Power transmission line icing picture sample generation method based on diffusion algorithm

Through the transmission line ice-covered image sample generation method based on diffusion algorithm, the problem of limited sample size and safety hazards in the prior art is solved, high-quality sample generation and identification accuracy is improved, and the drone acquisition process is simplified.

CN118941952AActive Publication Date: 2024-11-12NANJING YOUKUO ELECTRICAL TECH

Patent Information

Application Number
CN202410975395.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-19
Publication Date
2024-11-12
Estimated Expiration
2044-07-19

AI Technical Summary

Technical Problem

The existing transmission line ice-covered detection mainly relies on manual inspection and image recognition technology. The sample size is limited, the generalization ability is insufficient, and there are safety hazards in extreme weather conditions, making it difficult to achieve automated batch sample generation.

Method used

Using the transmission line ice-covered image sample generation method based on diffusion algorithm, we collect initial sample pictures and environmental information, identify and label target elements and key elements, generate new sample pictures, and write them to the model material library, and use images to generate models Carry out the graph production process and generate multiple new sample pictures close to the real situation for training the model.

Benefits of technology

It realizes the generation of a large number of high-quality training samples when there is insufficient training samples, reduces the workload of manual collection and labeling, improves the sample quality, ensures that the newly generated samples are close to the real situation, simplifies the drone collection process, and improves the Identification accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118941952A_ABST
    Figure CN118941952A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of artificial intelligence, and provides a power transmission line icing picture sample generation method based on a diffusion algorithm, and the method comprises the steps: collecting an initial sample picture of a power transmission line and first environment information; target elements and key elements in the initial sample picture are identified and extracted, and the target elements and the key elements are labeled based on the first environment information to obtain a first picture sample; writing each first picture sample into a model material library and triggering a picture generation process to obtain a new sample picture; and marking each new sample picture and the initial sample picture based on the first environment information to obtain a second picture sample. Based on the method, the method can generate images through images, expand model training samples, and reduce the manual collection cost and risk.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method for generating transmission line icing image samples based on a diffusion algorithm, a transmission line icing monitoring method, and a transmission line monitoring system. Background Art

[0002] With the rapid development of image processing and machine learning technologies, automated transmission line monitoring technology has become an important part of the power industry.

[0003] At present, the detection of ice cover on power transmission lines mainly adopts manual inspection combined with image recognition technology, and the recognition model is trained through manually collected pictures. However, this method not only has a limited sample size and limitations in sample collection, which also leads to insufficient generalization ability of the training model, but also poses a safety hazard when manually collecting samples under extreme weather conditions. Summary of the invention

[0004] In order to realize the automatic batch generation of sample images and reduce the burden of manual sample collection, an embodiment of the present application provides a method for generating transmission line ice-covered image samples based on a diffusion algorithm, the method comprising the steps of: collecting an initial sample image of a transmission line and first environmental information; identifying and extracting target elements and key elements in the initial sample image, and annotating each of the target elements and the key elements based on the first environmental information to obtain a first image sample; wherein the image generation process includes generating the new sample image based on an image generation model, and the image generation model is implemented based on a diffusion algorithm; writing each of the first image samples into a model material library and triggering the image generation process to obtain a new sample image; annotating each of the new sample images and the initial sample image based on the first environmental information to obtain a second image sample.

[0005] Based on the above technical solution, multiple new sample images can be generated based on the initial sample image, realizing the automatic expansion of model training samples. In addition, in the image generation process, the elements used are based on real collected images, thereby ensuring that the new images are more in line with the actual situation. Compared with directly relying on large models for image generation, the newly generated images can better meet the model training requirements.

[0006] In one implementation, the image generation process includes: retrieving a historical first image sample whose annotation information matches the first environmental information from the model material library; constructing an image output requirement based on the target element, the key element, and the historical first image sample; and processing the output requirement based on an image generation model to obtain the new sample image.

[0007] Based on the above technical solution, the matching historical first image sample is obtained from the model material library, which ensures that the elements used in the image generation process are all from real images, and the image output requirements are constructed based on these real materials, so that the image generation range of the image generation model is limited, thereby further ensuring that the output results are closer to the actual situation.

