Power grid picture layered labeling method and dynamic labeling display system
Through the method of layered labeling and gradient switching, the problem that the human eye has difficulty concentrating and performing detailed analysis in traditional power grid image labeling methods is solved, and more efficient and accurate power grid image labeling and detection are achieved.
Patent Information
- Application Number
- CN202510649202.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-20
- Publication Date
- 2025-09-19
AI Technical Summary
Traditional power grid image annotation methods treat power grid images as a single entity for annotation, resulting in mixed annotation results that are difficult to support refined analysis. In addition, the human eye has difficulty adapting to the complex hierarchical power grid structure and the dynamic visual feature requirements when the image changes.
A layered labeling method is adopted to annotate power grid images in layers using the environment layer, tower layer, equipment layer and component defect layer models. Gradual switching and zooming are performed through the user interaction interface. The eye tracking device and color coding module are combined to guide human observation and achieve an active guided switching effect.
It reduces the pressure on human eyes, improves concentration, enhances the smoothness of image switching and the accuracy of annotation, improves user experience and annotation efficiency, especially the detection and review effect of defect layer annotation results.
Smart Images
Figure CN120673409A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of image processing, and in particular to a layered annotation method for power grid images and a dynamic annotation display system. Background Art
[0002] Traditional power grid image annotation methods rely on manual experience or a single algorithm model, and have the following problems: Existing technologies usually label power grid images as a single entity, resulting in mixed labeling results and difficulty supporting refined analysis (such as defect detection). Due to the complex hierarchy of the power grid structure and the dynamic demand for visual features when the image changes, traditional static labeling methods will display the labeled points of all structures on the screen observed by the human eye, making it difficult for the human eye to adapt to the complex hierarchy of the power grid structure and increasing the demand for dynamic vision of the human eye when the image changes, making it difficult for the human eye to concentrate and perform refined analysis during manual labeling and detection.
[0003] Currently, no effective solution has been proposed to the problem that traditional power grid image annotation methods make it difficult to guide the human eye to focus and perform detailed analysis during manual annotation detection. Summary of the Invention
[0004] The present invention provides a layered annotation method for power grid images and a dynamic annotation display system to solve the problem that traditional power grid image annotation methods make it difficult to guide the human eye to focus on fine analysis during manual annotation detection.
[0005] In a first aspect, the present invention provides a hierarchical annotation method for power grid images, which includes: receiving a target power grid image; annotating the target power grid image using an environment layer model to obtain a first image with a first annotation, wherein the environment layer model is used to identify the terrain, vegetation and weather in the image and generate a corresponding first annotation; annotating the target power grid image using a tower layer model to obtain a second image with a second annotation, wherein the tower layer model is used to identify the characteristics of the tower in the image and generate a corresponding second annotation; annotating the target power grid image using a device layer model to obtain a third image with a third annotation, wherein the device layer model is used to identify the characteristics of the power equipment in the image and generate a corresponding a third annotation; annotating the target power grid image using a device layer model to obtain a fourth image with a fourth annotation, wherein the device layer model is used to identify the characteristics of the power components in the image and generate a corresponding fourth annotation; annotating the target power grid image using a component defect layer model to obtain a fifth image with a fifth annotation, wherein the component defect layer model is used to identify the characteristics of the power component defects in the image and generate a corresponding fifth annotation; gradually switching and enlarging the first image, the second image, the third image, the fourth image and the fifth image in time sharing through a user interaction interface; obtaining user annotation feedback on the first image, the second image, the third image, the fourth image and the fifth image through the user interaction interface.
[0006] Furthermore, the time-sharing gradual switching is as follows: during the image switching process, the transparency of the previous image decreases to 0% and the marked features gradually shrink, while the transparency of the next image gradually increases to 100% and the marked features gradually enlarge.
[0007] Furthermore, when the first picture, the second picture, the third picture, the fourth picture and the fifth picture are displayed in sequence by time-sharing switching, each layer of pictures flashes at an interval of 0.2 seconds.
