current image information is greater than the peak signal-to-noise ratio value corresponding to the original image information and the difference is greater than a threshold, the current image information is stored in a data set and the corresponding original image information is deleted; otherwise, the current image information is processed again by a mean filtering method.

[0018] Preferably, the image feature matching unit includes an image gradient analysis module and a feature vector generation module. After the image gradient analysis module constructs the image scale space using Gaussian filtering, it traverses all images in the image scale space, subtracts the pixel points of adjacent images, determines the local extreme points and specific positions in the subtracted images, uses the local extreme points as key points, and analyzes the image gradients around the key points to determine the main direction and auxiliary direction of each key point. After the feature vector generation module selects the scale and direction, it measures the local gradient of the image in the area around each key point, and constructs the feature points and feature vectors corresponding to each image based on the local gradient of the image.

[0019] Preferably, the image feature matching unit also includes a matching optimization module, which performs rough matching of feature points of adjacent images using a FLANN algorithm, and then uses a genetic algorithm and a RANSAC algorithm to adjust the matching points generated in the rough matching and delete the mismatched points contained therein. The FLANN algorithm is a feature matching and recognition algorithm, the genetic algorithm is a method for searching for the optimal solution by simulating a natural evolution process, and the RANSAC algorithm is an iterative algorithm based on random sampling.

[0020] Preferably, the image geometry transformation unit includes a coordinate translation module and a transformation matrix analysis module. The coordinate translation module translates the coordinates of each feature matching point so that its center of mass is moved to the origin, and scales these points according to a set ratio so that the average distance of all points to the origin is equal to a preset value. The transformation matrix analysis module uses a DLT algorithm to calculate the corresponding projection transformation matrix based on the coordinates of the feature matching points, aligns the overlapping parts of the image, analyzes the offsets corresponding to the feature matching points on the overlapping parts, and translates the coordinates of these points to obtain a new projection transformation matrix. The DLT algorithm is a direct linear transformation algorithm.

[0021] Preferably, the image geometric transformation unit also includes an image stitching generation module, which uses a feature refinement model to delete outliers in the image information, and then adjusts the projection transformation matrix corresponding to the non-overlapping part of the image through a smoothing constraint algorithm, obtains the final image stitching result based on the optimized projection transformation matrix, and outputs it through a visualization interface.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] The present invention divides a shooting area into a plurality of sub-areas through an area division unit, and numbers each sub-area one by one, so as to prevent the occurrence of omission of area shooting by the drone during the acquisition process, and also makes the subsequent image stitching smoother. At the same time, the image preprocessing unit performs distortion correction on the acquired image information, so that the pixel points of the offset position in the image information can be adjusted to the actual position, reducing the error in the image information, and setting the number of pixel points corresponding to the image grayscale level, reducing the color difference between adjacent images, so that the result of image stitching can be more significant, and the pixel analysis algorithm is used to calculate the peak signal-to-noise ratio values ​​corresponding to the original image information and the processed image information, so as to ensure that the noise points in the image can be more comprehensively deleted, and further improve the quality of the image information, and the adjacent image information is smoothly stitched through the image feature matching unit and the image geometric transformation unit, and the genetic algorithm and the RANSAC algorithm are used in the image feature matching unit to adjust the existing feature matching points, so that the erroneous matching points can be discovered and deleted in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Provides a schematic diagram of the overall system flow for an embodiment of the present invention;

[0025] Figure 2 An internal module block diagram of an image preprocessing unit provided by an embodiment of the present invention;

[0026] Figure 3 An internal module block diagram of an image feature matching unit provided by an embodiment of the present invention;

[0027] Figure 4 This is a block diagram of the internal modules of the image geometric transformation unit provided in an embodiment of the present invention.