[0008] In one implementation, the first environmental information includes temperature data and humidity data; the tag information matches the first environmental information including: the difference between the temperature data and humidity data in the tag information and the corresponding data in the first environmental information are both within a corresponding preset range.

[0009] Based on the above technical solution, by screening the historical first image sample according to the temperature and humidity data, historical elements with the same or similar environmental information as the first image sample are obtained, thereby ensuring the rationality and relevance of each element in the new sample image and improving the sample quality.

[0010] In one implementation, the temperature data and humidity data are collected by sensors installed on a drone when taking the initial sample image.

[0011] Based on the above technical solution, it can be ensured that the collected temperature and humidity data are consistent with the relevant data in the environment where the transmission line is located, which is more authentic and reliable than obtaining these data based on meteorological information.

[0012] In one implementation, the first environmental information includes altitude data, and the method further includes: screening the new sample image based on the altitude data.

[0013] Based on the above technical solution, new sample images are screened according to height data, which can improve the rationality of new sample images used to generate training samples.

[0014] Based on the same inventive concept, an embodiment of the present application also provides a method for monitoring icing of transmission lines, which is applied to a transmission line monitoring system, wherein the system includes a data acquisition device and a data analysis device, and the method includes: the data analysis device receives a real-time image of the transmission line to be monitored and second environmental information sent by the data acquisition device; captures at least one picture to be identified from the real-time image; identifies the picture to be identified and the second environmental information based on a panoramic image recognition model, and outputs a first recognition result; wherein the training sample of the panoramic image recognition model is the second picture sample obtained based on the above-mentioned transmission line icing picture sample generation method based on the diffusion algorithm; and performs monitoring and analysis based on the first recognition result.

[0015] Based on the above technical solution, a large number of second image samples can be generated for training the panoramic image recognition model when there are few initial sample images, so that the panoramic image recognition model can be put into use as soon as possible to avoid the extension of the system development cycle due to insufficient samples. In addition, the recognition accuracy of the panoramic image recognition model can be improved through multi-modal sample input.

[0016] In one implementation, the method also includes: identifying and extracting target elements and key elements in the picture to be identified, and generating a first identification picture and a second identification picture accordingly; identifying the first identification picture, the second identification picture and the second environmental information based on a scene-by-scene image recognition model, and outputting a second identification result; and performing monitoring and analysis based on the first identification result and the second identification result.

[0017] Based on the above technical solution, by introducing a scene-by-scene image recognition model to identify the target elements and key elements in the picture to be identified, the system can be assisted in judging the recognition results of the panoramic image recognition model. Especially in the initial stage of application, when the panoramic image model is insufficiently trained, the final recognition result can be determined by combining the recognition results of the two, which can provide recognition accuracy.

[0018] In one implementation, capturing at least one picture to be identified from the real-time image includes capturing an image corresponding to at least one key frame from the real-time image as the picture to be identified.

[0019] In addition, an embodiment of the present application also provides a power transmission line monitoring system, the system including a data acquisition device and a data analysis device, the data analysis device is used to implement the above-mentioned power transmission line icing monitoring method.

[0020] An embodiment of the present application also provides an electronic device, which includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction implements the above method when executed by the processor.

[0021] In summary, the embodiments of the present application at least include the following beneficial technical effects: 1. Generate images from one image. When there are insufficient training samples, multiple training sample images can be generated based on one initial sample image, thereby reducing the workload of manual collection and sample labeling.

[0022] 2. To a certain extent, ensure that the newly generated sample images are close to the actual situation and improve the sample quality.

[0023] 3. Collect video images of the power transmission lines through drones, and extract the pictures corresponding to the key frames in the video images as the pictures to be identified. This not only ensures the integrity and clarity of the pictures to be identified, but also reduces the difficulty of drone collection, facilitates the rapid completion of collection work, and reduces the failure rate of drones. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application.

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0026] Figure 1 A flow chart of a method for generating ice-covered image samples of power transmission lines based on a diffusion algorithm provided in an embodiment of the present application is shown.

[0027] Figure 2 A flow chart of the image generation method provided in an embodiment of the present application is shown.