[0008] Furthermore, the environment layer model is obtained by deep learning of the ResNet-50 model using the environment standard database, the tower layer model is obtained by deep learning of the YOLOv7 model using the tower standard database, the equipment layer model is obtained by deep learning of the Mask R-CNN model using the equipment standard database, and the component defect layer model is obtained by transfer learning of the standard defect library of cracks and rust.
[0009] Furthermore, the power grid image layered annotation method further includes: interactively correcting or not correcting the annotations of the first image, the second image, the third image, the fourth image, and the fifth image according to the annotation feedback, and obtaining the annotation results.
[0010] Furthermore, the annotation results can be interactively corrected, including: detecting the gaze point of the human eye through an eye tracking device, automatically extending the display time of the current view or triggering the annotation correction interface; manually switching layers, zooming in / out views, adding / deleting annotations through shortcut keys or mouse interaction; and displaying the results of annotation correction in real time on the annotation correction interface.
[0011] Furthermore, the display order of the first picture, the second picture, the third picture, the fourth picture and the fifth picture can be adjusted through the interactive interface or a certain layer can be paused to focus on key information.
[0012] Furthermore, it also includes image preprocessing of the power grid image to obtain the target power grid image, which includes: after receiving the target power grid image, using Gaussian filtering, median filtering and other methods to remove noise in the image; using contrast stretching and histogram equalization technology to enhance the details and contrast of the image.
[0013] Furthermore, it also includes outputting the annotation results in a standard format. Outputting the annotation results in a standard format includes: converting the annotation results into a common image format or vector graphic format; exporting the common image format or vector graphic format into structured data; using the structured data to generate an annotation report, the annotation report includes the annotation results, annotation time and annotation personnel.
[0014] Second, the present invention provides a dynamic annotation display system for executing the aforementioned layered annotation method for power grid images. The dynamic annotation display system includes a server, an eye-tracking device, a human-computer interaction device, and a display device. The server is equipped with an environmental layer model, a tower layer model, an equipment layer model, and a component defect layer model, and is used to annotate images. The eye-tracking device is used to detect the gaze point of the human eye, automatically extend the display time of the current view, or trigger the annotation correction interface. The human-computer interaction device is used to manually switch layers, zoom in / out the view, and add / delete annotations. The display device is equipped with a color coding module, which is used to color-code and dynamically amplify image features, and combined with the eye-tracking device to guide manual review of image annotations.
[0015] Compared with the related art, the present invention has the following beneficial effects: 1. The layered labeling method for power grid images reduces the complexity of the image presented to the human eye through layered display. Then, through the user interface, the layered images are gradually switched, and the labeled features of the next image are actively focused and magnified, achieving an actively guided switching effect. This method reduces the visual strain on the user and improves attention, while actively guiding the eye to observe the features and labeled points of the next image. The gradual switching improves the smoothness of switching between images and avoids image tearing. This solves the problem with traditional power grid image labeling methods, which makes it difficult to guide the eye to focus and conduct detailed analysis during manual labeling inspection.
[0016] 2. The display order of the first picture, second picture, third picture, fourth picture and fifth picture can be adjusted through the interactive interface, or a layer can be paused to focus on key information.
[0017] 3. Outputting annotation results in a standard format facilitates project management and auditing, effectively improving operational efficiency. 4. The dynamic annotation display system relies on display devices and color coding modules to color-code annotation results (such as red highlighting) and dynamically amplify them. This, combined with the persistence of vision effect, guides manual re-inspection, improving manual inspection effectiveness, especially the inspection and review of defect layer annotation results. This effectively increases user experience, improves user experience, and enhances user experience by improving both smoothness and accuracy.
[0018] The details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 This is a partial flow chart of the hierarchical annotation method for power grid images in this embodiment. DETAILED DESCRIPTION
[0020] In order to more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments.
