Flight simulation method and system based on unmanned aerial vehicle-mounted backlight imaging technology

Abstract

According to the flight simulation method and system based on the unmanned aerial vehicle-mounted backlight imaging technology, the key scientific problem of target perception and recognition of an unmanned aerial vehicle platform in the backlight environment is solved, and an innovative scheme is provided for cross research of high-dynamic-range imaging, target detection and intelligent flight control. A backlight imaging camera and a dynamic aperture control technology are introduced, and a multi-point calibration and multi-scale noise reduction method is combined, so that the traditional problems of insufficient imaging dynamic range, loss of image details, random noise interference and the like in a backlight environment are effectively solved. Meanwhile, based on the image enhancement method of the generative adversarial network, high-fidelity reconstruction of target texture and structural features under a complex illumination condition is realized, and the common problems of artifacts and color drift in a traditional enhancement algorithm are avoided. Furthermore, through the combination of a reinforced and optimized YOLO detection framework and a high-resolution feature pyramid network, a small-target high-precision detection model adapted to a backlight scene is successfully constructed, and the detection robustness and real-time performance are remarkably improved while the detection precision is kept.

Description

Technical Field

[0001] The present invention belongs to the field of computer vision and unmanned aerial vehicle application technology, and specifically relates to a technology for image enhancement, detection and tracking of a target in a backlit or high dynamic lighting scene. Background Art

[0002] With the development of drone technology, drones are increasingly used in intelligent monitoring, rescue search, ecological monitoring and other fields. However, in complex lighting environments, especially backlight conditions, drone target perception performance still faces significant technical bottlenecks. Such environments usually lead to uneven illumination of the imaging target, decreased contrast, and loss of details, which seriously affects the accuracy of target detection and tracking.

[0003] Backlighting is a high dynamic range lighting scenario where the background light source (such as the sun) is much stronger than the light reflected by the target, resulting in: 1) the target is too dark or completely covered by shadows: ordinary imaging systems have difficulty distinguishing the details of the background and the target; 2) overexposure of highlight areas: the excessive brightness of the background will mask image details and cause serious information loss; 3) blurred target outlines: light scattering or strong edge halo effects make it difficult to extract the target edge. Traditional imaging technologies such as high dynamic range imaging can partially improve the brightness range, but usually require the fusion of multiple image frames, which is difficult to meet the requirements of drone real-time perception tasks.

[0004] Current backlit image processing technologies are mainly divided into the following categories: 1) Brightness enhancement based on global adjustment: Enhance the target area by increasing the overall brightness of the image. However, this method easily causes background overexposure or insufficient contrast between the target and the background, and cannot accurately separate the target; 2) Local enhancement based on the Retinex model: The Retinex model achieves target enhancement through local contrast adjustment, but it is often accompanied by artifacts (such as chromatic aberration) or texture detail loss in backlit scenes; 3) Image enhancement based on deep learning: Use generative adversarial networks or convolutional neural networks to enhance backlit scene images. Although it improves the ability to restore details, the model is usually highly dependent on training data and is difficult to generalize to actual complex scenes. In addition, this type of method has high requirements for hardware computing resources and is difficult to implement in real-time drone applications.

[0005] Under backlight conditions, target detection and tracking technology faces the following challenges: 1) Low-contrast targets are difficult to detect: Due to the large contrast in light intensity, the distinction between the target and the background is reduced, and the accuracy of existing detection algorithms (such as YOLO or SSD) is significantly reduced; 2) Small target detection is unstable: When the drone flies at a high altitude, the resolution of small targets is limited and the target is easily disturbed by background noise; 3) Multiple target occlusion and discontinuous tracking: Under backlight conditions, the target may not be tracked continuously due to occlusion or blurred edges, resulting in loss of tracking trajectory. Summary of the invention

[0006] The technical problems to be solved by the present invention are:

[0007] In order to solve the problems of poor robustness and real-time performance of target perception in target detection and tracking technology under backlight conditions, limited resolution of small targets when the UAV is flying at a high altitude, and multi-target occlusion and tracking discontinuity, the present invention provides a flight simulation method and system based on UAV-mounted backlight imaging technology.

