CN111695576A

CN111695576A - Electric energy meter appearance rapid identification method based on neural network

Info

Publication number: CN111695576A
Application number: CN202010517193.3A
Authority: CN
Inventors: 马红明; 史轮; 李倩; 陶鹏; 赵国鹏; 马笑天; 张知; 杨丽
Original assignee: State Grid Corp of China SGCC; Electric Power Research Institute of State Grid Hebei Electric Power Co Ltd; Marketing Service Center of State Grid Hebei Electric Power Co Ltd
Current assignee: State Grid Corp of China SGCC; Electric Power Research Institute of State Grid Hebei Electric Power Co Ltd; Marketing Service Center of State Grid Hebei Electric Power Co Ltd
Priority date: 2020-06-09
Filing date: 2020-06-09
Publication date: 2020-09-22

Abstract

The invention relates to a method for quickly identifying the appearance of an electric energy meter based on a neural network, which cuts an original picture into a plurality of subdivided verification classification problems after graying, binary edge detection and correction, combines the cut image, and identifies and collects the appearance information of the electric energy meter by using means. The invention reduces a large amount of sorting work such as manpower and material resources, repeatability, mechanicalness and the like; the working accuracy and the working efficiency are improved; corresponding verification units and verification schemes are changed and added on the original line body, an automatic appearance verification system of new and old electric energy meters is compatible, and the new and old electric energy meters can be subjected to mixed online verification at the same time.

Description

Electric energy meter appearance rapid identification method based on neural network

Technical Field

The invention relates to the technical field of automatic verification systems of electric energy meters, in particular to a method for quickly identifying the appearance of an electric energy meter based on a neural network.

Background

With the rapid popularization of big data in multiple fields and the comprehensive construction of smart power grids, new technologies such as 'big cloud migration' are continuously and deeply applied to the field of power industry, and various marketing data, power consumption data, power distribution data and the like also become massive resources which can be utilized. "according to the construction requirement of the large marketing system of the national grid company, 27 provincial companies of the grid province have completed the provincial concentration of the metering business, and the centralized verification and distribution quality control of the metering assets becomes one of the core businesses of the metering center. Lean management can realize online monitoring, closed-loop management and quality control of the whole business process of asset measurement only by depending on full-life cycle management, so that the specialized management requirement is met. The main informatization systems in the marketing electricity utilization link are an electricity utilization information acquisition system and a marketing business application system, and the data characteristics of the system are that the acquisition points are multiple, the coverage range is wide, the data volume is large, various business data have large scale, and the mass data provide possibility for the management of the whole life cycle of the metering assets.

Since the electric energy meter is popularized and applied in a large scale, research and application of electric power big data are developed on the basis of the electric energy meter, large research institutions in China, China electric academy of sciences, domestic and overseas universities and national grid companies. In foreign countries, big data analysis systems are developed by IBM and C3-Energy and are aimed at smart grid usage, and power companies in the united states, canada, france, spain, etc. develop research and application of user power data, and research focuses on power supply quality service, power demand prediction, and other aspects. In China, a plurality of intelligent power grid big data research projects are started by national power grid companies, power companies such as Jiangsu, Zhejiang, Tianjin, Beijing, Shanghai and the like begin to build marketing big data intelligent analysis systems and carry out big data application research of measurement asset management prediction, at present, each company obtains certain achievements in the aspect of cleaning and integration mechanisms of data models, is still in an exploration research stage for the actual application of data, and lacks deep research achievements of systems and clear theoretical frameworks and actual application guidance.

In order to realize the vision of building the top-class power grid of the world, national grid companies have been dedicated to smart grid construction in recent years. As a key device for interaction between the tail end of the smart grid and a user, the electric energy meter plays an increasingly remarkable role in non-metering function besides the role of traditional legal metering, and supports the development of company marketing major and other major business applications.

At present, 2200 thousands of electric energy meters are installed in the Hebei south network in an accumulated way, and the full coverage of the electric energy meters is realized. The national network company highly attaches importance to the quality control of the electric energy meter, and forms a full life cycle management system with the electric energy meter as the core for bidding purchase, product detection, online monitoring and operation maintenance on the basis of documents such as an electric energy meter quality supervision and management method, a meter disassembly and sorting management method and the like. Since the electric energy meter is subjected to centralized bidding purchase in 2010, the number of faults is gradually increased along with the increase of the operation time, and a new challenge is brought to high-quality service. Fault handling and trend analysis of electric energy meters have become an important issue to be solved urgently.

Disclosure of the invention

The invention aims to provide a method for quickly identifying the appearance of an electric energy meter based on a neural network, which cuts an original picture into a plurality of subdivided verification classification problems after graying, binary edge detection and correction, combines the cut image, and identifies and collects the appearance information of the electric energy meter by using means.