[0028] Figure 3 A flow chart of a method for generating ice-covered image samples of power transmission lines based on a diffusion algorithm is shown in another embodiment of the present application.

[0029] Figure 4 The embodiment of the present application also provides a flow chart of a method for monitoring ice coating on a power transmission line. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0031] In the description of the embodiments of the present application, unless otherwise specified, "multiple" means two or more, and "first", "second" and various digital numbers are only distinguished for the convenience of description and are not used to limit the scope of the embodiments of the present application.

[0032] The features, structures or characteristics in this application may be combined in one or more embodiments in any suitable manner. In various embodiments of this application, the size of the sequence number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0033] Some optional features in the embodiments of the present application may be implemented independently in some scenarios without relying on other features to solve corresponding technical problems and achieve corresponding effects. They may also be combined with other features in some scenarios as needed.

[0034] In this application, unless otherwise specified, the same or similar parts between the various embodiments can refer to each other. In the various embodiments of this application, if there is no special description and logical conflict, the terms and / or descriptions between different embodiments are consistent and can be referenced to each other, and the technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationships. The implementation methods of this application do not constitute a limitation on the scope of protection of this application.

[0035] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0036] Please refer to Figure 1 The embodiment of the present application provides a method for generating a transmission line ice-covered image sample based on a diffusion algorithm, which is executed by an image generation server and specifically comprises the following steps: S110, collecting initial sample images of transmission lines and first environmental information.

[0037] In a specific implementation, the method for collecting the initial sample images of the power transmission line may include manual collection based on an image acquisition device, and collection by a drone equipped with an image acquisition device. The collected initial sample images of the power transmission line are images of ice-covered power transmission lines taken in real scenes.

[0038] In one example, the first environmental information includes temperature data and humidity data. The temperature data and humidity data are used to indicate the environment in which the ice-covered power transmission line is located at that time, and may be the ambient temperature and humidity at the time when the initial sample image is collected. These data may be collected by temperature and humidity sensors when the initial sample image is taken. The temperature and humidity sensors may be installed on image acquisition equipment or directly on drones. Since they are close to the ice-covered power transmission line, they may more realistically reflect the environmental conditions of the power transmission line.

[0039] S120, identifying and extracting target elements and key elements in the initial sample image, and marking each target element and key element based on the first environmental information to obtain a first image sample.

[0040] The target element is an image of an ice-covered power transmission line, and the key element is an environmental element in the background of the image, such as mountains, trees, buildings, fields, rivers, birds, and the sky. In the example, each element in the initial sample image can be identified and extracted based on the image feature extraction technology to generate independent element images, and these element images are labeled based on the first environmental information and the identified element names.

[0041] In one implementation, the target element and the key element annotation information also include ice thickness information, which is used to indicate the ice thickness of the transmission line in the target element, which can be a specific value or a thickness level, and the present application is not limited thereto. The thickness information can be obtained manually in combination with actual experience, or it can be calculated based on image processing technology. Preferably, the thickness information can be detected synchronously during the process of collecting the initial sample image.

[0042] It is worth noting that although the key elements do not include transmission lines, they are in the same environment as the target elements, so they can be used to indirectly reflect the possible ice thickness of the transmission lines in the current environment. Based on this, the thickness information of the key elements is annotated, which is also used to realize the estimation of the ice coverage of the transmission lines based on the background environment elements.

[0043] In one implementation, the annotation information of the target element also includes an ice-covered position. In one example, the ice-covered position can be characterized by the distance between the target element and a key component, wherein the key component may include a pole tower.

[0044] S130, writing each first image sample into the model material library and triggering the image generation process to obtain a new sample image.

[0045] Specifically, the model material library is used to store the first image samples obtained through the above steps. These samples can be used to generate new sample images. In one implementation, a trigger can be added to the model material library to trigger the image generation process when new data is written. Based on this, even if there is no data in the model material library before the first image sample is written, the first image sample can also be used as the image generation material this time, thereby achieving rapid image generation.

[0046] Please refer to Figure 2 , the graph generation process method specifically includes: S210: Retrieve a historical first picture sample whose annotation information matches the first environmental information from the model material library.