[0008] The technical solution adopted by the present invention to solve the above technical problems is:

[0009] The present invention proposes a flight simulation method and system based on unmanned aerial vehicle backlighting imaging technology, the method comprising the following steps:

[0010] Step 1: construct a UAV platform, which includes a backlight imaging camera, a stabilized gimbal, a real-time data transmission module and an image processor mounted on the UAV. The backlight imaging camera is installed on the stabilized gimbal, and the backlight imaging camera transmits the perceived image data of the UAV during flight to the image processor through the real-time data transmission module;

[0011] Step 2: Fix the backlight imaging camera on a stabilized gimbal to ensure that its field of view is consistent with the flight direction of the drone and has the ability to adjust in real time;

[0012] Step 3: Use a multi-point calibration method to accurately calibrate the internal and external parameters of the backlight imaging camera (backlight imaging device) to ensure that lens distortion and geometric errors are eliminated;

[0013] Step 4: The backlight imaging camera is equipped with a dynamic aperture control mechanism to achieve automatic exposure adjustment under drastic changes in lighting conditions and optimize the imaging quality of the target area;

[0014] Step 5: Introduce a multi-scale noise reduction algorithm based on wavelet transform to suppress random noise interference caused by high dynamic range under backlight conditions;

[0015] Step 6: Build a lightweight generative adversarial network model, input the backlit image and reconstruct the target texture and structural features through multi-layer feature extraction, avoiding artifacts or color drift problems caused by traditional methods;

[0016] Step 7: Based on the enhanced and optimized YOLO detection framework, combined with the prior backlight scene data, a real-time target detection model is constructed to extract the bounding boxes of multi-category targets and their classification probabilities; a high-resolution feature pyramid network is used for small target areas to ensure the accuracy and robustness of target detection; a multi-feature fusion method is used to further extract target key points and their spatial geometric characteristics, providing a standardized high-dimensional data description;

[0017] Step 8: Superimpose the UAV flight status parameters and light distribution parameters in real time to complete the flight simulation display of the UAV-mounted backlight imaging technology.

[0018] Furthermore, the specific process of step one includes:

[0019] Equipped with a backlight imaging camera, a three-axis mechanically stabilized gimbal and a high-bandwidth real-time data transmission module, it forms a complete UAV perception and transmission system framework.

[0020] Furthermore, the specific process of step 2 includes:

[0021] The gimbal uses high-precision servo control technology to ensure that the camera's line of sight is stable and can be precisely adjusted under complex flight conditions.

[0022] Furthermore, the specific process of step three includes:

[0023] The multi-point calibration method is used to accurately calibrate the internal parameters (focal length, principal point position, distortion coefficient) and external parameters (camera posture and position) of the backlight camera. Then, the calibration process eliminates errors such as lens distortion, geometric offset and optical axis inconsistency to ensure the geometric accuracy of the acquired image.

[0024] Furthermore, the specific process of step 4 includes:

[0025] The deployment of a dynamic aperture control module based on adaptive light perception can quickly respond to light changes in backlit environments, thereby realizing a linkage adjustment mechanism for aperture, shutter speed, and ISO parameters to ensure that the imaging quality of the target area is optimal in both bright and dark areas.

[0026] Furthermore, the specific process of step five includes:

[0027] A multi-scale denoising algorithm based on wavelet transform is introduced to perform noise suppression on backlit imaging data. In the denoising process, the high-frequency and low-frequency components of the image are separated to eliminate random noise interference while retaining the detailed features of the target area.

[0028] Furthermore, the specific process of step six includes:

[0029] A lightweight generative adversarial network model is designed. The original image data under backlighting conditions is input, and the texture details and structural features of the target are reconstructed through multi-layer feature extraction and illumination compensation. Then, the generative network is optimized using adversarial loss and perceptual loss to avoid artifacts and color drift problems in traditional algorithms in backlighting enhancement.

[0030] Furthermore, the specific process of the flight simulation demonstration of the UAV-mounted backlight imaging technology in step eight includes:

[0031] The backlight imaging camera carried by the drone collects environmental image data during flight, including target objects, background scenes, and light distribution information. The collected data must be optimized by dynamic aperture and backlight perception imaging technology to ensure clear imaging under conditions of strong light changes. In the backlight scene, the flight status parameters of the drone are superimposed on the screen to dynamically present the flight status. Then, intuitive superimposed annotations (such as bounding boxes, classification labels, and key points) are used to help analyze the target perception ability under backlight conditions.