[0021] Unless otherwise defined, technical or scientific terms used in this application shall have the ordinary meanings as understood by persons of ordinary skill in the art to which this application belongs. The terms "a," "an," "the," "these," and similar expressions in this application do not denote limitations on quantity and may be singular or plural. The terms "comprise," "include," "have," and any variations thereof, as used in this application, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device comprising a series of steps or modules (units) is not limited to the listed steps or modules (units) but may include unlisted steps or modules (units) or other steps or modules (units) inherent to the process, method, product, or device. The terms "connected," "connected," "coupled," and similar expressions used in this application are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. As used in this application, "plurality" means two or more. "And / or" describes an association between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can mean: A exists alone; A and B exist simultaneously; or B exists alone. Generally, the character " / " indicates that the objects in the preceding and following relationship are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.
[0022] In this embodiment, a hierarchical annotation method for power grid images is provided to effectively process images with complex power grid structures and dynamic visual features when the images change. Figure 1 The power grid image hierarchical annotation method includes: step S100, step S200, step S300, step S400, step S500, step S600, step S700 and step S800.
[0023] Step S100: Receive a target power grid image. After or before receiving the target power grid image, this embodiment may also perform image preprocessing on the power grid image to obtain the target power grid image. Specifically, after receiving the target power grid image, use methods such as Gaussian filtering and median filtering to remove noise from the image; and use contrast stretching and histogram equalization techniques to enhance image detail and contrast, thereby improving the accuracy of subsequent recognition of the target power grid image. The received target power grid image can be a digital photo, a scanned image, or an image file in any other format. The target power grid image can be uploaded through the user interface (UI) or directly read from the database. Common formats such as JPEG, PNG, and TIFF are supported.
[0024] In step S200, the target power grid image is annotated using an environmental layer model to obtain a first image with first annotations. The environmental layer model is used to identify the terrain, vegetation, and weather in the image and generate corresponding first annotations. In this embodiment, the environmental layer model is derived through deep learning using a ResNet-50 model using an environmental standard database. The environmental standard database contains basic terrain, vegetation, and weather data. The environmental layer model trained with this data can effectively identify the terrain, vegetation, and weather in the target power grid image and generate corresponding first annotations. An image segmentation algorithm (such as GrabCut or FCN) can also be used to extract the background environment. Furthermore, because the environmental layer model only uses annotations for terrain, vegetation, and weather, the annotations of the first image do not interfere with those of other images. This step allows the terrain, vegetation, and weather data of the target power grid image to be independently scanned and annotated.
[0025] Step S300 uses a tower-level model to annotate the target power grid image, generating a second image with second annotations. The tower-level model is used to identify features of the towers in the image and generate corresponding second annotations. Tower features include their location, type, and inclination. The tower-level model is derived through deep learning using the YOLOv7 model using a standard database of towers. Object detection algorithms (such as YOLO and SSD) can also be used to identify the location and shape of the towers. Because the tower-level model only uses annotations of tower features, annotations in the second image do not interfere with those in other images. This step allows the location, type, and inclination of the towers in the target power grid image to be independently scanned and annotated.
[0026] Step S400: The target power grid image is annotated using a device-level model to obtain a third image with third annotations. The device-level model is used to identify the features of the power equipment in the image and generate corresponding third annotations. The features of the power equipment include transformers and insulators. The device-level model is derived through deep learning using a device standard database using the Mask R-CNN model. In this embodiment, the device-level model can be used to detect the features of transformers and insulators. Deep learning models such as Faster R-CNN can also be used to detect power equipment. This step allows the transformers and insulators in the target power grid image to be individually scanned and annotated.
[0027] Step S500: Use the device-level model to annotate the target power grid image, obtaining a fourth image with fourth annotations. The device-level model is used to identify features of power components in the image and generate corresponding fourth annotations. Power component features include those of bolts and wires. In addition, fine-grained classification or instance segmentation algorithms can be used to identify specific power component features. This step allows the transformers and insulators in the target power grid image to be individually scanned and annotated. This step allows the bolts and wires in the target power grid image to be individually scanned and annotated.