[0047] The matching of the annotation information with the first environment information includes: the difference between the temperature data and the humidity data in the annotation information and the temperature data and the humidity data in the first environment information are both within the corresponding preset range.

[0048] Specifically, the image generation server stores a temperature preset range and a humidity preset range, which are set based on the impact of historical meteorological data on ice thickness. For example, when the temperature is above 0°C, the ice and snow may be in the melting stage, and the ice thickness changes rapidly. Therefore, the temperature preset range value is set to 2°C, that is, the temperature change range is within 2°C, and the ice coverage of the transmission line is roughly the same; when the temperature is below 0°C, the ice and snow melt slowly and change slowly, and the temperature preset range value can be set to 5°C. Similarly, the humidity preset range can be set according to different humidity ranges, which can be determined based on the historical data of the corresponding region.

[0049] Based on this, the historical first picture samples that meet the requirements, namely the historical target elements and historical key elements, are retrieved through the preset range to provide effective basic materials for the subsequent image generation steps, which can improve the rationality of the new pictures. It is worth noting that the historical first picture samples are retrieved from the model material library, so the first picture samples are also included in the historical first picture samples.

[0050] S220, constructing image output requirements based on the target element, the key element and the first historical image sample.

[0051] In one example, the image generation server may construct output requirements based on the target element, key elements, and historical target elements or historical key elements contained in the historical first image sample, including generating a new image based on the target element and at least one historical key element, or generating a new image based on the historical target element and at least one key element, etc.

[0052] S230, processing the output requirement based on the image generation model to obtain a new sample image.

[0053] In one example, the image generation model can be constructed based on a diffusion algorithm, and the output requirements are input into the image generation model as prompt words. The model replaces elements of the initial sample image based on the diffusion algorithm to obtain a new sample image.

[0054] In one example, in order to improve the efficiency of image generation, a pipelined image generation method can be implemented based on the workflow of ComfyUI, that is, the folder storing the initial sample images is set as the source path, and the workflow of reading the initial sample images from the source path to ComfyUI in sequence is implemented through a script. The workflow is pre-loaded with an image generation model, based on which new sample images can be generated in batches, that is, the newly generated sample images corresponding to all the initial sample images in the folder are batch output. Users only need to store the initial sample images in the source path.

[0055] ComfyUI's workflow is a node-based graphical user interface (GUI) designed for StableDiffusion to create complex image generation workflows. By dragging and connecting different nodes, users can build the entire process from loading models to generating images.

[0056] In one implementation, in order to enrich the sample style, the new sample images output by the model can also be processed in a process-based manner based on the image processing workflow, and the images obtained after each step of the process-based processing are saved as new sample images.

[0057] In one example, an image processing workflow can be implemented based on ComfyUI. Specifically, an image processing workflow can be built based on ComfyUI. For example, the image can be dimmed, filtered, and style converted in sequence to finally obtain a target sample image and intermediate sample images corresponding to each processing node. These sample images are all saved as new sample images for subsequent operations. In this way, multiple new sample images can be generated based on an initial sample image. When the initial sample images are small, the number of samples is greatly increased.

[0058] It is worth noting that the above-mentioned image generation process based on ComfyUI is different from the native function of ComfyUI, so it is necessary to implant a custom script in the ComfyUI software to implement it. For example, by running the script, the front-end request mechanism of ComfyUI is modified to automatically scan the specified directory (source path), and automatically build a processing request based on the scanned new initial sample image, and send it to the back-end service to request the initial sample image to be processed based on the workflow. In this way, the image can be automatically read and processed, saving a lot of manual operations.

[0059] In another implementation, each node in the workflow can be monitored through a script. When a new image is generated at the node, the image data is copied and output to a specified path for storage. Based on this, the intermediate sample images in the workflow can be output without affecting the processing flow of the workflow.

[0060] Please go back Figure 1 , the method further comprises: S140: Label each new sample image and the initial sample image based on the first environment information to obtain a second image sample.

[0061] Specifically, the annotation method of the second sample image is the same as that in the implementation method of step S120, and the annotated content is also the same, that is, the annotation content of the second image sample may include ice thickness data and ice position data, etc.