[0032] According to another aspect of the present invention, a flight simulation system (software) based on unmanned aerial vehicle backlight imaging technology is proposed, which includes:

[0033] A backlight imaging camera system is used to obtain environmental image data under backlight conditions; the dynamic aperture control mechanism configured therein optimizes the imaging quality of the target area through automatic exposure adjustment; the high-quality environmental image data obtained under backlight conditions is input to the data processing and image enhancement system in the image processor; corresponding to steps one, three, and four, and used to realize the functions of steps one, three, and four;

[0034] The stabilization control system is used to ensure the stability of the imaging device during flight, the field of view adjustment capability, and the quality of image acquisition; the servo system of the stabilization gimbal controls the precise motion state of the mechanical device, especially the precise operation in terms of position, speed, and acceleration; corresponding to step 2, and is used to realize the function of step 2;

[0035] The image processor is used to read high-quality environmental image data under backlight conditions and provide basic processing for target perception and recognition. It is equipped with a data processing and image enhancement system and a target detection and feature extraction system.

[0036] A data processing and image enhancement system, which includes a wavelet transform multi-scale noise reduction algorithm, a lightweight generative adversarial network and a multi-scale image enhancement algorithm, and is used to reduce noise, enhance and reconstruct target features of original image data under backlight conditions, thereby improving perception and recognition performance; corresponding to step five, and is used to implement the functions of step five;

[0037] The target detection and feature extraction system uses the YOLO target detection framework to detect multiple categories of targets in backlit scenes in real time, extract key features and generate standardized descriptions to provide input for flight decisions. It corresponds to steps six and seven and is used to implement the functions of steps six and seven.

[0038] The beneficial technical effects of the present invention are:

[0039] The present invention effectively solves the key scientific problems of target perception and recognition of UAV platforms in backlit environments, and provides an innovative solution for the cross-research of high dynamic range imaging, target detection and intelligent flight control. By introducing backlit imaging cameras and dynamic aperture control technology, combined with multi-point calibration and multi-scale noise reduction methods, the traditional problems such as insufficient imaging dynamic range, loss of image details and random noise interference in backlit environments are effectively overcome. At the same time, the image enhancement method based on the generative adversarial network achieves high-fidelity reconstruction of target texture and structural features under complex lighting conditions, avoiding the common artifacts and color drift problems in traditional enhancement algorithms. Furthermore, by combining the enhanced and optimized YOLOv8 detection framework with a high-resolution feature pyramid network, a high-precision detection model for small targets adapted to backlit scenes was successfully constructed, which significantly improved the robustness and real-time performance of the detection while maintaining the detection accuracy.

[0040] The advantages of the present invention are reflected in many aspects: first, this technology breaks through the dependence of traditional target detection methods on standard lighting conditions, especially in complex scenes such as backlighting, and achieves significant improvement in imaging and detection performance; second, through the deep collaboration between the UAV flight control system and the backlighting perception module, the real-time and stability problems of data under high dynamic lighting changes are solved; finally, through the three-dimensional flight simulation display platform, the UAV backlighting perception data and flight status parameters are integrated in real-time dynamic visualization for the first time, which not only verifies the effectiveness of the perception and detection algorithms, but also provides a new research tool for UAV mission planning and performance optimization in complex scenes.

[0041] The present invention is a comprehensive solution combining illumination modeling, image enhancement and target detection, which fully considers the dynamics and hardware limitations of drones, significantly improves the robustness and real-time performance of target perception under backlight conditions, and provides new technical support for the application of drones in complex environments. The patented technology of the present invention can ensure clear imaging of backlight targets while reducing the complexity of drone imaging systems, support drone applications in more scenarios, and improve drone autonomous perception systems.

[0042] The present invention is suitable for image enhancement, detection and tracking of targets in backlit or high dynamic lighting scenes, and is used in the fields of intelligent monitoring, emergency rescue, disaster monitoring and full-angle autonomous perception and navigation of unmanned aerial vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The present invention can be more deeply understood by referring to the description provided below in combination with the accompanying drawings. These drawings and the following detailed description together constitute an important part of this specification, which not only intuitively demonstrates the technical framework and implementation process of the present invention, but also further explains the core principles and technical advantages of the present invention through specific preferred embodiments. Through these drawings and descriptions, it is possible to clearly understand how the present invention achieves high-quality imaging, precise target perception and recognition, and the overall technical solution and its practical application effect of dynamic flight simulation display in a backlit environment.

[0044] Figure 1 It is a flow chart of a flight simulation method based on unmanned aerial vehicle backlighting imaging technology according to an embodiment of the present invention.

[0045] Figure 2 Schematic diagram of constructing a UAV platform equipped with a backlight imaging system in an embodiment of the present invention.

[0046] Figure 3 It is a schematic diagram of multi-point calibration of a backlight camera mounted on a drone in an embodiment of the present invention.