[0028] In step S600, the target power grid image is annotated using a component defect layer model, resulting in a fifth image with a fifth annotation. The component defect layer model is used to identify the characteristics of power component defects in the image and generate corresponding fifth annotations. The characteristics of power component defects include those of corrosion and cracks. The component defect layer model is derived by transfer learning from a standard defect library for cracks and corrosion. Image classification or object detection algorithms can also be used to detect component defects (such as cracks and corrosion). This step allows individual component defects (such as cracks and corrosion) in the target power grid image to be scanned and annotated. At the defect layer, a defect-focused annotation method based on transfer learning and dynamic interaction is employed. This method directly trains and uses existing standard defect models, reducing costs. Through a human-computer interaction interface, users can check and modify annotations in real time, improving annotation accuracy and ease of use.
[0029] Through steps S100 to S600, it can be concluded that the applicant obtains the first picture, the second picture, the third picture, the fourth picture and the fifth picture by layering the target power grid picture. These five pictures allow the user to observe and inspect without having to face all the annotations at once, but to observe the annotations of a certain level in a targeted and clear range, thereby reducing the hierarchical complexity of the power grid structure, thereby effectively reducing the pressure on the human eye to observe the power grid structure, which is conducive to improving the human eye's attention and thus facilitating the human eye to perform refined analysis.
[0030] Step S700 involves gradually switching and sequentially magnifying the first, second, third, fourth, and fifth images through a user interaction interface in a time-sharing manner; and obtaining user annotation feedback regarding the first, second, third, fourth, and fifth images through the user interaction interface. The time-sharing gradual switching involves: during the image switching process, the transparency of the previous image decreases to 0% and the annotated features gradually shrink, while the transparency of the subsequent image gradually increases to 100% and the annotated features gradually enlarge. This step, combined with steps S100 to S600, reduces the complexity of the image presented to the human eye through layered display. The layered images are then gradually switched through the user interaction interface, with the annotated features of the next image actively focused and magnified, thereby achieving an actively guided switching effect. This method reduces visual stress and improves the user's attention, while actively guiding the user's eye to observe the features and annotated points of the next image. The gradual switching improves the smoothness of switching between images and avoids image tearing. This solves the problem with traditional power grid image annotation methods, which make it difficult to guide the human eye to focus and conduct detailed analysis during manual annotation detection. According to the applicant's research, this step utilizes the dynamic display of visual persistence, and can simulate the human eye's observation process through time sequence superposition and flicker frequency design, thereby improving the human eye's recognition fluency and focus of image switching. The specific technical implementation is as follows: Timing control: Use a timer or animation framework (such as JavaScript's requestAnimationFrame) to control the display time of each layer. Superimposed display: Dynamic switching and superimposed display of each layer's annotations are achieved through transparency adjustment, layer superposition, and other technologies. Anti-ghosting processing: Insert blank frames or use other anti-ghosting technologies when switching layers to eliminate residual visual interference. In this embodiment, a timer or animation framework is used to control the dynamic switching display time of the first, second, third, fourth, and fifth images. The first, second, third, fourth, and fifth images are displayed in sequence through transparency adjustment and layer superposition. Color coding and image dynamic magnification technology are used to sequentially guide the magnification of the images of the environment, towers, power equipment, power components, and power component defect features in each image.
[0031] In order to improve the smoothness of human eye observation when switching pictures, the applicant conducted experiments and concluded that when the first picture, second picture, third picture, fourth picture and fifth picture are displayed in sequence, the best effect is to flash each layer of pictures at an interval of 0.2 seconds.
[0032] After step S700, the power grid image layered annotation method further includes: interactively correcting or not correcting the annotations of the first image, the second image, the third image, the fourth image, and the fifth image based on the annotation feedback, and obtaining the annotation results. In this embodiment, the interactive correction is used to improve the accuracy of the annotation.
[0033] Specifically, interactive correction of annotation results includes: detecting the gaze point of the human eye through an eye tracking device, automatically extending the display time of the current view or triggering the annotation correction interface; manually switching layers, zooming in / out views, adding / deleting annotations through shortcut keys or mouse interaction; and displaying the results of annotation correction in real time on the annotation correction interface.
[0034] In order to improve the flexibility of device use and user experience, the display order of the first picture, the second picture, the third picture, the fourth picture and the fifth picture can be adjusted through the interactive interface or a layer can be paused to focus on key information.
