Lightweight AI distribution network unmanned aerial vehicle inspection image automatic acquisition method

Abstract

The invention discloses a distribution network unmanned aerial vehicle inspection image automatic acquisition method based on lightweight AI, and the method comprises the steps: 1, enabling an unmanned aerial vehicle to start to operate, and obtaining an inspection image when the unmanned aerial vehicle arrives at an inspection shooting point; 2, positioning target power grid equipment in the inspection image based on a target detection and recognition algorithm, and outputting characteristic parameters of the target power grid equipment; 3, calculating the position deviation between the target equipment and the center of the inspection image according to the characteristic parameters of the target power grid equipment, and adjusting the attitude of the holder and the position of the unmanned aerial vehicle according to the position deviation to enable the target equipment to be located in the center of the inspection image; and step 4, obtaining illumination features through an ambient light sensor, adjusting camera parameter combination through an automatic dimming algorithm based on the ambient illumination features, and then executing a photographing action to obtain a target power grid equipment image. According to the invention, distribution network inspection key component image acquisition can be completed without manual intervention, and the inspection image one-time shooting qualified rate and operation efficiency are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection technology for power distribution networks, and in particular to a method for automatic image acquisition by lightweight AI-based drones for power distribution network inspection. Background Technology

[0002] With the continuous expansion of power distribution network scale, the inspection of power distribution lines, as a key link in ensuring power supply reliability, is facing increasingly higher work intensity and quality requirements. Traditional power distribution network inspection mainly relies on manual pole climbing or ground visual inspection, which is not only inefficient, but also poses safety risks such as falls from heights and electric shocks in complex terrain environments such as mountainous areas and lake and riverbanks. At the same time, the workload of grassroots inspection personnel is heavy and can hardly meet the operation and maintenance needs of large-scale power distribution networks.

[0003] To address the pain points of traditional manual inspections, drone inspection technology is gradually being promoted and applied in power distribution network inspections. However, its existing technical solutions still have significant shortcomings, resulting in unstable inspection image quality and insufficient release of data value, as specifically manifested as follows: Currently, the target alignment process for drone inspections relies on manual operation by the pilot, who must simultaneously manage the drone's flight and aim at critical components of the control tower. Factors such as pilot experience, fatigue, and hand tremors can easily cause target components to deviate from the image center, tilt, or even miss critical inspection areas, making subsequent image review impossible due to substandard target positioning.

[0004] Power distribution network inspection scenarios are complex, often facing varying lighting conditions such as backlighting, strong light, and cloudy days. However, the camera parameter adjustments of existing drones mainly rely on manual operation by the pilot. It is difficult for pilots to accurately judge changes in ambient light in real time and match the optimal parameters, resulting in problems such as overexposed highlights or completely black shadows in the acquired images. The detailed features of key components are lost, which cannot meet the needs of defect identification or manual image review.

[0005] Current drone inspections mostly use fixed-focus shooting or manual zoom mode, which makes it difficult to cover both distant views and close-up details. Manual zoom operation is cumbersome and prone to blurring due to data transmission delays and pilot errors. At the same time, there is a lack of automatic judgment and correction mechanism for image blur. Once focus failure occurs, it is necessary to fly again to take a new picture, which greatly reduces inspection efficiency.

[0006] Currently, the target identification, alignment, parameter adjustment, focusing, and photography stages of drone inspections are independent and all require manual intervention and coordination, resulting in a "human-powered" operation mode. This not only leads to a fragmented inspection process and long inspection times for single towers, but also results in a low first-pass image quality rate, requiring frontline personnel to frequently re-fly the drones, leading to low data utilization. The core need to reduce the burden on grassroots staff has not been effectively met, thus hindering the digital and intelligent transformation of power distribution network inspections. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of existing power distribution network drone inspection technology, which lacks an automated and collaborative image acquisition mechanism and requires manual intervention in target positioning, parameter adjustment, and focusing. This results in problems such as inaccurate target positioning, unbalanced lighting, and blurry focus, leading to low inspection efficiency and low first-shot pass rate. This invention provides a lightweight AI-based automatic image acquisition method for power distribution network drone inspection. By sequentially executing target power grid equipment positioning, automatic correction, automatic dimming, and automatic zooming, the method cyclically completes image acquisition for all shooting points, achieving automated image acquisition for power distribution network inspection, improving the first-shot pass rate and inspection efficiency, and reducing the burden on grassroots units.