[0047] Figure 4 It is a flow chart of image processing and target detection and recognition in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] like Figure 1-4 As shown, the present invention relates to a flight simulation method and system based on unmanned aerial vehicle backlight imaging technology, focusing on solving the technical problems of target perception and recognition in backlight scenes, and realizing high-quality flight simulation display through multi-technical collaboration. The following describes the implementation method in detail from four aspects: system construction, image processing, target detection and recognition, and flight simulation display.

[0049] The present invention first designs a hardware platform of a backlight imaging system, including a backlight imaging camera, a dynamic aperture module, a stabilized gimbal, a real-time data transmission module and a drone body.

[0050] Step 1: Build a UAV platform equipped with a backlight imaging system;

[0051] The backlight imaging camera is installed on a three-axis stabilized gimbal, which realizes attitude adjustment through linkage with the UAV flight control system to ensure that the backlight imaging camera's field of view is consistent with the target area. The overall structure of the UAV platform has been optimized to reduce wind resistance and improve flight stability.

[0052] Step 2: Fix the backlight imaging camera on the gimbal;

[0053] The control of the stabilized gimbal uses a three-dimensional rotation matrix, and the calculation formula is as follows:

[0054] R=R z (ψ)R y (θ)R x (φ) (1)

[0055] Among them, R z (ψ),R y (θ),R x (φ) represents the rotation matrix around the x, y, and z axes respectively. The gimbal adjusts the camera viewing angle in real time through feedback closed-loop control. The formula is:

[0056] e angle =θ desired -θ current (2)

[0057] The controller is based on the angle error e angle Output the corresponding driving signal.

[0058] Step 4: Configure a dynamic aperture control mechanism to achieve automatic exposure adjustment under drastic changes in lighting conditions and optimize the imaging quality of the target area;

[0059] The dynamic aperture module adjusts the aperture size in real time according to the ambient light conditions and adopts an exposure value optimization strategy. The aperture adjustment formula is:

[0060]

[0061] Among them, A is the aperture diameter, ΔL is the light difference, k is the adjustment coefficient, and t is the exposure time. Through the dynamic combination of aperture adjustment and exposure time, the system can accurately image the target area under strong backlight.

[0062] Step 5: Introduce a multi-scale noise reduction algorithm based on wavelet transform to suppress random noise interference caused by high dynamic range under backlight conditions;

[0063] The backlight imaging camera is generated by encoding and restoring the mask of the target in the strong light background. The imaging quality of the optical system depends on the phase distribution of the exit pupil, so the aberration in the optical system will cause the phase change of the exit pupil. In physical optics, the generalized pupil function is often used to represent the phase distribution of the pupil. Considering the system defocus τ, the specific expression of the exit pupil function can be given:

[0064] P(x,y,τ)=p(x,y)exp{i[f(x,y)+τ(x 2 +y 2 )]} (4)

[0065]

[0066] Among them, i 2=-1, τ represents the phase distribution of the defocus aberration on the pupil, D is the diameter of the system pupil, and f(x, y) represents the phase mask function of the optical system when the phase plate is inserted into the defocused state.

[0067] After knowing the exit pupil function of the system, the point spread function used to characterize the imaging characteristics of the system in the spatial domain can be given:

[0068] h(x′,y′,τ)=κ{F[P(x,y,τ)]F * [P(x,y,τ)]} (6)

[0069] Where F represents Fourier transform, κ is a normalization constant, and * represents conjugation.

[0070] After obtaining the point spread function of optical imaging, Fourier transform can be performed to obtain the optical transfer function that characterizes the spatial frequency transfer characteristics of the system:

[0071] H(u,v,τ)=F[h(x′,y′,τ)] (7)

[0072] The backlight imaging camera is able to calculate the optical transfer function in advance through the phase plate shape, and the physical system can perform a Fourier transform on the point spread function of the point light source imaging acquisition system to obtain the optical transfer function of the system.

[0073] Wiener filtering is called minimum mean square error filtering. This method is to treat the image and noise as random processes, and the ultimate goal is to obtain a statistical error function: The minimum estimate of Where f is the unprocessed original image information, is the minimum estimate of the original image, and E is the expectation of the difference between the original image and the minimum estimate. The expression formula is:

[0074]

[0075] Among them, |H(u,v) 2 =H * (u,v)H(u,v),H * (u,v) represents the complex conjugate of H(u,v), S n (u,v)=|N(u,v) 2 represents the noise power spectrum of image processing; S f (u,v)=|F(u,v) 2 Represents the power spectrum of the original image.

