CN-121982686-A - Lightning arrester instrument data image identification method, system, equipment and medium

Abstract

The invention discloses a lightning arrester instrument data image recognition method, a system, equipment and a medium, which comprise the steps of obtaining an original image of a lightning arrester instrument, adaptively adjusting acquisition parameters according to ambient illumination, carrying out dynamic contrast enhancement and multi-scale noise suppression through an illumination decoupling model after geometric correction and quality evaluation to generate a standardized image, inputting a light-weight mobile terminal target detection model, positioning and cutting an instrument display area, enhancing sub-images and extracting edge characteristics, judging instrument types, dividing character areas and recognizing by a special network to obtain initial readings, checking validity in a reasonable range by combining preset physical quantities to generate final readings, automatically associating current inspection tasks, attaching time, positions and image indexes, storing the additional time, the position and image indexes into a local database through transaction operation, and supporting to generate a structured inspection report. The invention improves the accuracy and efficiency of the inspection work and ensures the reliability and traceability of the data.

Inventors

• ZHAO WEI
• Xie Yanjiang
• YU CHAOYAN
• YU JUNJIE
• YU RULIAN
• ZHANG YINGLONG
• LI LIN
• ZHANG YAO
• WANG YOUJUN
• ZHENG XIAOHU
• SI YANG
• ZHANG JIE
• Li Fenzong
• LI KANG
• LI GUANGXUN
• QI YANBO
• YAO QUN

Assignees

• 贵州电网有限责任公司

Dates

Publication Date

20260505

Application Date

20251218

Claims (10)

1. 1. A method for identifying data images of lightning arrester instruments, comprising the steps of: Acquiring an original image of a lightning arrester instrument, adaptively adjusting image acquisition parameters according to ambient illumination conditions, performing geometric correction and quality evaluation on the original image, and performing dynamic contrast enhancement and multi-scale noise suppression based on an illumination decoupling color space model to generate a standardized image; Inputting the standardized image into a lightweight target detection model deployed at a mobile terminal, and positioning and cutting out a target sub-image containing an instrument display area; Performing image enhancement and edge feature extraction on the target sub-image, judging the instrument type based on edge distribution characteristics, and performing region segmentation and special identification network analysis on the display content of the digital instrument to obtain initial instrument reading; And carrying out validity verification on the initial instrument reading based on a preset physical quantity reasonable range, generating a final instrument reading, associating the final instrument reading with a current inspection task session, automatically attaching time information, a geographic position and an original image index, storing the time information, the geographic position and the original image index into a local structured database through transactional operation, and supporting to generate a structured report containing a recognition result and image evidence according to the inspection task.
2. 2. The method for identifying the data image of the lightning arrester instrument according to claim 1, wherein the color space model based on illumination decoupling performs dynamic contrast enhancement and multi-scale noise suppression, and the method comprises the following steps: converting the geometrically corrected image to a decoupling color space comprising a structural luminance component and an illumination intensity component; Performing bi-directional gamma correction on the illumination intensity component; Carrying out multi-scale Gaussian pyramid fusion on the image subjected to the bidirectional gamma correction; And performing edge-preserving guide filtering denoising on the image obtained by the multi-scale Gaussian pyramid fusion to generate a standardized image.
3. 3. The method for identifying the data image of the lightning arrester instrument according to claim 1 or 2, wherein the lightweight target detection model deployed at the mobile terminal comprises the following steps: The target detection model obtained by training based on the deep learning framework is converted from an intermediate representation format; Operator fusion is carried out on a convolution layer, a batch normalization layer and an activation function layer in the model, and structured pruning is carried out on redundant channels; and carrying out statistical analysis on the dynamic range of the weight and the activation value, and then implementing eight-bit integer quantization to generate a model file suitable for the lightweight inference engine.
4. 4. The method for identifying the data image of the lightning arrester instrument according to claim 3, wherein the steps of carrying out image enhancement and edge feature extraction on the target sub-image comprise the following steps: Constructing a circular and linear structure element library containing multiple scales, and selecting structure elements adapting to the characteristic scale of the current region according to the local variance analysis result; Performing corrosion operation by adopting small-scale circular structural elements in a high noise area, and performing directional corrosion and conditional expansion by adopting linear structural elements along characteristic trend in a pointer and scale area; and calculating gradient amplitude by using a first-order differential operator, combining multi-scale phase consistency characteristics, generating an edge probability map through self-adaptive weight fusion, and dynamically setting a high-low threshold value based on local gradient distribution to carry out edge tracking and Bezier curve bridging of direction constraint.
5. 5. The method for recognizing the data image of the lightning arrester instrument according to claim 4, wherein the area division and the special recognition network analysis are performed on the display content of the digital instrument, and the method comprises the following steps: Positioning a rectangular digital display area by utilizing edge detection and morphological closing operation; performing vertical projection analysis on the rectangular digital display area image, and dividing a single symbol unit according to a projection valley value; After normalizing each symbol unit to a uniform size, a lightweight convolutional neural network formed by depth separable convolution is input to judge the symbol category.
6. 6. The method for identifying an arrester meter data image as set forth in claim 5, wherein said associating the final meter reading with the current inspection task session and storing it to the local structured database by transactional operations comprises: creating a unique session identifier when a patrol task is started, and associating all identification records by taking the session identifier as an external key; packaging into atomic transaction operation when writing reading, time information, geographic position and image index; when the identification confidence is lower than a preset threshold, marking the corresponding record in the database as a state of 'to be manually checked', and reserving an original image index.
7. 7. The method for recognizing the data image of the lightning arrester instrument according to claim 6, wherein the steps of obtaining the original image of the lightning arrester instrument, and adaptively adjusting the image acquisition parameters according to the ambient lighting conditions include: simultaneously starting wide-angle and long-focus double lenses to collect images; dynamically adjusting exposure time and white balance parameters based on a picture brightness distribution histogram and local contrast analysis; And feeding back the image definition score in real time in the acquisition process, and only keeping the image frames with the scores higher than a preset threshold value.
8. 8. A lightning arrester meter data image recognition system employing the method of any of claims 1-7, comprising: The standardized preprocessing module is used for acquiring an original image of the lightning arrester instrument, adaptively adjusting image acquisition parameters according to ambient illumination conditions, performing geometric correction and quality evaluation on the original image, and performing dynamic contrast enhancement and multi-scale noise suppression based on an illumination decoupling color space model to generate a standardized image; the area positioning module is used for inputting the standardized image into a lightweight target detection model deployed at the mobile terminal, and positioning and cutting out a target sub-image containing an instrument display area; The reading analysis module is used for carrying out image enhancement and edge feature extraction on the target sub-image, judging the type of the instrument based on the edge distribution characteristics, and carrying out region segmentation and special identification network analysis on the display content of the digital instrument to obtain an initial instrument reading; And the patrol data fusion module is used for carrying out validity verification on the initial instrument reading based on a preset physical quantity reasonable range, generating a final instrument reading, associating the final instrument reading with a current patrol task session, automatically attaching time information, a geographic position and an original image index, storing the time information, the geographic position and the original image index into a local structured database through transactional operation, and supporting the generation of a structured report containing an identification result and image evidence according to the patrol task.
9. 9. An electronic device, comprising: a memory for storing a program; a processor for loading the program to perform the steps of the method according to any one of claims 1-7.
10. 10. A computer readable storage medium storing a program, which when executed by a processor, implements the steps of the method according to any one of claims 1-7.