[0008] The objective of this invention is achieved through the following technical solution: A method for automatic image acquisition for power distribution network inspection using lightweight AI-based drones includes the following steps: Step 1: The drone starts operating and arrives at the inspection and shooting point to acquire inspection images; Step 2: Locate the target power grid equipment in the inspection image based on the target detection and recognition algorithm, and output the feature parameters of the target power grid equipment; Step 3: Calculate the positional deviation between the target equipment and the center of the inspection image based on the characteristic parameters of the target power grid equipment, and adjust the gimbal attitude and the UAV position according to the positional deviation so that the target equipment is located at the center of the inspection image. Step 4: Obtain illumination characteristics through an ambient light sensor, adjust camera parameter combinations based on ambient light characteristics using an automatic dimming algorithm, and then perform a photo-taking action to obtain an image of the target power grid equipment; Step 5: The drone arrives at the next inspection and shooting point and repeats steps 1 to 4 until images of the target power grid equipment at all inspection and shooting points are acquired, thus completing the automatic acquisition process of distribution network drone inspection images.

[0009] Preferably, step 2 involves locating the target power grid equipment in the inspection image based on a target detection and recognition algorithm, and outputting the feature parameters of the target power grid equipment. Specifically: The target detection and recognition algorithm is YOLOv5 or YOLOv8. It detects the feature information of the inspection image and matches it with the preset power grid equipment features, and outputs the position coordinates, size ratio and attitude angle data of the target power grid equipment.

[0010] Preferably, step 3 involves calculating the positional deviation between the target equipment and the center of the inspection image based on the characteristic parameters of the target power grid equipment, and adjusting the gimbal attitude and UAV position according to the positional deviation to position the target equipment at the center of the inspection image. Specifically: Based on the characteristic parameters of the target power grid equipment, the target power grid equipment in the continuous inspection image sequence is continuously tracked; by comparing the position coordinates of the target power grid equipment with the center coordinates of the inspection image, the positional deviations of the two in the horizontal and vertical directions are calculated, and the horizontal offset status is determined by combining the attitude angle of the target power grid equipment; the gimbal is driven to rotate using a control algorithm, and the relative position of the UAV is adjusted so that the target power grid equipment is both located in the center of the inspection image and maintains a horizontal attitude, and the positional deviation and horizontal offset do not exceed the preset threshold.

[0011] Preferably, the control algorithm is an event-triggered loop control algorithm based on the gimbal attitude position error threshold. The error threshold includes: horizontal position deviation less than 5 pixels, vertical position deviation less than 5 pixels, and horizontal offset angle less than 0.5. When the deviation or offset exceeds the error threshold, the control algorithm triggers the gimbal and the UAV to adjust.

[0012] Preferably, in step 4, the ambient light sensor is installed on the drone. The lighting characteristics include the average brightness, contrast, proportion of highlights, proportion of shadows, and backlight or frontlight scene indicators of the target power grid equipment area. The above lighting characteristics are input into a preset automatic dimming algorithm. The automatic dimming algorithm evaluates and determines the lighting type of the current drone scene through multi-dimensional evaluation, and dynamically outputs an appropriate combination of camera parameters according to the lighting type.

[0013] Preferably, the lighting type includes normal, backlight, or strong light.

[0014] Preferably, the camera parameter combination includes camera ISO, shutter speed, and exposure compensation. The ISO is controlled within a low noise threshold range, the shutter speed matches the flight stability requirements of the drone, and the exposure compensation is fine-tuned for the target area.

[0015] Preferably, in step 4, before taking a picture to acquire an image of the target power grid equipment, the camera is automatically zoomed, specifically as follows: The Laplacian operator is used to calculate the blur of the inspection image. If the blur exceeds the set threshold, the image is determined to be blurry and the camera is refocused. If the blur still exceeds the set threshold after refocusing, the camera is restarted and refocused. If the blur still exceeds the set threshold, the lens is determined to be faulty and relevant personnel are alerted.

[0016] The beneficial effects of this invention are as follows: By constructing a fully automated image acquisition mechanism that includes target positioning, automatic correction, automatic dimming, automatic zooming, and cyclic acquisition, this invention can complete the acquisition of images of key components in power distribution network inspection without manual intervention. This completely solves the problems of inaccurate target positioning, unbalanced lighting, and blurry focus caused by the reliance on drone operators in existing technologies, and significantly improves the first-time pass rate of inspection images and operational efficiency.