[0076] Under normal conditions, the distribution of most signals and noise cannot produce accurate results, so equation (8) can be simplified to:

[0077]

[0078] There will inevitably be some random noise in the image perceived by the backlight imaging camera. In order to suppress the random noise in the high dynamic range image, multi-scale wavelet transform is used for noise reduction. The input image I(x, y) is decomposed into low-frequency component L and high-frequency component H i , the specific formula is:

[0079]

[0080] Noise reduction is done by soft thresholding, the formula is:

[0081]

[0082] Among them, H i ′ is the high-frequency component after noise reduction, and λ is the noise threshold, which is automatically estimated through noise statistics.

[0083] Step 6: Design a lightweight generative adversarial network model, input the backlit image and reconstruct the target texture and structural features through multi-layer feature extraction, avoiding artifacts or color drift problems caused by traditional methods;

[0084] Step 7: Based on the enhanced and optimized YOLOv8 detection framework and combined with prior backlight scene data, a real-time target detection model is constructed to extract the bounding boxes of multi-category targets and their classification probabilities;

[0085] The target detection system uses the enhanced and optimized YOLOv8 framework. Input backlit image I input The detection network extracts multi-scale features and outputs the bounding box and category probability of the target through the fully connected layer. The loss function L of the bounding box box for:

[0086]

[0087] in, is the predicted bounding box, is the real bounding box. The classification loss uses the cross entropy function, and the formula is:

[0088]

[0089] The complete detection loss is:

[0090] L total =L box +L cls (14)

[0091] On the basis of detecting the target, the key point information is further extracted. The key point is determined by the response function R:

[0092]

[0093] Among them, Ω(k i ) is the pixel area around the key point, and F(p) is the eigenvalue. Multi-eigenvalue fusion is performed through the feature pyramid network to integrate high-resolution features and low-resolution global features to generate the final target description.

[0094] Verification of the technical effect of the present invention:

[0095] The deep learning-based method for recognizing weak targets in backlit backgrounds proposed in the present invention effectively solves the problems of complex backgrounds, weak target contrast, and blurred target features under backlit conditions by combining image preprocessing, deep convolutional neural networks (CNNs), and data enhancement technology. The following is a specific implementation case of the present invention, which demonstrates its technical effects in actual scenarios.

[0096] Case 1: Pedestrian Detection in Autonomous Driving

[0097] In autonomous driving systems, pedestrian detection is a crucial task. Especially in backlit environments such as early morning or dusk, the strong sunlight in front will cause the background to be overexposed, the pedestrian's outline to be blurred, or even completely unrecognizable. In this case, traditional feature-based pedestrian detection methods are difficult to cope with.

[0098] Implementation process:

[0099] Dataset preparation: First, a large number of images with backlighting backgrounds were generated through backlighting camera imaging. This includes autonomous driving datasets for different backlighting scenarios, and the datasets were enhanced. The enhancement methods include simulating direct sunlight of different intensities and angles, as well as changes in lighting under various weather conditions.

[0100] Image preprocessing and enhancement: Adaptive contrast enhancement is then performed on the original image to improve the local contrast of the image and ensure the visibility of pedestrians in backlight conditions. Secondly, the details of the pedestrians are retained by removing background highlights. Noise and shadows in the image are also effectively removed.

[0101] Deep learning model training and detection: Use the YOLOv5 model for target detection. The model extracts the features of pedestrians in backlit backgrounds through multiple convolutional layers, and combines feature fusion technology to enhance the feature expression of low-contrast areas.

[0102] Test results: During the test, the system was able to accurately detect pedestrians in different backlighting environments, especially in low contrast and high light intensity conditions, with recognition accuracy improved by about 15% compared to traditional methods.

[0103] This implementation case shows that the target detection method based on deep learning, especially through the combination of backlight camera perception imaging, image preprocessing and deep convolutional neural network, can effectively improve the accuracy and robustness of pedestrian recognition in backlight background. The application of this method in the autonomous driving system can significantly reduce the false recognition rate caused by backlighting and improve the reliability of pedestrian detection.

[0104] Case 2: Vehicle identification in security monitoring

[0105] In the field of security monitoring, especially in road traffic monitoring, vehicle identification under backlight conditions is a long-standing problem. In particular, when the camera is set in an inconsistent direction with the angle of sunlight, the details of the vehicle will be completely lost due to strong light exposure, affecting the normal operation of the monitoring system.

[0106] Implementation process:

[0107] Dataset preparation: We collected road traffic backlight perception video data under various backlight conditions. The dataset contains a variety of weather conditions, light intensities, and time periods to ensure the diversity of training data.