[0035] In step S800, the layered annotation method for power grid images also includes outputting the annotation results in a standard format. Outputting the annotation results in a standard format includes: converting the annotation results into common image formats (such as JPEG, PNG, etc.) or vector graphics formats (such as SVG, etc.); exporting the common image or vector graphics formats into structured data in formats such as CSV and JSON to facilitate subsequent data analysis and mining; and generating an annotation report using the structured data. The annotation report includes the annotation results, annotation time, and annotator, facilitating project management and auditing. This effectively improves efficiency.
[0036] In summary, this layered power grid image annotation method reduces visual complexity by displaying images in layers. It then uses a user interface to gradually switch between layers, actively focusing on and magnifying the annotated features of the next image, thus achieving an actively guided switching effect. This method reduces visual strain and improves attention while actively guiding the eye to observe the features and annotated points of the next image. The gradual switching improves the smoothness of switching between images and avoids image tearing. This addresses the difficulty traditional power grid image annotation methods face in guiding the eye to focus and conduct detailed analysis during manual annotation inspection. Interactive correction improves annotation accuracy. The interface allows for adjusting the display order of the first, second, third, fourth, and fifth images, or pausing a layer to focus on key information. Annotation results are output in a standard format to facilitate project management and auditing, effectively improving efficiency. This method enhances defect recognition: the recognition rate of small defects (<1 mm) has increased from 68% to 92%, and the mislabeling rate has decreased by 18%. Optimized human-machine collaboration: Eye trackers and shortcut keys reduce the number of steps by 70, aligning with the work habits of power inspectors. Improved annotation efficiency: Layered annotation reduces the amount of data processed at a time, and combined with dynamic display to optimize human-machine collaboration, annotation speed increases by over 30%. Enhanced accuracy and reliability: Independent sub-models reduce cross-layer interference, reducing the error rate of defective layers to less than 5%. Reduced labor costs: Dynamic visual persistence guidance reduces manual re-inspection time, resulting in a 40% overall cost savings.
[0037] Second, the present invention provides a dynamic annotation display system for executing the aforementioned layered annotation method for power grid images. The dynamic annotation display system includes a server, an eye-tracking device, a human-computer interaction device, and a display device. The server is equipped with an environmental layer model, a tower layer model, an equipment layer model, and a component defect layer model, and is used to annotate images. The eye-tracking device is used to detect the gaze point of the human eye, automatically extending the display time of the current view or triggering the annotation correction interface. The human-computer interaction device is used to manually switch layers, zoom in / out the view, and add / delete annotations. The display device is equipped with a color coding module, which is used to color-code and dynamically amplify image features, and, in combination with the eye-tracking device, guides manual review of image annotations. When in use, the server is responsible for receiving the target power grid image and marking it. During the manual review and marking stage, the eye tracking device can be used to easily capture which layer of the image the human eye is focusing on, so that this layer of image can stay for a long time for manual review. At the same time, the images can be switched at will. When modifying the annotation, the modification operation is implemented through human-computer interaction devices such as keyboard and mouse. When switching between images, the display device and color coding module are mainly relied on to color-code the annotation results (such as red highlight) and dynamically amplify the visual persistence effect to guide manual re-inspection, improve the manual inspection effect, especially the inspection and review of the defective layer annotation results, thereby effectively improving the fluency and accuracy of user use and improving the user experience.
[0038] It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit it. Based on the embodiments provided in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0039] Obviously, the accompanying drawings are merely examples or embodiments of the present application. A person skilled in the art can also apply the present application to other similar situations based on these drawings without inventive effort. Furthermore, it is understandable that, although the work involved in this development process may be complex and lengthy, certain design, manufacturing, or production changes based on the technical content disclosed in this application are merely routine technical means for a person skilled in the art and should not be considered to constitute a deficiency in the disclosure of the present application.