[0017] This invention automatically corrects deviations to ensure the target is centered, automatically adjusts brightness to adapt to complex lighting conditions, and automatically zooms to precisely focus on details. These three features work together to output standardized images with good composition, clear details, and suitable lighting. These images can be directly integrated into defect identification models or used for rapid manual review, significantly reducing the proportion of invalid data and fully releasing the application value of inspection data.

[0018] The solution of this invention is suitable for flight altitudes of 30m-80m and various power distribution environments such as urban, rural and mountainous areas. It has good robustness to working conditions such as strong winds, backlight and complex terrain. At the same time, it greatly reduces the frequency of manual pole climbing, avoids safety risks such as high-altitude operations and electric shock, and takes into account both full coverage of inspection and operational safety. Attached Figure Description

[0019] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0020] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.

[0021] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0022] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0023] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0024] Example: A method for automatic image acquisition for power distribution network inspection using lightweight AI-based drones, such as Figure 1 As shown, it includes the following steps: Step 1: The drone starts operating and arrives at the inspection and shooting point to acquire inspection images; Step 2: Locate the target power grid equipment in the inspection image based on the target detection and recognition algorithm, and output the feature parameters of the target power grid equipment; Step 3: Calculate the positional deviation between the target equipment and the center of the inspection image based on the characteristic parameters of the target power grid equipment, and adjust the gimbal attitude and the UAV position according to the positional deviation so that the target equipment is located at the center of the inspection image. Step 4: Obtain illumination characteristics through an ambient light sensor, adjust camera parameter combinations based on ambient light characteristics using an automatic dimming algorithm, and then perform a photo-taking action to obtain an image of the target power grid equipment; Step 5: The drone arrives at the next inspection and shooting point and repeats steps 1 to 4 until images of the target power grid equipment at all inspection and shooting points are acquired, thus completing the automatic acquisition process of distribution network drone inspection images.

[0025] Step 2, which involves locating the target power grid equipment in the inspection image based on the target detection and recognition algorithm, and outputting the feature parameters of the target power grid equipment, specifically includes: The target detection and recognition algorithm is YOLOv5 or YOLOv8. It detects the feature information of the inspection image and matches it with the preset power grid equipment features, and outputs the position coordinates, size ratio and attitude angle data of the target power grid equipment.

[0026] YOLO (YouOnlyLookOnce) is a single-stage object detection algorithm widely used in real-time object detection tasks due to its fast detection speed and high accuracy. The core idea of ​​this algorithm is to take the entire image as input and predict the location and category of all objects in the image at once. Specifically, it divides the input image into an S×S grid, with each grid responsible for predicting the presence, category, and location of objects within that region. When the center point of an object falls within a grid, that grid is responsible for detecting that object. YOLO uses bounding boxes to locate objects, represented by four parameters: center coordinates, width, and height. Furthermore, each bounding box predicts a confidence score, representing the probability of an object being within the bounding box and the accuracy of the bounding box.

[0027] The main advantages of the YOLO algorithm are as follows: High speed: YOLO can process large amounts of image data in a short time and detect targets in real time; Strong generalization ability: YOLO is highly adaptable to image data from different domains and can be applied to various target detection tasks; Open source nature: YOLO is an open source system, and its code and model are publicly available online, facilitating research and improvement by researchers and developers.

[0028] Step 3, based on the characteristic parameters of the target power grid equipment, calculates the positional deviation between the target equipment and the center of the inspection image, and adjusts the gimbal attitude and the UAV position according to the positional deviation to place the target equipment at the center of the inspection image. Specifically: Based on the characteristic parameters of the target power grid equipment, the target power grid equipment in the continuous inspection image sequence is continuously tracked; by comparing the position coordinates of the target power grid equipment with the center coordinates of the inspection image, the positional deviations of the two in the horizontal and vertical directions are calculated, and the horizontal offset status is determined by combining the attitude angle of the target power grid equipment; the gimbal is driven to rotate using a control algorithm, and the relative position of the UAV is adjusted so that the target power grid equipment is both located in the center of the inspection image and maintains a horizontal attitude, and the positional deviation and horizontal offset do not exceed the preset threshold.

[0029] During drone inspection operations, manually keeping the target centered in the frame requires the operator to continuously adjust the drone's attitude and gimbal angle. This is cumbersome and prone to causing the target to frequently deviate from the field of view due to hand tremors and work fatigue. Especially in complex environments such as strong winds and sudden changes in lighting, the aiming accuracy drops significantly, seriously affecting inspection efficiency and data quality.