[0108] Image preprocessing and enhancement: Adaptive contrast enhancement and local histogram equalization methods are used to optimize the brightness and contrast of vehicle images under backlight, especially in strong sunlight, to enhance the outline and texture of the vehicle.

[0109] Deep learning model training and detection: Use the Faster R-CNN model for vehicle detection. During the training process, add backlit backgrounds and images under different lighting conditions so that the network can learn the characteristics of vehicles in backlit environments.

[0110] Results: In real monitoring scenarios, the accuracy of the vehicle detection system under backlight conditions is improved by more than 20%, especially in low-contrast and small target conditions, the system can still maintain a high accuracy rate.

[0111] This implementation case demonstrates the ability of the present invention to accurately identify vehicles in backlit environments through a deep learning model. By combining image enhancement with deep learning, the system can effectively identify low-contrast, blurred targets in backlit environments, thereby improving the stability and accuracy of the security monitoring system.

[0112] Case 3: Drone Target Identification

[0113] When drones are conducting aerial inspections or monitoring, they often encounter backlighting conditions, especially when photographing distant targets. Due to changes in the sun's position, the outline of the target object is blurred or overexposed, resulting in target recognition failure.

[0114] Implementation process:

[0115] Dataset preparation: We selected an aerial image dataset taken by a backlight camera mounted on a drone, which includes target objects under various backlight conditions. The dataset contains images from different time periods and different environments.

[0116] Image preprocessing and enhancement: Perform backlight enhancement on the image, use adaptive brightness adjustment and shadow removal technology to improve the visibility of the target area. Use image detail recovery methods for low-contrast areas to make the target appear clearer in the image.

[0117] Deep learning model training and detection: The improved YOLOv4 model is used for target detection, especially in backlit scenes with poor image quality. The model enhances the recognition ability of target areas through multi-layer feature extraction and context information fusion.

[0118] Results: During the aerial inspection, the system successfully identified multiple low-contrast targets, especially under backlight conditions, and the system's recognition accuracy was improved by 17%.

[0119] This implementation case demonstrates the high efficiency of the deep learning-based target detection method in drone backlighting photography. By optimizing the backlighting environment, this method can improve the visibility of the target and enhance the detection accuracy, greatly improving the target recognition capability in drone mission execution.

Claims

1. A flight simulation method based on unmanned aerial vehicle backlighting imaging technology, characterized in that: The following steps are involved: Step 1: construct a UAV platform, which includes a backlight imaging camera, a stabilized gimbal, a real-time data transmission module and an image processor mounted on the UAV. The backlight imaging camera is installed on the stabilized gimbal, and the backlight imaging camera transmits the perceived image data of the UAV during flight to the image processor through the real-time data transmission module; Step 2: Fix the backlight imaging camera on a stabilized gimbal to ensure that its field of view is consistent with the flight direction of the drone and has the ability to adjust in real time; Step 3: Use the multi-point calibration method to accurately calibrate the internal and external parameters of the backlight imaging camera to ensure that lens distortion and geometric errors are eliminated; Step 4: The backlight imaging camera is equipped with a dynamic aperture control mechanism to achieve automatic exposure adjustment under drastic changes in lighting conditions and optimize the imaging quality of the target area; Step 5: Introduce a multi-scale noise reduction algorithm based on wavelet transform to suppress random noise interference caused by high dynamic range under backlight conditions; Step 6: Build a lightweight generative adversarial network model, input the backlit image and reconstruct the target texture and structural features through multi-layer feature extraction, avoiding artifacts or color drift problems caused by traditional methods; Step 7: Based on the enhanced and optimized YOLO detection framework, combined with the prior backlight scene data, a real-time target detection model is constructed to extract the bounding boxes of multi-category targets and their classification probabilities; a high-resolution feature pyramid network is used for small target areas to ensure the accuracy and robustness of target detection; a multi-feature fusion method is used to further extract target key points and their spatial geometric characteristics, providing a standardized high-dimensional data description; Step 8: Overlay the UAV flight status parameters and light distribution parameters in real time to complete the flight simulation display of the UAV-mounted backlight imaging technology.

2. The flight simulation method based on unmanned aerial vehicle backlighting imaging technology according to claim 1, characterized in that: In step one, the stabilized gimbal adopts a three-axis mechanical stabilized gimbal, the real-time data transmission module is a high-bandwidth real-time data transmission module, and the backlight imaging camera, the three-axis mechanical stabilized gimbal and the high-bandwidth real-time data transmission module form a complete UAV perception and transmission system framework.