[0062] Based on the above technical solution, it is possible to generate multiple new sample images based on an initial sample image, and the materials generated by the new sample images are all from the real collected sample images, which can improve the effectiveness of the new sample images. Furthermore, by selecting the corresponding raw image basis based on temperature and humidity data, the degree to which the new samples are close to the real images can be further improved, thereby ensuring the quality of the training samples.

[0063] In a preferred embodiment, in order to ensure the quality of the second image sample, the new sample image may be screened by introducing height data to ensure the rationality of the new sample image.

[0064] Please refer to Figure 3 , Figure 3 A method for generating ice-covered image samples of power transmission lines based on a diffusion algorithm is shown in another embodiment of the present application, comprising the steps of: S310, collecting initial sample images of transmission lines and first environmental information.

[0065] The difference between this step and the above-mentioned step S110 is that the first environmental information also includes altitude data, that is, the altitude of the image acquisition device when taking the image of ice-covered power transmission lines.

[0066] S320, identifying and extracting target elements and key elements in the initial sample image, and marking each target element and key element based on the first environment information to obtain a first image sample.

[0067] S330, write each first image sample into the model material library and trigger the image generation process to obtain a new sample image.

[0068] S340, screening new sample images based on the height data.

[0069] Specifically, the height data can be used to judge the rationality of key elements. For example, above a certain altitude, key elements should not include shorter objects such as trees and houses. Therefore, the correspondence between height data and key element types can be established in advance, based on which new sample images can be screened, and new sample images with higher rationality can be retained for subsequent sample generation. This improves the rationality of new sample images to a certain extent.

[0070] S350: annotate the screened new sample images and the initial sample images based on the first environment information to obtain second image samples.

[0071] The operation of the above steps S320, S330 and S350 is substantially the same as that of steps S120, S130 and S140, except that the annotation information further includes height data.

[0072] Based on the same inventive concept, an embodiment of the present application also provides a method for monitoring icing of a power transmission line, which is applied to a power transmission line monitoring system. The system includes a data acquisition device and a data analysis device, and the two exchange data through a network.

[0073] Please refer to Figure 4 , the method is performed by a data analysis device, comprising: S410, receiving a real-time image of the power transmission line to be monitored and second environmental information sent by a data acquisition device.

[0074] In one example, the data acquisition device is a drone equipped with a camera and temperature and humidity sensors. The drone is remotely controlled to collect real-time images of the power transmission line and send the real-time images, temperature data, humidity data and flight altitude data to the data analysis device. It is worth noting that in order to ensure the quality of the image, the data acquisition device captures a video image.

[0075] S420, capturing at least one picture to be identified from the real-time image.

[0076] Among them, at least one image corresponding to a key frame can be intercepted from the real-time image as the image to be identified. Generally speaking, the image corresponding to the key frame is clearer and it is easier to identify the target elements and key elements therein. Therefore, the image corresponding to all the key frames in the real-time image can be directly used as the image to be identified, or several images can be selected from them as the images to be identified.

[0077] S430: Identify the image to be identified and the second environmental information based on the panoramic image recognition model, and output a first recognition result.

[0078] Among them, the training sample of the panoramic image recognition model is the second picture sample obtained by the method for generating ice-covered transmission line pictures based on the diffusion algorithm provided in the embodiment of the present application. The input layer of the panoramic image recognition model is all the pictures to be recognized and the second environmental information obtained in this collection. The panoramic image recognition model can be implemented based on late fusion (LateFusion) or Transformer and self-attention mechanism, so that the recognition efficiency of the model can be improved through the multimodal input layer design. The output layer of the panoramic image recognition model may include the ice coverage of the transmission line, such as whether it is covered with ice, the thickness of the ice, and the location of the ice.

[0079] S440: Perform monitoring and analysis based on the first recognition result.