Description

Lightning arrester instrument data image identification method, system, equipment and medium Technical Field The invention relates to the technical field of image recognition, in particular to a lightning arrester instrument data image recognition method, a system, equipment and a medium. Background The lightning arrester is a very critical protection device in the power system, and has the main functions of limiting lightning overvoltage and operation overvoltage, and has good running state, and is directly related to the safety and stability of the whole power grid. At present, the monitoring of the meter reading of the lightning arrester mainly depends on manual inspection or uses some traditional image processing technology. The manual inspection is low in efficiency and high in labor cost, the experience level, the mental state and even the fatigue degree of inspection personnel can influence the judgment result, and the consistency and the timeliness of data are difficult to ensure. In the traditional image processing method, when the actual power field faces the conditions of severe illumination change, inclined instrument installation angle, strong dial reflection or serious background interference, the identification accuracy is often remarkably reduced, and the robustness is poor. Although in recent years, research has been carried out on a target detection algorithm (such as YOLO series) based on deep learning to be used for instrument recognition, and some effects are obtained, the models are usually large in parameters and calculation amount, most of the models can only run on a server, and low-delay and high-frame-rate real-time processing is difficult to achieve on mobile devices such as mobile phones or patrol terminals. Moreover, the existing method has a common problem that no fine design is made for the own characteristics of the lightning arrester instrument, such as poor utilization of geometric constraint of a round dial plate, insufficient feature extraction of scale marks, inaccurate pointer positioning and lack of reliable logic in automatic calculation of continuous readings, so that the system stability is insufficient under the complex working condition of the scene, and misreading frequently occurs. Disclosure of Invention The present invention has been made in view of the above-described problems occurring in the prior art. Therefore, the invention provides a lightning arrester instrument data image identification method, a system, equipment and a medium, which solve the problems of insufficient identification stability and limited practicability in field application of the existing lightning arrester instrument monitoring technology. In order to solve the technical problems, the invention provides the following technical scheme: In a first aspect, the present invention provides a lightning arrester instrument data image recognition method, including: Acquiring an original image of a lightning arrester instrument, adaptively adjusting image acquisition parameters according to ambient illumination conditions, performing geometric correction and quality evaluation on the original image, and performing dynamic contrast enhancement and multi-scale noise suppression based on an illumination decoupling color space model to generate a standardized image; Inputting the standardized image into a lightweight target detection model deployed at a mobile terminal, and positioning and cutting out a target sub-image containing an instrument display area; Performing image enhancement and edge feature extraction on the target sub-image, judging the instrument type based on edge distribution characteristics, and performing region segmentation and special identification network analysis on the display content of the digital instrument to obtain initial instrument reading; And carrying out validity verification on the initial instrument reading based on a preset physical quantity reasonable range, generating a final instrument reading, associating the final instrument reading with a current inspection task session, automatically attaching time information, a geographic position and an original image index, storing the time information, the geographic position and the original image index into a local structured database through transactional operation, and supporting to generate a structured report containing a recognition result and image evidence according to the inspection task. As a preferable scheme of the lightning arrester instrument data image recognition method, the color space model based on illumination decoupling carries out dynamic contrast enhancement and multi-scale noise suppression, and the method comprises the following steps: converting the geometrically corrected image to a decoupling color space comprising a structural luminance component and an illumination intensity component; Performing bi-directional gamma correction on the illumination intensity component; Carrying out mu