Furthermore, the electric energy meter appearance imaging is preprocessed in a gray level conversion mode, so that the noise influence of the complex environment of the image is reduced, and meanwhile, the image processing efficiency is improved.

Further, the cut pictures are classified and identified according to requirements, so that the electric energy meter is subdivided, and the subdivided state conditions are summarized into an overall appearance judgment conclusion.

Further, an edge detection method based on different operators and binaryzation is adopted to clean the noise of the image of the gray-scale electric energy meter, the noise which does not belong to the meter body in the imaging is eliminated, and the imaging is primarily cut into the image only containing the electric energy meter body.

Further, irregular electric energy meter imaging data after preliminary cutting are unified by using a method of distortion correction and interpolation smoothing of offset angle mapping, and the size of the data scale is unified, so that the electric energy meters imaged at different angles are subjected to pixel-level comparison.

Further, the imaging data with the uniform size after the distortion correction is cut according to 21 appearance verification items, verification conclusions are respectively made on the 21 cut images, and all the subentry conclusions are gathered.

Further, a Multi-digital number classification method is adopted for identifying the number collection of the cut nameplate image aiming at the fixed-length digital sequence, and remark information which is taken as a verification result is output.

Further, the model information of the cut nameplate image is matched by similarity calculation of a fixed template, meter models are distinguished, and remark information of a verification result is output.

Furthermore, OCR recognition of a single line is carried out on manufacturer information of the cut nameplate image, data collection is perfected by automatically recognizing the nameplate information, and finally the serial number, the model number and the equipment information are integrated into additional information of a verification result.

Furthermore, the 21 appearance verification subentries and the additional information are integrated to generate a complete one-time appearance verification result, verification sample data is analyzed in batches, and high-quality data are provided for data mining and analyzing tasks.

A neural network-based electric energy meter appearance identification method for rapid verification comprises edge detection, distortion correction, picture cutting, character identification and picture classification.

Edge detection: the purpose is to identify points in the digital image where the brightness changes are significant. Significant changes in image attributes typically reflect significant events and changes in the attributes. These include: discontinuities in depth, surface orientation discontinuities, material property variations, and scene lighting variations. The image edge detection greatly reduces the data volume, eliminates information which can be considered irrelevant, and retains important structural attributes of the image.

In the electric energy meter snapshot, a large amount of information other than the electric energy meter is included, and this is considered as noise information.

Firstly, the image edge is separated from the background, and then the image detail can be perceived to recognize the image outline. Computer vision is just this process of mimicking human vision. Therefore, when the edge of the object is detected, the contour points are roughly detected, the originally detected contour points are connected through a link rule, and meanwhile, the missing boundary points are detected and connected and the false boundary points are removed. The edge of an image is an important feature of the image and is the basis of computer vision, pattern recognition and the like, so edge detection is an important link in image processing. However, edge detection is a difficult problem in image processing, since the edges of the actual scene image are often a combination of various types of edges and their blurred results, and the actual image signal is noisy.

Therefore, the actual imaging effect of the appearance detection based on the existing verification assembly line is divided into five parts:

graying of an image: graying a color image, namely carrying out weighted average according to sampling values of all channels of the image.

Gaussian filtering: the image Gaussian filtering can be realized by two one-dimensional Gaussian kernels which are weighted twice respectively, and can also be realized by one convolution of one two-dimensional Gaussian kernel.

Calculating gradient and amplitude: the gradient of the image gray values can be approximated using a first order finite difference, which results in two matrices of partial derivatives of the image in the x and y directions.

Gradient amplitude non-maxima suppression: and searching the local maximum value of the pixel point, and setting the gray value corresponding to the non-maximum value point as 0, so that most non-edge points can be eliminated. After the suppression of the non-maximum value is completed, a binary image is obtained, and the gray values of the non-edge points are all 0.

Thresholding: two thresholds are chosen (the selection method of the thresholds is discussed in the extension), and an edge image is obtained according to the high threshold, so that the image has few false edges, but because the threshold is high, the generated image edge may not be closed, and another low threshold is adopted in the unsolved problem. The edges are linked into the contour in the high-threshold image, when the end point of the contour is reached, the algorithm searches for a point meeting the low threshold value in 8 adjacent points of the breakpoint, and then collects new edges according to the point until the edge of the whole image is closed.

Distortion correction: the key point of the distortion correction is to find the corresponding relation of the point positions before and after the distortion.

Cutting pictures: for the picture after distortion correction, firstly, interpolation smoothing is carried out on the corrected picture, then, according to the difference of the appearance verification items, the area cutting is carried out on the appearance verification snapshot according to the fixed position of the pixel, and image data of specific detection items, such as lead sealing, indicator lights, readings and the like, are generated. Thereby subdividing the appearance verification into classification questions for each verification item.