Claims
1. A hierarchical labeling method for power grid images, characterized in that: include: Receive target power grid image; Annotating the target power grid image using an environmental layer model to obtain a first image with a first annotation, wherein the environmental layer model is used to identify terrain, vegetation, and weather in the image and generate corresponding first annotations; Annotating the target power grid image using a tower layer model to obtain a second image with a second annotation, wherein the tower layer model is used to identify features of towers in the image and generate corresponding second annotations; Annotating the target power grid image using the device layer model to obtain a third image with third annotations, wherein the device layer model is used to identify features of power equipment in the image and generate corresponding third annotations; Annotating the target power grid image using the device layer model to obtain a fourth image with fourth annotations, wherein the device layer model is used to identify features of power components in the image and generate corresponding fourth annotations; annotating the target power grid image using a component defect layer model to obtain a fifth image with a fifth annotation, wherein the component defect layer model is used to identify characteristics of power component defects in the image and generate corresponding fifth annotations; Displaying the first picture, the second picture, the third picture, the fourth picture, and the fifth picture in a time-sharing and gradual manner through a user interaction interface; Obtain user annotation feedback on the first picture, the second picture, the third picture, the fourth picture, and the fifth picture through the user interaction interface.
2. The grid image layered labeling method according to claim 1, characterized in that: The time-sharing gradual switching is as follows: during the image switching process, the transparency of the previous image decreases to 0% and the marked features gradually shrink, while the transparency of the next image gradually increases to 100% and the marked features gradually enlarge.
3. The grid image layered labeling method according to claim 2, characterized in that: When the first picture, the second picture, the third picture, the fourth picture and the fifth picture are displayed in sequence, each layer of pictures flashes at an interval of 0.2 seconds.
4. The grid image layered labeling method according to claim 1, characterized in that: The environment layer model is obtained by deep learning of the ResNet-50 model using the environmental standard database. The tower layer model is obtained by deep learning of the YOLOv7 model using the tower standard database. The equipment layer model is obtained by deep learning of the Mask R-CNN model using the equipment standard database. The component defect layer model is obtained by transfer learning from the standard defect library of cracks and rust.
5. The power grid image layered labeling method according to claim 1, characterized in that: Also includes: According to the annotation feedback, the annotations of the first picture, the second picture, the third picture, the fourth picture, and the fifth picture are interactively corrected or not corrected, and an annotation result is obtained.
6. The power grid image layered labeling method according to claim 5, characterized in that: Interactively modify annotation results, including: By using an eye tracking device to detect the gaze point of the human eye, the system can automatically extend the display time of the current view or trigger the annotation correction interface. Manually switch layers, zoom in / out views, and add / delete annotations using shortcut keys or mouse interaction; The result of annotation correction is displayed in real time on the annotation correction interface.
7. The power grid image layered labeling method according to claim 6, characterized in that: The display order of the first picture, the second picture, the third picture, the fourth picture and the fifth picture is adjusted through the interactive interface, or a layer is paused to focus on key information.
8. The power grid image layered labeling method according to claim 1, characterized in that: The method also includes performing image preprocessing on the power grid image to obtain a target power grid image, which includes: After receiving the target power grid image, Gaussian filtering and median filtering methods are used to remove noise from the image; Use contrast stretching and histogram equalization techniques to enhance image details and contrast.
9. The power grid image layered labeling method according to claim 5, characterized in that: It also includes outputting the annotation results in a standard format. Outputting the annotation results in a standard format includes: Convert the annotation results into common image formats or vector graphics formats; Export common image formats or vector graphics formats as structured data; Generate annotation reports using structured data. The annotation reports include annotation results, annotation time, and annotation personnel.
10. A dynamic annotation display system, characterized in that: It is used to execute the power grid image layered annotation method according to any one of claims 1 to 9, and the dynamic annotation display system includes: The server is equipped with an environment layer model, a tower layer model, an equipment layer model, and a component defect layer model, and is used to annotate images; Eye tracking devices, which detect where the eyes are looking and automatically extend the display time of the current view or trigger a annotation correction interface; Human-computer interaction device, which is used to manually switch layers, zoom in / out views, and add / delete annotations; The display device is equipped with a color coding module, which is used to color-code and dynamically amplify image features, and is combined with an eye tracking device to guide manual review of image annotations.