[0030] Automatic target deviation correction technology detects, extracts, and tracks targets in image sequences acquired by UAVs, obtaining data such as the target's position, velocity, acceleration, and trajectory. Then, control algorithms are used to operate the gimbal and UAV, ensuring the detected target remains centered in the image based on changes in the image, thereby guaranteeing the quality of the acquired image data.

[0031] The target tracking algorithm employs a linear matching algorithm. A loss matrix is ​​generated from the first and second frame detections, recording the loss values ​​of the two detections between the two frames. A higher probability of a match between detections results in a lower loss value; conversely, a lower probability of a match results in a higher loss value. The matching is based on the globally optimal result.

[0032] Currently, common target tracking techniques mainly include template matching-based methods (NCC matching, SSD matching, SAD matching, etc.), feature point-based methods (KLT tracking, SIFT tracking, SURF tracking, etc.), and deep learning-based methods (Siamese network, MDNet, SiamFC, etc.).

[0033] The control algorithm is an event-triggered loop control algorithm based on the gimbal attitude position error threshold. The error threshold includes: horizontal position deviation less than 5 pixels, vertical position deviation less than 5 pixels, and horizontal offset angle less than 0.5. When the deviation or offset exceeds the error threshold, the control algorithm triggers the gimbal and the UAV to adjust.

[0034] In step 4, an ambient light sensor is mounted on the drone. The lighting characteristics include the average brightness, contrast ratio, highlight pixel ratio, shadow pixel ratio, and backlight or front lighting scene indicators for the target power grid equipment area. These lighting characteristics are input into a preset automatic dimming algorithm. The algorithm evaluates and determines the lighting type of the current drone scene from multiple dimensions and dynamically outputs an appropriate combination of camera parameters based on the lighting type. The lighting type includes normal, backlight, or strong light.

[0035] The camera parameter combination includes camera ISO, shutter speed, and exposure compensation. The ISO is controlled within a low noise threshold range, the shutter speed is matched to the flight stability requirements of the drone, and the exposure compensation value is fine-tuned for the target area.

[0036] AI-powered automatic dimming technology is a visual recognition-based intelligent adjustment solution for drone-mounted cameras. This technology uses the drone's onboard vision system to capture real-time environmental images, employs AI algorithms to analyze and calculate the images, and dynamically adjusts the drone's camera parameters based on the analysis results to obtain clear images.

[0037] Image acquisition quality is significantly affected by lighting conditions, especially in complex lighting conditions such as backlighting, strong light, or cloudy days, which can easily lead to overexposure, underexposure, or contrast imbalance, resulting in the loss of critical details. In actual drone inspection operations, this will greatly increase the workload of personnel and expose the operation to more risks.

[0038] AI automatic dimming algorithm can combine multi-dimensional lighting characteristics for real-time comprehensive evaluation, and dynamically adjust camera parameters such as ISO, shutter speed and exposure compensation according to scene type. While ensuring moderate brightness of the image, it suppresses noise generation and improves the detail of dark and bright areas.

[0039] In step 4, before taking a picture to acquire an image of the target power grid equipment, the camera is automatically zoomed, specifically as follows: The Laplacian operator is used to calculate the blur of the inspection image. If the blur exceeds the set threshold, the image is determined to be blurry and the camera is refocused. If the blur still exceeds the set threshold after refocusing, the camera is restarted and refocused. If the blur still exceeds the set threshold, the lens is determined to be faulty and relevant personnel are alerted.

[0040] Traditional drone inspection operations often rely on fixed-focus photography, making it difficult to capture both distant views and details. When zooming is required, the process becomes cumbersome and inaccurate. Operators must frequently adjust the focus manually, which can lead to blurred targets due to operator fatigue or data delays. This is especially problematic in long-distance or fast-moving scenarios, where it is difficult to capture clear details in real time, impacting inspection efficiency and data quality.

[0041] Automatic zoom technology first calculates the blur level of the captured image. If the image is determined to be blurry, it refocuses. If the image is still blurry after refocusing, the gimbal camera is restarted and refocused again. If the image is still blurry, the lens is considered faulty and needs repair. This method pre-judges the blur level of the image before the camera takes a picture, ensuring the sharpness of the captured photo. The proposed research adopts an event-triggered cyclic focusing technology based on the Laplacian operator image blur discrimination threshold to ensure a high success rate of focusing, and to refocus and retake blurry photos.