Character recognition: the nameplate of the electric energy meter mainly comprises serial numbers, models and manufacturer information, and aims to realize appearance detection of compatible new and old meters and realize efficient data collection. And selecting an incompatible mode for identification according to the information type. Identifying the fixed-length digital sequence by using a multi-bit identification method aiming at the numbered fixed-length information; matching by using similarity calculation of a fixed template according to model information, and distinguishing meter models; a single line of OCR recognition is performed against the vendor information. And the data collection is completed by automatically identifying the nameplate information.

Classifying pictures: since the image has passed through the image data for each appearance verification item for which the cutting has been completed, similarity matching is performed. And evaluating the detection items through the similarity between the verification item picture and the normal picture.

Template matching/histogram: and respectively calculating histograms of the two images, normalizing the histograms, and measuring the similarity according to the distance measurement standard.

Peak signal-to-noise ratio: based on the error between the corresponding pixel points, i.e. based on the error-sensitive image quality evaluation.

Structural similarity: the image is blocked by utilizing a sliding window, the total number of blocks is N, the mean value, the variance and the covariance of each window are calculated by adopting Gaussian weighting in consideration of the influence of the window shape on the blocks, then the structural similarity SSIM of the corresponding block is calculated, and finally the mean value is used as the structural similarity measurement of the two images.

Perceptual hashing: a "fingerprint" string is generated for each image, and the fingerprints of the different images are then compared. And judging the image according to the fingerprint similarity.

The invention has the advantages that:

the invention reduces a large amount of sorting work such as manpower and material resources, repeatability, mechanicalness and the like; the working accuracy and the working efficiency are improved; corresponding verification units and verification schemes are changed and added on the original line body, an automatic appearance verification system of new and old electric energy meters is compatible, and the new and old electric energy meters can be subjected to mixed online verification at the same time.

Drawings

FIG. 1 is a flow chart of the present invention.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the application, its application, or uses. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.

The relative arrangement of the components and steps, the numerical expressions, and numerical values set forth in these embodiments do not limit the scope of the present application unless specifically stated otherwise. Meanwhile, it should be understood that the sizes of the respective portions shown in the drawings are not drawn in an actual proportional relationship for the convenience of description. Techniques, methods, and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail but are intended to be part of the specification where appropriate. In all examples shown and discussed herein, any particular value should be construed as merely illustrative, and not limiting. Thus, other examples of the exemplary embodiments may have different values. It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, further discussion thereof is not required in subsequent figures.

In the description of the present application, it is to be understood that the orientation or positional relationship indicated by the directional terms such as "front, rear, upper, lower, left, right", "lateral, vertical, horizontal" and "top, bottom", etc., are generally based on the orientation or positional relationship shown in the drawings, and are used for convenience of description and simplicity of description only, and in the case of not making a reverse description, these directional terms do not indicate and imply that the device or element being referred to must have a particular orientation or be constructed and operated in a particular orientation, and therefore, should not be considered as limiting the scope of the present application; the terms "inner and outer" refer to the inner and outer relative to the profile of the respective component itself.

Spatially relative terms, such as "above … …," "above … …," "above … …," "above," and the like, may be used herein for ease of description to describe one device or feature's spatial relationship to another device or feature as illustrated in the figures. It will be understood that the spatially relative terms are intended to encompass different orientations of the device in use or operation in addition to the orientation depicted in the figures. For example, if a device in the figures is turned over, devices described as "above" or "on" other devices or configurations would then be oriented "below" or "under" the other devices or configurations. Thus, the exemplary term "above … …" can include both an orientation of "above … …" and "below … …". The device may be otherwise variously oriented (rotated 90 degrees or at other orientations) and the spatially relative descriptors used herein interpreted accordingly.

It should be noted that the terms "first", "second", and the like are used to define the components, and are only used for convenience of distinguishing the corresponding components, and the terms have no special meanings unless otherwise stated, and therefore, the scope of protection of the present application is not to be construed as being limited.

The technical solution and structure of the present invention will be described in further detail with reference to the accompanying drawings.

The invention provides a method for quickly identifying the appearance of an electric energy meter based on a neural network, which cuts an original picture into a plurality of subdivided verification classification problems after graying, binary edge detection and correction, combines the cut image, and identifies and collects the appearance information of the electric energy meter by using means.

graying of an image: graying a color image, namely carrying out weighted average according to sampling values of all channels of the image. Taking color chart in RGB format as an example, the method adopted by the common graying mainly comprises

The method comprises the following steps: gray = (R + G + B)/3;

the method 2 comprises the following steps: gray =0.299R +0.587G + 0.114B;

The one-dimensional kernel vector can be obtained by determining the parameters

The convolution operator employed in the Canny algorithm is expressed as follows:

the mathematical expressions of the first-order partial derivative matrix, the gradient amplitude and the gradient direction of the x direction and the y direction are as follows:

Distortion correction: the key point of the distortion correction is to find the corresponding relation of the point positions before and after the distortion. Assuming that before distortion, the pixel coordinates of each point in the image can be obtained by the formula:

if no distortion is present, then ideally the coordinate transformation for the camera image can be calculated as above. In the case of distortion, the distorted coordinates are:

at the same time, the new coordinates of each pixel of the distorted image can be obtained as follows:

thus, an image is obtained from a camera coordinate system, then the image is distorted, and finally the mapping relation of each point in the whole coordinate transformation process of the distorted image is obtained.