[0028] In the figure: 1. region division unit; 2. image acquisition unit; 3. image preprocessing unit; 301. distortion correction module; 302. gray level analysis module; 303. noise removal module; 304. signal-to-noise ratio determination module; 305. signal-to-noise ratio analysis module; 4. image feature matching unit; 401. image gradient analysis module; 402. feature vector generation module; 403. matching optimization module; 5. image geometry transformation unit; 501. coordinate translation module; 502. transformation matrix analysis module; 503. image stitching generation module. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] See also Figure 1-4 ,The present invention provides a technical solution: an image acquisition and data processing system for a vertical take-off and landing UAV, comprising an image geometric transformation unit 5;

[0031] The region division unit 1 divides the photographed region into a plurality of sub-regions, numbers each sub-region one by one, and constructs a corresponding data set according to the number;

[0032] Image acquisition unit 2, image acquisition unit 2 uses the HY-T5 aerial survey camera carried by the drone to take pictures of each sub-area. The resolution of the HY-T5 aerial survey camera is 6252×4168 pixels. After setting the shooting height of the drone, 200 images are collected for each sub-area, and these images are stored as initial image information in the corresponding data set;

[0033] The image preprocessing unit 3 extracts the initial image information in the data set corresponding to each sub-region, calculates the theoretical imaging coordinates of each pixel in the image information using the image distortion correction model, adjusts the pixel position in the image information, and obtains the image information after distortion correction, sets the number of pixels corresponding to the gray level of the current image according to the gray level distribution probability density function, calculates the image information through the pixel analysis algorithm, obtains the peak signal-to-noise ratio corresponding to each image information, deletes the noise points contained in the image information through the mean filtering method, calculates the pixel mean of the current image information according to the pixel values ​​corresponding to the processed image information using the pixel analysis algorithm, determines the peak signal-to-noise ratio corresponding to the current image information, and compares the peak signal-to-noise ratio value corresponding to the original image information with the peak signal-to-noise ratio value corresponding to the current image information;

[0034] Image feature matching unit 4, image feature matching unit 4 uses Gaussian filtering to construct an image scale space, subtracts the pixels of adjacent images to determine the local extreme points and their specific positions, constructs the feature points and feature vectors corresponding to each image according to the local gradient of the image, and uses genetic algorithm and RANSAC algorithm to adjust the matching points generated from the feature points of adjacent images and delete the mismatched points.

[0035] The image preprocessing unit 3 includes a distortion correction module 301 and a grayscale analysis module 302. After the distortion correction module 301 extracts the initial image information in the data set corresponding to each sub-region, it uses the image distortion correction model to calculate the theoretical imaging coordinates of each pixel in the image information according to the radial distortion coefficient and the pixel distance between each pixel and the image center, and then uses the theoretical imaging coordinates to adjust the pixel position in the image information to obtain the image information after distortion correction, which is stored in the corresponding data set, and the corresponding initial image information is deleted. The grayscale analysis module 302 processes the grayscale in the image information through a transformation function to obtain the corrected image grayscale information, and uses a probability density analysis algorithm to calculate according to the corrected image grayscale information to obtain the grayscale distribution probability density function corresponding to the current image, and sets the number of pixels corresponding to the grayscale of the current image according to the grayscale distribution probability density function to obtain new image information, which is stored in the corresponding data set, and the original image information is deleted. The image distortion correction model is specifically as follows:

[0036]

[0037] Among them, x corrected ,y corrected represents the theoretical imaging point of the image point (x, y) in the plane coordinate system, r represents the pixel distance between the image point (x, y) and the image center (x0, y0), and kn represents the radial distortion coefficient;

[0038] The specific probability density analysis algorithm is:

[0039]

[0040] Where T(r) represents a monotonically increasing transformation function in the interval [0,1], r represents the grayscale of the original image, s represents the grayscale of the corrected image, and P r (r) represents the gray level distribution probability density of the original image, P s (s) represents the gray level distribution probability density of the corrected image;

[0041] The image preprocessing unit 3 also includes a noise point removal module 303 and a signal-to-noise ratio determination module 304. The noise point removal module 303 calculates the image information through a pixel analysis algorithm, obtains the peak signal-to-noise ratio corresponding to each image information, determines the number of noise points in the image information, and deletes the noise points contained in the image information through a mean filtering method. After the signal-to-noise ratio determination module 304 determines the pixel value corresponding to each pixel point in the current image information, it calculates the pixel mean of the current image information according to these pixel values ​​using a pixel analysis algorithm, thereby obtaining the peak signal-to-noise ratio corresponding to the current image information. The pixel analysis algorithm is specifically as follows:

[0042]

[0043] Where (x, y) represents the pixel coordinates, m represents the number of pixels in the neighborhood S, M represents the number of rows of image pixels, N represents the number of columns of image pixels, and P max Represents the maximum pixel value in the image, R p represents the peak signal-to-noise ratio, f(x,y) represents the pixel value of any point (x,y), and g(x,y) represents the pixel mean;

[0044] The image preprocessing unit 3 also includes a signal-to-noise ratio analysis module 305, which compares the peak signal-to-noise ratio value corresponding to the original image information with the peak signal-to-noise ratio value corresponding to the current image information. If the peak signal-to-noise ratio value corresponding to the current image information is greater than the peak signal-to-noise ratio value corresponding to the original image information and the difference is greater than a threshold, the current image information is stored in the data set and the corresponding original image information is deleted. Otherwise, the current image information is processed again by a mean filtering method.

[0045] The image feature matching unit 4 includes an image gradient analysis module 401 and a feature vector generation module 402. After the image gradient analysis module 401 constructs an image scale space using Gaussian filtering, it traverses all images in the image scale space, subtracts the pixel points of adjacent images, determines the local extreme value points and specific positions in the subtracted images, takes the local extreme value points as key points, and analyzes the image gradients around the key points to determine the main direction and auxiliary direction of each key point. After the feature vector generation module 402 selects the scale and direction, it measures the local gradient of the image in the area around each key point, and constructs the feature points and feature vectors corresponding to each image according to the local gradient of the image.

[0046] The image feature matching unit 4 also includes a matching optimization module 403. After the matching optimization module 403 performs rough matching on the feature points of adjacent images by using the FLANN algorithm, it uses the genetic algorithm and the RANSAC algorithm to adjust the matching points generated in the rough matching and delete the mismatched points contained therein. The FLANN algorithm is an algorithm for feature matching and recognition, the genetic algorithm is a method for searching for the optimal solution by simulating the natural evolution process, and the RANSAC algorithm is an iterative algorithm based on random sampling.

[0047] The image geometric transformation unit 5 includes a coordinate translation module 501 and a transformation matrix analysis module 502. The coordinate translation module 501 translates the coordinates of each feature matching point so that its centroid point moves to the origin, and scales these points according to a set ratio so that the average distance of all points to the origin is equal to a preset value. The transformation matrix analysis module 502 uses a DLT algorithm to calculate the corresponding projection transformation matrix based on the coordinates of the feature matching points, aligns the overlapping parts of the image, analyzes the offset corresponding to the feature matching points on the overlapping parts, and translates the coordinates of these points to obtain a new projection transformation matrix. The DLT algorithm is a direct linear transformation algorithm.

[0048] The image geometric transformation unit 5 also includes an image stitching generation module 503. After deleting outliers in the image information using the feature refinement model, the image stitching generation module 503 adjusts the projection transformation matrix corresponding to the non-overlapping part of the image through a smooth constraint algorithm, obtains the final image stitching result according to the optimized projection transformation matrix, and outputs it through a visual interface. The smooth constraint algorithm is specifically as follows:

[0049] g s (x,y)=ηg(x,y)

[0050]

[0051] Among them, [x l ,x u] represents the boundary of the overlapping area in the x direction, [y l ,y u ] represents the boundary of the overlapping area in the y direction, d s represents the smooth transition width, g(x,y) represents the deformation function, g s (x,y) represents the constrained deformation function, η represents the scale parameter, and x,y represent the parameters.