[0080] In one implementation, when the first recognition result indicates that the transmission line is covered with ice, the data analysis server can instruct the drone to continue to collect images of other sections of the transmission line and identify them, and then construct an ice image of the entire line according to the recognition results of different sections of the same transmission line, establish an ice three-dimensional model, and then combine the second environmental information and meteorological information to predict the duration of ice and the thickness change trend, add the predicted time data to the ice three-dimensional model to obtain a demonstration animation of the ice three-dimensional model within the predicted time, so as to show the ice change of the transmission line in the predicted period, and combine various warning data to perform fault warning analysis based on the ice three-dimensional model, including the current ice position, ice degree, ice change in the predicted time period and the corresponding warning level, and finally visualize the warning dynamic picture constructed based on the ice three-dimensional model through the monitor. It is worth noting that in order to ensure the real-time monitoring, the ice three-dimensional model is only used to display the ice condition of the transmission line within a preset distance range (such as within 60km), which can reduce the amount of data processing and improve the system analysis efficiency. The system can monitor the transmission lines in different ranges by setting up multiple data acquisition devices and multiple data analysis devices to ensure the real-time monitoring and analysis.

[0081] Furthermore, in order to achieve self-updating of training samples, the data analysis equipment can determine the collection location of the picture to be identified based on the positioning data of the drone, that is, the actual ice-covered location corresponding to the picture, and verify the actual ice-covered location with the ice-covered location in the model recognition result to determine the recognition accuracy of the model. When the recognition error is large, it is determined to be a misrecognition, and the scene-by-scene image recognition model is introduced for further recognition. At the same time, the picture to be identified is stored in the source path as a new initial sample picture. In this way, the sample data can be automatically collected, thereby realizing the continuous training of the image recognition model.

[0082] As mentioned above, in actual applications, due to the fact that the initial recognition accuracy of the panoramic image recognition model is not ideal, or the collected images may not be clear enough due to snowy weather, resulting in misrecognition or unrecognition of the pictures to be recognized. Based on this, the pictures to be recognized can be re-recognized by introducing a scene-by-scene image recognition model.

[0083] In one implementation, the method further includes: identifying and extracting target elements and key elements in the image to be identified, and correspondingly generating a first identification image and a second identification image; identifying the first identification image, the second identification image and the second environmental information based on a scene-by-scene image recognition model, and outputting a second identification result; and performing monitoring and analysis based on the first identification result and the second identification result. The scene-by-scene image recognition model is obtained by training based on the first image sample in the model material library.

[0084] Based on the split-scene image recognition model, even when only part of the picture to be recognized is clear, the recognition result can be obtained based on the recognition of elements. Especially in the early stage of system use, the recognition accuracy of the panoramic image recognition model may not be high due to insufficient sample data. Therefore, the first recognition result can be verified or corrected by combining the split-scene image recognition model. For example, when the results of the two are consistent, the confidence of the panoramic image recognition model can be improved. If they are inconsistent, the result data can be fed back to the user, and new training samples can be constructed based on the suggestions of user feedback for optimization training of the panoramic image recognition model and the split-scene image recognition model, thereby continuously improving the recognition accuracy of the model.

[0085] It is worth noting that since the recognition processing process of the split-scene image recognition model is relatively complicated and is relatively limited by image recognition technology, in some implementations of the present application, the split-scene image recognition model is used for auxiliary verification only when the recognition accuracy of the panoramic image recognition model is not ideal. When the recognition accuracy (i.e., confidence) of the panoramic image recognition model meets the preset target, there is no need to use the split-scene image recognition model in the image recognition process, thereby saving system computing resources.

[0086] It is understandable that when there are few initial sample images and the model training is insufficient, the recognition accuracy of the split-scene image recognition model is higher than that of the panoramic image recognition model because: the range of the first image sample is wider than that of the second image sample, the difference between the samples is larger, and the recognition objects include target elements and key elements. Even if some elements are not clear enough, the results can be output based on the recognition results of other elements. Therefore, in the early stage of system use, the panoramic image recognition model can be verified and the confidence level can be judged by introducing the split-scene image recognition model, which can improve the image recognition accuracy in the early stage of application and ensure the normal use of the system.

[0087] In addition, an embodiment of the present application also provides an electronic device, which includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements a method as in any one of the implementations in the embodiments of the present application; wherein the processor may adopt a general-purpose central processing unit (CPU), a microprocessor, an application specific integrated circuit (ASIC), a graphics processing unit (GPU), or one or more integrated circuits, for executing relevant programs to implement the method in any one of the implementations in the embodiments of the present application.