3. A flight simulation method based on unmanned aerial vehicle backlighting imaging technology according to claim 1 or 2, characterized in that: In step two, the stabilized gimbal uses high-precision servo control technology and utilizes the servo system to achieve precise control of the motion state of the mechanical equipment, especially precise operation in terms of position, speed and acceleration, to ensure that the camera's line of sight is stable and can be precisely adjusted under complex flight conditions.

4. The flight simulation method based on unmanned aerial vehicle backlighting imaging technology according to claim 3 is characterized in that: Step three is as follows: The multi-point calibration method is used to accurately calibrate the internal and external parameters of the backlight camera. Then the calibration process is used to eliminate the errors of lens distortion, geometric offset and optical axis inconsistency to ensure the geometric accuracy of the acquired image. The internal parameters include focal length, principal point position, and distortion coefficient; the external parameters include camera posture and position.

5. The flight simulation method based on unmanned aerial vehicle backlighting imaging technology according to claim 4 is characterized in that: Step 4 is as follows: Deploying a dynamic aperture control module based on adaptive light perception can quickly respond to light changes in backlit environments to achieve a linkage adjustment mechanism for aperture, shutter speed, and ISO parameters, ensuring that the imaging quality of the target area in both bright and dark areas is optimal.

6. The flight simulation method based on unmanned aerial vehicle backlighting imaging technology according to claim 5 is characterized in that: Step 5 is as follows: A multi-scale denoising algorithm based on wavelet transform is introduced to perform noise suppression on backlit imaging data. In the denoising process, the high-frequency and low-frequency components of the image are separated to eliminate random noise interference while retaining the detailed features of the target area.

7. The flight simulation method based on unmanned aerial vehicle backlighting imaging technology according to claim 6 is characterized in that: Step 6 is as follows: A lightweight generative adversarial network model is designed. The original image data under backlighting conditions is input, and the texture details and structural features of the target are reconstructed through multi-layer feature extraction and illumination compensation. The adversarial loss and perceptual loss are then used to optimize the generative network to avoid the artifacts and color drift problems of traditional algorithms in backlighting enhancement.

8. The flight simulation method based on unmanned aerial vehicle backlighting imaging technology according to claim 7 is characterized in that: The specific process of the flight simulation demonstration of the UAV-mounted backlight imaging technology in step 8 includes: The backlight imaging camera carried by the drone collects environmental image data during the flight, including target objects, background scenes and light distribution information. The collected data needs to be optimized by dynamic aperture and backlight perception imaging technology to ensure clear imaging under conditions of drastic changes in light. In the backlight scene, the flight status parameters of the drone are superimposed on the picture to dynamically present the flight status. Then, intuitive superimposed annotations are used to help analyze the target perception ability under backlight conditions. The superimposed annotations include bounding boxes, classification labels and key points.