[0052] Working principle: The present invention constructs a corresponding data set for each sub-region through the region division unit 1, and uses the image acquisition unit 2 to collect 200 image information for each sub-region, and stores these image information as initial image information in the corresponding data set. The distortion correction module 301 in the image preprocessing unit 3 adjusts the pixel position in the image information to obtain the image information after distortion correction, and stores it in the corresponding data set. The grayscale analysis module 302 sets the number of pixels corresponding to the grayscale of the current image to obtain new image information, and stores it in the corresponding data set. The noise deletion module 303 deletes the noise contained in the image information through the mean filtering method. The signal-to-noise ratio determination module 304 uses the pixel analysis algorithm to calculate the peak signal-to-noise ratio corresponding to the current image information. The peak signal-to-noise ratio value corresponding to the original image information and the peak signal-to-noise ratio value corresponding to the current image information are compared by the signal-to-noise ratio analysis module 305. If the current image signal When the peak signal-to-noise ratio value corresponding to the new information is greater than the peak signal-to-noise ratio value corresponding to the original image information and the difference is greater than the threshold, the current image information is stored in the data set, and the image gradient analysis module 401 in the image feature matching unit 4 is used to determine the main direction and auxiliary direction of each key point. The feature vector generation module 402 constructs the feature points and feature vectors corresponding to each image. The matching optimization module 403 uses the genetic algorithm and the RANSAC algorithm to adjust the matching points generated in the feature points of adjacent images. The coordinate translation module 501 in the image geometry transformation unit 5 is used to make the average distance of the coordinates of all feature matching points to the origin equal to a preset value. The transformation matrix analysis module 502 obtains a new projection transformation matrix. The image stitching generation module 503 adjusts the projection transformation matrix corresponding to the non-overlapping part of the image through a smooth constraint algorithm. The final image stitching result is obtained according to the optimized projection transformation matrix and is output through a visual interface.

[0053] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0054] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An image acquisition and data processing system for a vertical take-off and landing unmanned aerial vehicle, comprising an image geometric transformation unit (5), characterized in that: A region division unit (1), wherein the region division unit (1) divides the photographed region into a plurality of sub-regions, numbers each sub-region one by one, and constructs a corresponding data set according to the number; An image acquisition unit (2), wherein the image acquisition unit (2) uses a HY-T5 aerial survey camera carried by the drone to take pictures of each sub-area, wherein the resolution of the HY-T5 aerial survey camera is 6252×4168 pixels. After setting the shooting height of the drone, 200 pieces of image information are collected for each sub-area, and the image information is stored as initial image information in a corresponding data set; An image preprocessing unit (3), wherein the image preprocessing unit (3) extracts initial image information from a data set corresponding to each sub-region, calculates theoretical imaging coordinates of each pixel in the image information using an image distortion correction model, adjusts the positions of the pixels in the image information, thereby obtaining image information after distortion correction, sets the number of pixels corresponding to the grayscale of the current image according to a grayscale distribution probability density function, calculates the image information using a pixel analysis algorithm, obtains a peak signal-to-noise ratio corresponding to each image information, deletes noise points contained in the image information using a mean filtering method, calculates a pixel mean of the current image information based on pixel values ​​corresponding to the processed image information using a pixel analysis algorithm, determines a peak signal-to-noise ratio corresponding to the current image information, and compares the peak signal-to-noise ratio value corresponding to the original image information with the peak signal-to-noise ratio value corresponding to the current image information; An image feature matching unit (4) is provided, wherein the image feature matching unit (4) uses Gaussian filtering to construct an image scale space, subtracts pixel points of adjacent images to determine local extreme value points and their specific positions, constructs feature points and feature vectors corresponding to each image according to local image gradients, and uses a genetic algorithm and a RANSAC algorithm to adjust matching points generated from feature points of adjacent images and delete mismatched points.

2. The image acquisition and data processing system for a vertical take-off and landing unmanned aerial vehicle according to claim 1, characterized in that: The image preprocessing unit (3) comprises a distortion correction module (301) and a grayscale analysis module (302). After extracting the initial image information in the data set corresponding to each sub-region, the distortion correction module (301) uses an image distortion correction model to calculate the theoretical imaging coordinates of each pixel in the image information according to the radial distortion coefficient and the pixel distance between each pixel and the image center, and then uses the theoretical imaging coordinates to adjust the pixel position in the image information to obtain the image information after distortion correction, which is stored in the corresponding data set, and the corresponding initial image information is deleted. After processing the grayscale in the image information through a transformation function, the grayscale analysis module (302) obtains the corrected image grayscale information, and uses a probability density analysis algorithm to calculate based on the corrected image grayscale information to obtain the grayscale distribution probability density function corresponding to the current image. According to the grayscale distribution probability density function, the number of pixels corresponding to the grayscale of the current image is set to obtain new image information, which is stored in the corresponding data set, and the original image information is deleted.