[0042] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0043] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for automatic image acquisition during power distribution network drone inspections based on lightweight AI, characterized by: Includes the following steps: Step 1: The drone starts operating and arrives at the inspection and shooting point to acquire inspection images; Step 2: Locate the target power grid equipment in the inspection image based on the target detection and recognition algorithm, and output the feature parameters of the target power grid equipment; Step 3: Calculate the positional deviation between the target equipment and the center of the inspection image based on the characteristic parameters of the target power grid equipment, and adjust the gimbal attitude and the UAV position according to the positional deviation so that the target equipment is located at the center of the inspection image. Step 4: Obtain illumination characteristics through an ambient light sensor, adjust camera parameter combinations based on ambient light characteristics using an automatic dimming algorithm, and then perform a photo-taking action to obtain an image of the target power grid equipment; Step 5: The drone arrives at the next inspection and shooting point and repeats steps 1 to 4 until images of the target power grid equipment at all inspection and shooting points are acquired, thus completing the automatic acquisition process of distribution network drone inspection images.

2. The method for automatic image acquisition of power distribution network inspection by UAV based on lightweight AI according to claim 1, characterized in that, Step 2, which involves locating the target power grid equipment in the inspection image based on the target detection and recognition algorithm, and outputting the feature parameters of the target power grid equipment, specifically includes: The target detection and recognition algorithm is YOLOv5 or YOLOv8. It detects the feature information of the inspection image and matches it with the preset power grid equipment features, and outputs the position coordinates, size ratio and attitude angle data of the target power grid equipment.

3. The method for automatic image acquisition of power distribution network drone inspection based on lightweight AI according to claim 2, characterized in that, Step 3, based on the characteristic parameters of the target power grid equipment, calculates the positional deviation between the target equipment and the center of the inspection image, and adjusts the gimbal attitude and the UAV position according to the positional deviation to place the target equipment at the center of the inspection image. Specifically: Based on the characteristic parameters of the target power grid equipment, the target power grid equipment in the continuous inspection image sequence is continuously tracked. By comparing the position coordinates of the target power grid equipment with the center coordinates of the inspection image, the positional deviations of the two in the horizontal and vertical directions are calculated. At the same time, the horizontal offset status is determined by combining the attitude angle of the target power grid equipment. The control algorithm drives the gimbal to rotate and adjusts the relative position of the UAV so that the target power grid equipment is both located in the center of the inspection image and maintains a horizontal attitude, and the positional deviation and horizontal offset do not exceed the preset threshold.

4. The method for automatic image acquisition of power distribution network drone inspection based on lightweight AI according to claim 3, characterized in that, The control algorithm is an event-triggered loop control algorithm based on the gimbal attitude position error threshold. The error threshold includes: horizontal position deviation less than 5 pixels, vertical position deviation less than 5 pixels, and horizontal offset angle less than 0.

5. When the deviation or offset exceeds the error threshold, the control algorithm triggers the gimbal and the UAV to adjust.

5. The method for automatic image acquisition of power distribution network drone inspection based on lightweight AI according to claim 1, characterized in that, In step 4, an ambient light sensor is installed on the drone. The lighting characteristics include the average brightness, contrast, percentage of highlights, percentage of shadows, and backlight or frontlight scene indicators of the target power grid equipment area. The above lighting characteristics are input into a preset automatic dimming algorithm. The automatic dimming algorithm evaluates and determines the lighting type of the current drone scene through multi-dimensional evaluation and dynamically outputs an appropriate combination of camera parameters according to the lighting type.

6. The method for automatic image acquisition of power distribution network inspection by UAV based on lightweight AI according to claim 5, characterized in that, The lighting type includes normal, backlight, or strong light.

7. The method for automatic image acquisition of power distribution network drone inspection based on lightweight AI according to claim 5, characterized in that, The camera parameter combination includes camera ISO, shutter speed, and exposure compensation. The ISO is controlled within a low noise threshold range, the shutter speed is matched to the flight stability requirements of the drone, and the exposure compensation is fine-tuned for the target area.

8. The method for automatic image acquisition of power distribution network drone inspection based on lightweight AI according to claim 1, characterized in that, In step 4, before taking a picture to acquire an image of the target power grid equipment, the camera is automatically zoomed, specifically as follows: The Laplace algorithm is used to calculate the blur of the inspection image. If the blur exceeds the set threshold, the image is determined to be blurry, and the camera is refocused. If the blur still exceeds the set threshold after refocusing, the camera is restarted and refocused. If the blur still exceeds the set threshold, the lens is determined to be faulty, and relevant personnel are alerted.

Images