9. The flight simulation method based on the unmanned aerial vehicle backlighting imaging technology according to claim 1, 2, 4, 5, 6, 7 or 8, wherein the specific process is: Step 1: Build a UAV platform equipped with a backlight imaging system; The backlight imaging camera is installed on a three-axis stabilized gimbal, which realizes attitude adjustment through linkage with the UAV flight control system to ensure that the field of view of the backlight imaging camera is consistent with the target area; Step 2: Fix the backlight imaging camera on the gimbal; The control of the stabilized gimbal uses a three-dimensional rotation matrix, and the calculation formula is as follows: R=R z (ψ)R y (i)R x (f) (1) in, R z (ψ),R y (θ),R x (φ) represents the rotation matrix around the x, y, and z axes respectively; The gimbal adjusts the camera viewing angle in real time through feedback closed-loop control. The formula is: e angle =θ desired -θ current (2) The controller is based on the angle error e angle Output corresponding driving signal; Step 4: Configure a dynamic aperture control mechanism to achieve automatic exposure adjustment under drastic changes in lighting conditions and optimize the imaging quality of the target area; The dynamic aperture module adjusts the aperture size in real time according to the ambient light conditions and adopts an exposure value optimization strategy. The aperture adjustment formula is: Where A is the aperture diameter, ΔL is the illumination difference, k is the adjustment coefficient, and t is the exposure time. Through the dynamic combination of aperture adjustment and exposure time, the system can accurately image the target area under strong backlight. Step 5: Introduce a multi-scale noise reduction algorithm based on wavelet transform to suppress random noise interference caused by high dynamic range under backlight conditions; The backlight imaging camera is generated by encoding and restoring the mask of the target in the strong light background. Taking into account the system defocus τ, the specific expression of the exit pupil function is given: P(x,y,τ)=p(x,y)exp{i[f(x,y)+τ(x 2 +y 2 )]} (4) Among them, i 2 = -1, τ represents the phase distribution of the defocus aberration on the pupil, D is the diameter of the system pupil, and f(x, y) represents the phase mask function of the optical system when the phase plate is inserted in the defocus condition; After knowing the exit pupil function of the system, the point spread function used to characterize the imaging characteristics of the system in the spatial domain is given: h(x′,y′,τ)=κ{F[P(x,y,τ)]F * [P(x,y,τ)]} (6) Where, F represents Fourier transform, κ is the normalization constant, and * represents conjugation; After obtaining the point spread function of optical imaging, Fourier transform can be performed to obtain the optical transfer function that characterizes the spatial frequency transfer characteristics of the system: H(u,v,τ)=F[h(x′,y′,τ)] (7) The backlight imaging camera can calculate the optical transfer function in advance through the phase plate type, and the physical system can collect the point spread function of the point light source imaging system. By performing a Fourier transform on the point spread function, the optical transfer function of the system can be obtained. Wiener filtering is called minimum mean square error filtering. This method regards the image and noise as random processes, and the ultimate goal is to obtain a statistical error function: The minimum estimate of Where f is the unprocessed original image information, is the minimum estimate of the original image, and E is the expectation of the difference between the original image and the minimum estimate, expressed as: Among them, |H(u,v) 2 =H * (u,v)H(u,v),H * (u,v) represents the complex conjugate of H(u,v), S n (u,v)=|N(u,v) 2 represents the noise power spectrum of image processing; S f (u,v)=|F(u,v) 2 Represents the power spectrum of the original image; Formula (8) can be simplified as: In order to suppress random noise in high dynamic range images, multi-scale wavelet transform is used for denoising; the input image I(x, y) is decomposed into low-frequency component L and high-frequency component H i , the specific formula is: Noise reduction is done by soft thresholding, the formula is: Among them, H i ′ is the high-frequency component after noise reduction, λ is the noise threshold, which is automatically estimated through noise statistics; Step 6: Design a lightweight generative adversarial network model, input the backlit image and reconstruct the target texture and structural features through multi-layer feature extraction, avoiding artifacts or color drift problems caused by traditional methods; Step 7: Based on the enhanced and optimized YOLOv8 detection framework and combined with prior backlight scene data, a real-time target detection model is constructed to extract the bounding boxes of multi-category targets and their classification probabilities; The target detection system uses the enhanced and optimized YOLOv8 framework and inputs the backlit image I input The detection network extracts multi-scale features and outputs the bounding box and category probability of the target through the fully connected layer. The loss function of the bounding box is L box for: in, is the predicted bounding box, is the true bounding box, and the classification loss uses the cross entropy function, the formula is: The complete detection loss is: L total =L box +L cls (14) On the basis of detecting the target, the key point information is extracted, and the key point is determined by the response function R: Among them, Ω(k i ) is the pixel area around the key point, and F(p) is the eigenvalue; multiple eigenvalues ​​are fused through the feature pyramid network to integrate high-resolution features and low-resolution global features to generate the final target description.

10. A flight simulation system based on unmanned aerial vehicle backlight imaging technology, characterized in that: The system has a program module corresponding to the steps of any one of claims 1 to 9 above, and executes the steps of the flight simulation method based on the unmanned aerial vehicle backlighting imaging technology when running, which includes: The backlight imaging camera system is used to obtain environmental image data under backlight conditions; its configured dynamic aperture control mechanism optimizes the imaging quality of the target area through automatic exposure adjustment; and inputs the high-quality environmental image data obtained under backlight conditions into the data processing and image enhancement system in the image processor; Stabilization control system, used to ensure the stability of imaging equipment during flight, the ability to adjust the field of view, and the quality of image acquisition; control the servo system of the stabilized gimbal to accurately control the motion state of the mechanical equipment, especially the precise operation in terms of position, speed, and acceleration; The image processor is used to read high-quality environmental image data under backlight conditions and provide basic processing for target perception and recognition. It is equipped with a data processing and image enhancement system and a target detection and feature extraction system. Data processing and image enhancement system, which includes wavelet transform multi-scale noise reduction algorithm, lightweight generative adversarial network and multi-scale image enhancement algorithm, is used to reduce noise, enhance and reconstruct target features of raw image data under backlight conditions, and improve perception and recognition performance; The target detection and feature extraction system uses the YOLO target detection framework to detect multi-category targets in backlit scenes in real time, extract key features and generate standardized descriptions to provide input for flight decisions.

Images