[0088] The processor may also be an integrated circuit electronic device with signal processing capability. In the implementation process, each step of the method in any implementation of the embodiments of the present application may be completed by an integrated logic circuit of hardware in the processor or by instructions in software form.

[0089] The above-mentioned processor can also be a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware decoding processor to be executed, or the hardware and software modules in the decoding processor can be combined and executed.

[0090] The software module may be located in a random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, register or other mature storage media in the art. The storage medium is located in the memory, and the processor reads the information in the memory, and combines its hardware to complete the functions required to be performed by the units included in the data processing device of the embodiment of the present application, or executes the method in any one of the implementation modes in the embodiment of the present application.

[0091] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for generating ice-covered image samples of power transmission lines based on a diffusion algorithm, characterized in that: The method comprises the steps of: Collecting initial sample images of transmission lines and first environmental information; Identify and extract target elements and key elements in the initial sample image, and mark each of the target elements and the key elements based on the first environmental information to obtain a first image sample; Writing each of the first image samples into the model material library and triggering the image generation process to obtain a new sample image; wherein the image generation process includes generating the new sample image based on an image generation model, and the image generation model is implemented based on a diffusion algorithm; Based on the first environmental information, each of the new sample pictures and the initial sample picture is labeled to obtain a second picture sample.

2. The method according to claim 1, characterized in that The graph generation process includes: Retrieving a historical first picture sample whose annotation information matches the first environmental information from the model material library; Constructing a picture output requirement based on the target element, the key element, and the historical first picture sample; The output requirement is processed based on the image generation model to obtain the new sample image.

3. The method according to claim 2, characterized in that The first environmental information includes temperature data and humidity data; the tag information matches the first environmental information including: the difference between the temperature data and humidity data in the tag information and the corresponding data in the first environmental information are both within a corresponding preset range.

4. The method according to claim 3, characterized in that The temperature data and humidity data are collected by sensors installed on the drone when taking the initial sample pictures.

5. The method according to claim 1, characterized in that The first environmental information includes altitude data, and the method further includes: screening the new sample image based on the altitude data.

6. A method for monitoring ice coating on a power transmission line, characterized in that: Applied to a power transmission line monitoring system, the system includes a data acquisition device and a data analysis device, and the method includes: The data analysis device receives the real-time image of the power transmission line to be monitored and the second environmental information sent by the data acquisition device; Intercepting at least one picture to be identified from the real-time image; The image to be identified and the second environmental information are identified based on a panoramic image recognition model, and a first recognition result is output; wherein the training sample of the panoramic image recognition model is the second image sample obtained by the method for generating an ice-covered image sample of a power transmission line based on a diffusion algorithm according to any one of claims 1 to 5; Monitoring and analysis are performed based on the first identification result.

7. The method according to claim 6, characterized in that The method further comprises: Identify and extract target elements and key elements in the image to be identified, and generate a first identification image and a second identification image accordingly; Identify the first identification picture, the second identification picture and the second environment information based on the scene segmentation image recognition model, and output a second identification result; Monitoring and analysis are performed based on the first recognition result and the second recognition result.

8. The method according to claim 6, characterized in that The capturing at least one picture to be identified from the real-time image includes capturing an image corresponding to at least one key frame from the real-time image as the picture to be identified.

9. A power transmission line monitoring system, characterized in that: The system comprises a data acquisition device and a data analysis device, and the data analysis device is used to implement the method according to any one of claims 6 to 8.

10. An electronic device, characterized in that: The electronic device includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction implements the method according to any one of claims 1 to 8 when executed by the processor.

Citation Information

Patent Citations

  • Power transmission line icing monitoring method, device and system based on power internet of things and equipment

    CN114049576A

  • Image generation method and device, model training method and device, equipment and storage medium

    CN116823994A

  • Method for improving generalization ability of target detection model based on diffusion model

    CN117253110A

  • Power transmission line icing early warning method and system based on multi-element fusion

    CN117494040A

  • Artificial intelligence fault identification system and method based on power transmission line transient waveform

    WO2022183698A1

Cited By

  • Information identification method and device, medium and intelligent equipment

    CN119830934A