3. The image acquisition and data processing system for a vertical take-off and landing unmanned aerial vehicle according to claim 2, characterized in that: The image preprocessing unit (3) further comprises a noise point removal module (303) and a signal-to-noise ratio determination module (304). The noise point removal module (303) calculates the image information through a pixel analysis algorithm to obtain a peak signal-to-noise ratio corresponding to each image information, determines the number of noise points in the image information, and removes the noise points contained in the image information through a mean filtering method. The signal-to-noise ratio determination module (304) determines the pixel value corresponding to each pixel point in the current image information, and calculates the pixel mean of the current image information based on these pixel values ​​using a pixel analysis algorithm, thereby obtaining the peak signal-to-noise ratio corresponding to the current image information.

4. The image acquisition and data processing system for a vertical take-off and landing unmanned aerial vehicle according to claim 3, characterized in that: The image preprocessing unit (3) further comprises a signal-to-noise ratio analysis module (305), wherein the signal-to-noise ratio analysis module (305) compares a peak signal-to-noise ratio value corresponding to the original image information with a peak signal-to-noise ratio value corresponding to the current image information; if the peak signal-to-noise ratio value corresponding to the current image information is greater than the peak signal-to-noise ratio value corresponding to the original image information and the difference is greater than a threshold, the current image information is stored in a data set and the corresponding original image information is deleted; otherwise, the current image information is processed again by a mean filtering method.

5. The image acquisition and data processing system for a vertical take-off and landing unmanned aerial vehicle according to claim 1, characterized in that: The image feature matching unit (4) comprises an image gradient analysis module (401) and a feature vector generation module (402). After the image gradient analysis module (401) constructs an image scale space using Gaussian filtering, it traverses all images in the image scale space, subtracts pixel points of adjacent images, determines local extreme value points and specific positions in the subtracted images, uses the local extreme value points as key points, and analyzes image gradients around the key points to determine the main direction and auxiliary direction of each key point. After the feature vector generation module (402) selects the scale and direction, it measures the local gradient of the image in the area around each key point, and constructs feature points and feature vectors corresponding to each image based on the local gradient of the image.

6. The image acquisition and data processing system for a vertical take-off and landing unmanned aerial vehicle according to claim 5, characterized in that: The image feature matching unit (4) also includes a matching optimization module (403), which performs rough matching of feature points of adjacent images using a FLANN algorithm, and then uses a genetic algorithm and a RANSAC algorithm to adjust matching points generated in the rough matching, thereby deleting mismatched points contained therein.

7. The image acquisition and data processing system for a vertical take-off and landing unmanned aerial vehicle according to claim 1, characterized in that: The image geometric transformation unit (5) comprises a coordinate translation module (501) and a transformation matrix analysis module (502). The coordinate translation module (501) translates the coordinates of each feature matching point so that its centroid is moved to the origin, and scales these points according to a set ratio so that the average distance of all points to the origin is equal to a preset value. The transformation matrix analysis module (502) uses a DLT algorithm to calculate the corresponding projection transformation matrix based on the coordinates of the feature matching points, aligns the overlapping parts of the images, analyzes the offsets corresponding to the feature matching points on the overlapping parts, and translates the coordinates of these points to obtain a new projection transformation matrix.

8. The image acquisition and data processing system for a vertical take-off and landing unmanned aerial vehicle according to claim 7, characterized in that: The image geometric transformation unit (5) also includes an image splicing generation module (503). After deleting outliers in the image information using a feature refinement model, the image splicing generation module (503) adjusts the projection transformation matrix corresponding to the non-overlapping part of the image using a smoothing constraint algorithm, obtains a final image splicing result based on the optimized projection transformation matrix, and outputs it through a visual interface.

Images