As shown in fig. 1, in the appearance verification part, the snapshot of the appearance of the electric energy meter is processed. Due to the position of the camera, the snapshot angles are inconsistent. And firstly, converting a three-color channel picture into a single-channel image by utilizing a gray level conversion mode for snapshots with inconsistent angles, and realizing filtering of the image by one-time convolution of a two-dimensional Gaussian kernel. The gradient of the image gray value can be approximated by using a first-order finite difference, then the binarization processing of the picture is carried out, finally the edge identification content of the electric energy meter main body is obtained, and the picture of the electric energy meter main body is taken out from the snapshot. And then distortion correction is carried out on the taken out electric energy meter image main body. And mapping each pixel point based on distortion correction of coordinate conversion, and obtaining an imaging result with consistent snapshot angles of each electric energy meter after smooth interpolation. And respectively cutting out image materials required by corresponding verification according to the pixel positions of the 21 appearance verification items, and sequentially carrying out classification judgment on all verification item images.

The electric energy meter nameplate information is divided into a plurality of parts, and fixed length numbers, indefinite length text information and fixed templates are respectively identified and matched. And finally, updating the result of the appearance verification of each electric energy meter to a database in a form of the result containing the manufacturer, number, model and appearance items. The method can obtain all appearance verification results, and meanwhile, unknown abnormity can be avoided by comparing verified nameplate information with purchasing or asset information.

Claims

1. A method for quickly identifying the appearance of an electric energy meter based on a neural network is characterized by comprising the steps of edge detection, distortion correction, picture cutting, character identification and picture classification, wherein an original picture is subjected to image cutting after graying, binarization edge detection and correction, the cut image is combined to become a plurality of subdivided verification classification problem treatments, and meanwhile, the appearance information of the electric energy meter is identified and collected by using means.

2. The electric energy meter appearance rapid identification method based on the neural network as claimed in claim 1, characterized in that, the electric energy meter appearance imaging is preprocessed in a gray scale conversion mode, so that the noise influence of the complex environment of the image is reduced, and the image processing efficiency is improved.

3. The method for rapidly identifying the appearance of the electric energy meter based on the neural network as claimed in claim 1, wherein the cut pictures are classified and identified according to requirements, so that the electric energy meter is subdivided, and the subdivided states are summarized into an overall appearance judgment conclusion.

4. The method for rapidly identifying the appearance of the electric energy meter based on the neural network as claimed in claim 3, wherein an edge detection method based on different operators and binarization is adopted to clean noise of an image of the gray electric energy meter, noise which does not belong to a meter body in imaging is eliminated, and the imaging is primarily cut into the image only containing the meter body.

5. The method for rapidly identifying the appearance of the electric energy meter based on the neural network as claimed in claim 1 or 3, wherein the irregular electric energy meter imaging data after the preliminary cutting is unified by using a distortion correction and interpolation smoothing method of offset angle mapping, and the data scale size is unified, so that the electric energy meters imaged at different angles are subjected to pixel-level comparison.

6. The method for rapidly identifying the appearance of the electric energy meter based on the neural network as claimed in claim 5, wherein the unified size imaging data after distortion correction is cut for 21 appearance verification items, verification conclusions are made on 21 cut images respectively, and all the subentry conclusions are gathered.

7. The method for rapidly identifying the appearance of the electric energy meter based on the neural network as claimed in claim 5, wherein a Multi-digital number classification method is adopted for identifying a fixed-length number sequence for the collection of the serial numbers of the cut nameplate images, and remark information serving as a verification result is output.

8. The method for rapidly identifying the appearance of the electric energy meter based on the neural network as claimed in claim 3, wherein the model information of the cut nameplate image is matched by similarity calculation of a fixed template, the meter models are distinguished, and remark information which is a verification result is output.

9. The method as claimed in claim 3, wherein OCR recognition is performed on a single line for factory information of cutting a nameplate image, data collection is perfected by automatically recognizing nameplate information, and finally number, model and equipment information are integrated into additional information of verification results.

10. The method for rapidly identifying the appearance of the electric energy meter based on the neural network as claimed in claim 6, wherein 21 appearance verification subentries are integrated with additional information to generate a complete one-time appearance verification result, verification sample data is analyzed in batches, and high-quality data is provided for data mining and analysis tasks.